In 2018, the market was tumultuous from China trade wars, and the concept of an “AI data center” was not uttered by any stock analyst that I can recall. The AI blogs and Substacks you see today did not exist and the social media influencers that elaborate on the topic daily in 2026 were entirely focused on consumer tech and cloud (or had not yet begun investing in stocks). 

This history is important as it substantiates who is early to a trend, yet it gets buried very quickly as most of what we read today is designed to be ephemeral. Being early to a trend is not simply a bragging right; it’s how the majority of the money is made in the stock market. 

Jumping on a bandwagon yields lower returns as it means investing in what is already consensus. We avoid bandwagons, and instead, are often jumping off just as a trade gets too crowded. Below, I lay the groundwork for one of our boldest moves yet, which is to move to the sidelines with our Nvidia position. This move is driven by analysis that consistently questions: what is best for my Research Members right now? It is a question only afforded to the most independent research sites, of which there are very few.  

While technologists and AI developers can devour information without penalty; investors cannot. Where investors are quite different is that every new data point and every hot take can lead to costly redirection. To contrast, our site is built not only to identify major trends and lesser-known tech stocks early, but to help Members move forward with calm confidence. 

And in case it's gone unnoticed, the QQQs are flat this year. Many influencer-led tech ETFs are also flat to down – GRNY, IVES, and ARKK are all barely keeping pace with the broader tech index. In sharp contrast, we are up about 30% YTD, a meaningful outperformance during a bout of weakness in the markets and confusion around AI spending. Contributors include Bloom Energy, which we brought to our Research Members long before the stock became widely discussed — our initial entry near the 2025 lows is up over 1000% today. We also highlighted AAOI ahead of its 2026 surge, with the stock up nearly 300% YTD and over 650% since our lowest entry. We then doubled down on Lumentum in January with a 10% allocation, and the stock is up over 130% YTD. 

Similar to previous reports, the report below is our team stepping back up to the plate and pointing in the direction we think the ball will land. It is over 70 pages long and took three weeks to write, combining deep thematic work, fundamental analysis, and portfolio-level judgment.  

Although quite lengthy, this is about executing in the last inning when the game must be won. Whether you joined our site years ago or only since January, our official batting average is improving. Let's see if we can deliver for our Members again this quarter. 

AI Accelerators: Shifting from Raw Compute to Unit Economics 

AI accelerators are shifting their primary focus from raw compute to unit economics. Two years ago, Gartner had predicted 40% of existing AI data centers will be operationally constrained by power availability by 2027. That date is fast approaching, which means to continue selling GPUs or XPUs, leading AI semiconductor companies must actively work to solve for this constraint.  

More recently, Morgan Stanley stated data centers are facing a “power shortfall totaling as much as 20%” for data centers through 2028. It’s expected there will be 13 GWs of shortage through 2028, when factoring in behind-the-meter solutions (more on that under the Energy section below). 

While supplying power is outside of their domain, unit economics is something companies like Nvidia, AMD and Broadcom can help to improve. What this suggests is that the semiconductor supply chain will do everything within reach to lower the power requirements of AI systems from a design perspective.  

To illustrate this approach, recently, the I/O Fund team covered how Arm is tackling lower power requirements in our write-up Arm Stock Could Win as Agentic AI Shifts the Bottleneck to CPUs  Arm Stock Could Win as Agentic AI Shifts the Bottleneck to CPUs  stating: “Arm is taking this a step further with a fully-liquid cooled, 200kW open-standard rack in partnership with Super Micro, packing 168 blades, or 336 CPUs, delivering a total of up to 45,696 cores. Arm EVP of Cloud AI Mohamed Awad stated that while it is a ‘200-kilowatt rack. We actually will consume about half that much power. We ran out of space. That’s why we couldn’t put more cores in there.’ 

This is one of the key advantages – it is not just about offering 2X the performance of x86 chips, but providing that performance boost while freeing up power for more compute or for more networking […]  Arm says the new chip’s performance advantage over x86 could enable “up to $10B in capex savings per GW of AI data center capacity,” making it a compelling option for current and future agentic AI-optimized deployments to save money, save power and avoid Nvidia-lock in from its accelerator-agnostic nature.” 

To connect the dots here, these stats should not be glossed over. As opposed to the compute-driven era we are firmly exiting, the way forward will be architectural designs that can lower power requirements. Instead of asking “how much compute can we build?” Big Tech is waiting in interconnection queues and waiting for the completion of nuclear power plants, asking: “How much compute can we actually power?” 

Nvidia’s Systems are Becoming More Efficient:  

There are two primary ways that Nvidia plans to assist in the push for better unit economics, such as cost per token and performance per watt. The first is to make each system they sell more efficient, and the second is to increase GPU density within the same power envelope.  

In the March press releases from GTC, there is a subtle hint that Nvidia’s core KPI is not FLOPs anymore but rather tokens per watt.   

Nvidia’s GB300 NVL72s offer 50X better performance per watt and 35X lower cost per token compared to the H200s. When comparing the Vera Rubin NVL72 to Blackwell, the new system delivers 4X better training performance and up to 10X better inference performance per watt.  

In other words, if a data center has 100MW of power, then Vera Rubin allows 10X more inference in the same power envelope as Blackwell, which is critical for hyperscalers that are constrained by facility power. 

More GPUs in the same Power Envelope:  

As announced at GTC this year, Nvidia’s MGX racks will include rack-level energy storage capacitors to avoid the large load swings created by AI training and inference workloads. According to Nvidia, these spikes create stress on the grid and data center power infrastructure with these improvements resulting in lower peak current by up to 25% while Intelligent Power Smoothing will also help to reclaim stranded capacity:  

“NVIDIA Vera Rubin NVL72 now introduces Intelligent Power Smoothing. It features 6x more rack-level energy storage (400 J per GPU) versus prior generations, and introduces a new closed-loop system that enables the GPUs to continuously monitor the state of charge of the capacitors to more efficiently flatten power profiles. This achieves much smaller AC power variation per minute, reduces peak current demands by up to 25%, and eliminates the need for massive battery packs to protect against large-scale power transients. At the facility level, provisioning racks at static Max-P strands power capacity that could otherwise be used to generate tokens. It assumes homogeneous workloads that always require peak power, when in reality AI factories run a mix of workloads with varying power needs.” 

What Nvidia is describing here is that by flattening power spikes and by lowering peak current demand, Nvidia can put up to 30% more GPUs in the same facility power envelope. Max-P refers to static maximum power whereas Max-Q refers to allocating power more dynamically to accomplish better economics. In practice, the DSX Max-Q API is a software tool that Nvidia offers to achieve a token-per-watt goal rather than just raw performance.  

Following GTC in March, Nvidia is effectively agreeing that power is the defining constraint of the AI buildout. 

Join the Discovery tier for early access to stock ideas and to stay ahead of where the market is heading next. To subscribe to Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40 Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40 

The Importance of Cooling Technologies: 

We’ve covered direct liquid cooling for a few years, beginning with Supermicro in 2023 and later with a cohort of cooling stocks in 2024. The simplest thesis as to why my Q2 report is emphasizing cooling technologies at this critical juncture for lowering power for data centers is the following: 

“Cooling data center servers is responsible for 40% of the data center energy consumption. According to Dell, enclosed DLC solutions can save up to 23% of energy compared to traditional air-cooled racks. McKinsey places this number at 27% savings when there is 75% liquid cooled and 25% air cooled servers.:” – Liquid Cooling Leaders, June 2024 Liquid Cooling Leaders, June 2024  

There are a few reasons why we dropped direct liquid cooling from our coverage over the past year or so but are picking it back up again as an important AI thematic trend. Although Blackwell offered both liquid cooled and air-cooled options in the B200s and the MGX NVL36s, many deployments remained with air-cooled because the current data facilities are built for air-cooled. This includes neoclouds, which also have a strong preference for the B200s and NVL36 with industry analysts stating the 40kW rack requirements were an easier upgrade from Hopper’s 20kW rack requirements, achieved by skipping a row: “Since it is only 40kW per rack, the MGX NVL36 can be air cooled […] This makes the MGX NVL36 very easy for existing datacenter operators to deploy without reworking their infrastructure.” 

Rubin changes this as air-cooled is not an option as this generation reaches 180kW to 230kW per GPU. Nvidia is redesigning its Rubin racks to be “liquid cooled with high warm-water inlet temperatures,” which will help to lower facility power costs while freeing up more power for compute.  

For Max-Q to be achievable (mentioned above), systems must be cooled to 45 degrees Celsius or 113 degrees Fahrenheit. According to Nvidia, this approach leads to “significant data center power savings,” resulting in up to 10% more GPUs being deployed. Here is what was stated: 

“Operating at 45°C enables data centers in many climates to use ambient air and closed loop dry coolers for cooling, reducing the need for compressors, driving down PUE, and unlocking larger energy budgets for compute. Lower inlet temperatures of 35°C require data centers to divert massive amounts of facility power or water for cooling, while higher inlet temperatures maximize the amount of grid power converted directly into tokens. This yields significant data center power savings—enough to allocate up to 10% additional Vera Rubin NVL72 racks for more token generation in the same power budget.” 

Rather than “cool” data centers, Nvidia is proposing to use warm water, which will require less power for chilling servers and allow more power to be used toward more compute. To achieve this, Nvidia uses facility water loops and coolant distribution units (CDUs) rather than direct-to-chip cooling to recycle the warm water across the facility. 

All of the above marks an important change in tone for the Nvidia management team as it’s more about performance-per-watt and cost-per-token rather than raw performance or FLOPs. The first approach Nvidia is taking to decouple bigger racks from needing proportionate power is Intelligence Power Smoothing and warm-water cooling. 

As investors, we should take our cue from the leading AI management team that the constraint in AI has officially shifted to Energy. 

Nvidia’s Dominance Faces Its Biggest Test Yet 

Recently, I’ve reiterated my $20 trillion market cap thesis, which translates to about 400% returns over the next four years, yet to assume Nvidia achieves this through hardware would be incorrect, in my opinion. The thesis hinges on software advancements and the recurring revenue that will inevitably come from Nvidia’s lead in robotics and simulation. Notably, I’ve held this opinion on the importance of Nvidia’s software business relative to hardware since 2023. 

However, on the flip side, by saying software is central to the $20T thesis, I'm implying that Nvidia’s hardware moat becomes breached. Over 7 years ago, my original thesis on why Nvidia can become the world’s most valuable company when it was at a $100 billion market cap was centered on the moat the CUDA platform provides when I stated: “Developers will self-regulate the number of competitors for processing units due to a need for a universal platform that supports all frameworks.” 

However, programming GPUs with the CUDA platform is primarily a training exercise as this is the phase where engineers are experimenting and need the developer ecosystem, including extensive tools like cuDNN, NCCL, debugging, custom kernel support, and CUDA’s massive libraries. The ecosystem has been built for over 20 years, has over 4 million developers contributing and every ML framework is first optimized for CUDA. The switching costs are extraordinarily high for engineers. 

To contrast, inference is repetitive to where once a model is trained, the model is running millions of times per day. Serving platforms and inference frameworks like vLLM and TensorRT-LLM reduce dependency to develop on a specific software platform, like Nvidia’s CUDA. There is also more of a push toward open standards for the inference phase to reduce dependency on hardware specific code for serving paths, as tools like ONNX runtime, vLLM and the compiler Triton help to export models (or compile them) to be run agnostically on any AI accelerator. 

In response to CUDA's moat weakening in the inference phase, Nvidia has pushed for their inference stack to remain proprietary by offering inference optimization software called TensorRT-LLM. TensorRT-LLM analyzes and optimizes LLMs to improve performance by running multiple operations on a single GPU kernel, selecting the optimal precision and optimizing memory usage for the key-value cache. Overall, Nvidia states this leads to 2-5X faster model performance for inference.  

However, consider that Nvidia is needing this new attempt at vendor lock-in as the CUDA dynasty will not hold in the inference market. The open-source market is growing to become a serious contender to proprietary optimization software like TensorRT-LLM, as alternatives that are more community driven are available and accomplish something similar, such as vLLM and SGLang. Furthermore, large inference players like Cloudflare can build their own custom engines.  

Expectations for the Erosion of Nvidia’s Market Share: 

Below, I present what a few industry analysts are predicting. Although I believe these are aggressive, they help to illustrate what is in front of Nvidia as the hardware moat becomes breached.  

Counterpoint Research believes that by 2028, custom silicon will cross the 15-million mark to surpass GPU shipments as the top 10 hyperscalers will have deployed 40 million AI server compute ASIC chips cumulatively during 2024-2028, stating: “What is also supporting this unprecedented demand is AI hyperscalers building significant rack-scale AI infrastructure based on their in-house stacks, such as Google TPU Pods and AWS Trainium UltraClusters, enabling them to operate as one supercomputer.” 

TrendForce is the most aggressive forecast, stating GPU-based AI servers will account for 69.7% of shipments in 2026 with ASIC-based servers rising to 27.8%. This doesn’t account for GPU market share from AMD, which if you put that at 10%, would result in Nvidia’s market share being 59.7%.  

With the information that I have today, these forecasts are too aggressive. 

According to Broadcom, they’ll see $100B in AI chip revenue in 2027 and we’ve modeled another $50B in networking. If we allocate $60B to AMD and go with what we know of Nvidia’s stated trajectory to $1 trillion in revenue, then the split looks something more like this for 2027: 

• NVDA $500B 
• AVGO $150B to $200B (assuming mgmt team was being conservative we will use the $200B number) 
• AMD $60B 
• Total among top 3 silicon providers: $760B with NVDA at 66% market share 

However, one data point that complicates things is MediaTek could see 150,000 CoWoS wafers in capacity in 2027, compared to 20,000 in 2026. Thus, the landscape is evolving in terms of the number of competitors.  

Reference more CoWoS allocation notes under the AMD section. 

The Linchpin: Rubin Delay Related to HBM4 

The reason for closing our Nvidia is two-fold. As outlined above, custom silicon is expected to gain meaningful share in the coming years. Nvidia’s GPUs come at a significant premium, and Big Tech seeks to lower total cost of ownership. Additionally, inference prioritizes repetition and efficiency over general-purpose flexibility, where the CUDA moat matters less than it did during the training market.  

At the same time, custom silicon is designed for specific workloads, allowing for lower power requirements, which is a critical advantage during a window when power will be greatly constrained.  

To add to this, we are getting additional confirmation of an incoming Rubin delay. To be blunt, this is terrible timing for Nvidia as Big Tech was already diversifying with custom silicon. This makes a stronger case for having back-up orders with Broadcom, MediaTek and/or AMD.  

HBM4 validation times have been cited as one key factor behind the delays for Nvidia’s upcoming Vera Rubin generation – we have seen in the past that these qualification tests can extend as long as 18 months, such as in Samsung’s case with HBM3e. Currently, reports suggest this HBM4-related delay could persist for one quarter. 

Reports suggest this delay stems from Nvidia pushing suppliers to “request speeds of over 11 Gb/s per pin,” well above the JEDEC standard of 8Gb/s. More evidence for a delay is surfacing, with DigiTimes reporting on April 15 that SK Hynix is “considering reducing its planned 2026 shipments of high-bandwidth memory (HBM4) to Nvidia by about 20-30%.”  

We also have another report stating SK Hynix is delaying its HBM4 production ramp until Q3, instead of its original Q2 target, with the delay said to better align with Nvidia’s schedule. Any potential delays or shipment cuts at SK Hynix also could be a key factor in a Rubin delay, as SK Hynix reportedly secured more than 70% of HBM orders for the upcoming chip; on the other hand, Micron and Samsung both have announced that HBM4 is in mass production for Vera Rubin, easing some of the supply constraints.  

Memory: Pricing Power Takes on Market Doubts 

The market is presenting two very different extremes with surging DRAM and NAND pricing causing memory stocks to skyrocket, until recently, when this subsector came to a screeching halt following Google’s TurboQuant announcement. The announcement got a lot of attention as Google stated they can deliver up to a 6x compression on LLM memory and vector search.  

The Google TurboQuant announcement specifically addresses the KV cache, which essentially serves as a model’s long-term memory that is reused and extended throughout many steps or requests. KV cache capacity is a known pain point when working to balance long-context reasoning and memory capacity in inference workloads, as it can consume ~30% of GPU memory during deployment.  

TurboQuant directly addresses this pain point by compressing the vectors (queries from users and keys in the KV cache) via real-time quantization, reducing the amount of bits needed to store ‘high-dimensional’ or complex data such as image features. This is increasingly important with AI coding, natural language processing, and multi-agent workflows, as the more a model is used, the more memory the KV cache takes up as it stores all of the prior responses.  

Alphabet says TurboQuant can drive a 6X reduction in KV cache memory size, all while preserving model accuracy and accelerating speed up to 8X. However, there is some debate here as industry analysts have noted that “actual memory savings are around 2.7x, with speed improvements of about 4x.” 

The market first interpreted the TurboQuant release as a memory ‘demand killer’, though it’s possible that the true reality may be that this is another ‘DeepSeek moment’ (as stated by Cloudflare’s CEO), where these new optimizations simply drive more AI infrastructure and thus more memory demand. It may all boil down to what Alphabet itself acknowledged – that TurboQuant “lowers memory costs.”  

This could be another Jevon’s Paradox in the making, where the new efficiencies created on the KV cache side simply drive memory and AI infrastructure demand higher (not lower) by opening up new use cases and reducing memory costs.   

Looking at this from the lens of AI inference, or inference at the edge or on local devices, the ability to reduce KV cache footprint, boost speed, preserve accuracy, lengthen context windows and allow for more concurrent requests may drive faster adoption of applications such as multi-agent systems, coding and natural language processing.  

TurboQuant does not reduce the need for HBM in AI accelerators, as HBM will still be critical for parameter storage for training; it simply minimizes KV cache usage in inference. It also does not replace the NAND flash and SSD storage layer for training data, inference data retrieval and caching. Instead, what it likely will do is allow larger models to be run on the same accelerator footprint, and enable and broaden access to previously infeasible memory-intensive workloads (like multi-agent systems) to a wider range of users, from lowering memory usage and decreasing memory costs. 

While the market grew concerned over TurboQuant, the other side of the picture – prices – may have been lost in the noise as the release was cited as a key contributor to a slight pullback in DRAM prices over the last few weeks. The 10,000-foot view instead shows DRAM prices had risen >20X over the last year, with robust price momentum for both DRAM and NAND throughout Q4 and Q1 that is now extending into Q2.  

Key factors behind the rapid ascent in prices included major manufacturers shifting supply to prioritize AI-related HBM and LPDDR (server DRAM) demand, and new chips such as Nvidia’s Blackwell Ultra incorporating 50% more HBM capacity per chip versus Blackwell. This strong demand, increasing content of HBM attached to accelerators and DRAM content growth in AI servers is also why we saw Micron shutter its consumer DRAM unit to focus solely on AI. The NAND and enterprise SSD side faces extremely tight supply coupled with strong AI storage demand — Kioxia sold out of 2026 NAND capacity in January, and reports surfaced recently that controller supplier Phison’s CEO said that “every NAND manufacturer told us 2026 is sold out.” 

Closer Look at the Memory Pricing Surge 

It’s safe to say that looking back, the ascension and sheer pace of memory prices caught the entire industry off-guard as supply constraints worsened. 

The vertical ascent in memory prices first became visible late last year within DDR4/5 chips for PCs, with prices surging from roughly $6.84 in September to $27.20 by December as supply and inventories rapidly tightened. This surge quickly extended well beyond PC/consumer DRAM, as server DRAM, NAND flash and enterprise SSDs also witnessed prices rise sharply into year end and early 2026.  

For example, back in September, TrendForce had initially estimated conventional DRAM prices to rise 8-13% QoQ in Q4 driven by some supply constraints for DDR4/5 for PCs, and up 13-18% QoQ when including HBM. By November, reports were surfacing that quotes from Samsung were rising from $149 to $239 for 32GB DDR5, with other capacities all rising 30-50% since September, more than 3X the estimated quarterly increase.    

By the end of Q4, conventional DRAM prices were pegged at +45-50% QoQ, far above initial estimates, with HBM-blended prices up 50-55% QoQ. The price hikes were only projected to worsen moving through Q1, with TrendForce estimating conventional DRAM prices up 90-95% QoQ (on top of Q4’s increase), driven by PC DDR4/5 up 105-110% QoQ and LPDDR5 server DRAM up 88-93% QoQ. Samsung had reportedly doubled its DRAM contract prices in Q1 versus Q4, with Q2 expected to see another 30% increase on top of that.   

On the NAND side, prices followed a similar trajectory, with Q4 prices for enterprise SSDs estimated at 25-30% QoQ, with total NAND flash prices reported to be up 33-38% QoQ. For Q1, TrendForce had estimated at the start of February that the pace of NAND flash price hikes would quicken, rising 55-60% QoQ, with enterprise SSDs nearly matching that pace at 53-58% QoQ. By the end of March, barely eight weeks later, and NAND prices were estimated to be significantly higher, up 85-90% QoQ.  

Initial price estimates for Q2 signal robust momentum continues, with conventional DRAM forecast to increase 58-63% QoQ, and NAND flash rising 70-75% QoQ. However, DRAM prices have recently pulled back around ~20-30%, with reports pointing to two main factors for the decline – distributors beginning to sell off stockpiled inventory and TurboQuant. Even with this correction, it should be noted that DDR4/5 prices remain >20X of their early 2025 prices.  

For HDDs, pricing is more obscure. Major suppliers WDC and Seagate have already sold out of capacity for 2026 with price and volume conditions set in contracts, though 2027 pricing has not been set, and is the bigger catalyst on the horizon.  

WDC executives noted that last quarter, prices were up “2%, 3% on an ASP per terabyte basis,” with the pricing environment stable and expected to remain stable moving forward, implying similar steady growth through the rest of this year. Some reports suggested HDD contract prices rose ~4% QoQ, still a far cry from the rapid ascent seen in SSDs.  

Looking further ahead for HDDs, analysts from Morgan Stanley see 2027 pricing potentially being much stronger than currently expected. MS explained that they see “sustained hyperscaler demand strength, elongating customer visibility [and] firmer pricing into 2027,” with its initial channel checks suggesting hyperscalers are closer to paying $20 per terabyte for 2027 and 2028 capacity. This is significantly above current estimates for $13 to $15 per terabyte.  

Even with capacities selling out and prices rapidly advancing above expectations, the market is still finding a way to doubt the runway for memory stocks. This likely stems from memory’s cyclical nature, previous history of rather violent swings from peak to trough, and an expectation that current prices will soon peak and quickly reverse.  

However, the current environment has key ingredients for prices to remain strong. While it may be unlikely that we see prices rise >60% QoQ in each and every quarter this year, the persisting supply shortages coupled with the strength of AI-driven demand and timing of capacity expansion suggest that prices may not immediately reverse but rather remain elevated for longer – perhaps into 2027.  

Current expectations predict this supply shortage will persist through late 2027, though key industry executives are beginning to pencil it in lasting even longer. Intel CEO Lip-Bu Tan believes there will be “no relief” until 2028, while SK Group Chairman Chey Tae-won stated at Nvidia’s GTC that the shortage could persist another four to five years as wafer supply may lag demand by more than 20% at times.  

This is because capacity expansion efforts will take multiple quarters to materialize. For example, Micron detailed that initial output from its first Idaho fab is slated for mid-2027, while its Singapore fab and new Tongluo fab will add supply in late 2027 through 2028. Kioxia is planning to double its 2024 NAND capacity, but this will not be achieved until 2029, with equipment spending remaining below 2023 levels as NAND manufacturers remain cautious on spending to prevent oversupply.  

Samsung and SK Hynix are said to be prioritizing boosting 1c DRAM for HBM and DDR memory for AI applications, with NAND on the back burner for the two; SK Hynix is reportedly projecting a capacity boost in 1H 2027 to double 1c DRAM output, while  SK Hynix on the DRAM side, aiming to double its 1c DRAM output; Samsung is aiming to triple its 1c DRAM capacity by year-end 2026, supporting the ramp of HBM4. 

AI Networking:  

The stock market is like musical chairs, and you don't want to be the one left standing when the music stops. When it comes to networking companies gaining content on a platform or losing incremental share, nowhere does the supply chain shift faster than networking. This goes back to the differences between technologists and investors; an investor cannot afford to be the last one out whereas AI enthusiasts can devour information and debate without penalty. 

The reason networking sees immense volatility is straightforward: much of the market is tied to a single customer, Nvidia; and Nvidia is rolling out new architectural iterations at an unusually fast pace these days. 

On that note, we wanted to give our Members’ a heads-up last quarter that the copper-to-optics boundary was shifting, stating in the Q1 2026 AI Top 15 Report that: 

“While copper-based links remain essential for short-reach, low-latency connections—particularly within NVLink scale-up domains—the expansion of Ethernet fabrics, higher port counts, and the adoption of co-packaged optics are driving an inevitable shift toward optical content.   

Blackwell and Blackwell Ultra are fundamentally focused on solving scale-up problems, where the primary challenge is binding large numbers of GPUs into a single coherent node using ultra-dense, low-latency NVLink fabrics.   

Rubin, by contrast, is primarily focused on assisting higher bandwidth requirements, as the focus is now on sustaining inference and training workloads at scale without bottlenecks forming beyond NVLink. The limiting factor is how efficiently bandwidth can be delivered and distributed across racks and fabrics, resulting in higher port counts, faster link speeds (800G now and moving toward 1.6T). 

[…] The increasing amount of computing nodes (especially as Nvidia pushes towards the NVL576 with Rubin Ultra) along with increasing amount of interconnects means that bandwidth must also increase, from 400G to 800G and now to 1.6T, to ensure that low-latency, high-throughput communication remains across the entire platform.  

As a result, it’s expected that optics move closer to the switch, as copper and AEC content becomes constrained by reach and signal integrity. The result is a networking stack where silicon photonics capture incremental value, even though copper remains relevant and intact at the shortest distances. “ 

To continue on the theme for this report, networking is no longer optimized around compute. Instead, the road map for AI systems is forced to address fixed power envelopes while also preparing for larger clusters. Not surprisingly, the motivation to move to silicon photonics and co-packaged optics is also about reducing power consumption, stating CPO could “slash power consumption by 3.5X compared to traditional pluggable transceivers.” The press release goes onto say that by eliminating external DSPs and reducing the signal path from inches to millimeters, CPOs “dramatically boost power efficiency.” 

There are additional reasons, such as reducing component count and enhancing performance as adding DSPs can result in latency, as well. This becomes even more evident as data rates become faster with DSPs consuming half the power draw of a 1.6 Tbps transceiver.  

In a recent press release, Broadcom stated co-packaged optics can offer 65% power savings compared to re-timed pluggable optics. These may seem like big numbers but remember that compute is the bulk of the power draw, thus the overall impact is likely in the single digits.  

Although both of these companies have voracious appetites to control as much of the scale-up networking stack as possible, there are key areas where smaller vendors can participate, such as supplying ASIC switches, optical transceivers including lasers like EML, VSCEL, CW, Silicon Photonics chips and interconnects.  

To illustrate the opportunities for smaller vendors, Lumentum has been able to capitalize on tight EML supply and pricing power. In fact, EML shortages are so severe, that hyperscalers are accelerating the SiPho timeline by adopting alternatives like a combination of CW lasers and silicon photonics to forego waiting for more indium phosphide (InP) supply.  

The CW laser-SiPho combination combines a continuous-wave (CW) laser with a silicon photonics chip to handle modulation, which opens up the supplier base. Here is what a December TrendForce press release stated: 

“CW lasers offer a steady optical signal and are paired with silicon photonics chips produced at semiconductor foundries used as external modulators. Their simpler design stems from the absence of integrated modulation, which broadens supplier options. Consequently, CW lasers combined with silicon photonics have become the main alternative route for CSPs facing EML shortages. 

However, CW production faces increasing constraints due to several factors: long equipment lead times restrict expansion, and strict reliability standards necessitate labor-intensive die-cutting and aging tests. Consequently, many vendors outsource these steps, which adds to downstream bottlenecks. This situation is causing the CW ecosystem to approach a capacity crunch, leading suppliers to hasten their expansion efforts.” 

Yet, even if EML demand can source elsewhere to alleviate the bottleneck, a company like Lumentum remains in pole position to supply external InP-based CW lasers along with other suppliers. 

We also recently covered a long-haul networking supplier on the Discovery tier that specializes in long-haul networks, an area that Nvidia and Broadcom are unlikely to compete with vendors as it requires expertise in telecom networks and the ability to distribute AI traffic between data centers.  

Another growth opportunity is VSCEL lasers, which are ideal for high-volume and short-reach interconnects as they are low power and low cost. Broadcom recently emphasized VSCEL lasers for their ability to help AI clusters scale beyond copper while still providing short-reach efficiency. This is called near-packaged optics (NPO) and will bridge the limitation of copper today with the 1+ year deployment of co-packaged optics.  

Here is what Broadcom stated – note this is a longer quote but nicely summarizes how the shift toward optics is likely to play out: 

“The insatiable demand for compute power in AI and high-performance computing (HPC) is rapidly approaching a fundamental physical barrier: the limits of copper connectivity. As next-generation XPUs demand bandwidths soaring toward 28.8 Tbps, traditional copper interconnects are struggling to keep pace. 

With SerDes rates reaching 100 Gbps per lane, the effective reach of Direct Attached Copper (DAC) has shrunk to a mere 5 meters. For operators, this restricted electrical signal is a roadblock to building the massive, disaggregated AI clusters required for the next era of innovation. 

While the future of data center connectivity is undeniably optical, the path forward requires a pragmatic approach. Co-Packaged Optics (CPO) remains the "North Star" for energy-efficient, high-bandwidth scaling, but the industry needs a high-performance solution that can be deployed today. 

Vertical Cavity Surface Emitting Laser (VCSEL)-based Near-Package Optics (NPO) serves as that essential bridge. By leveraging readily available 100 Gbps VCSEL technology—with 200 Gbps solutions already in development—NPO offers a practical, high-performance alternative to traditional pluggable optics.” 

Note: We will be publishing a thematic piece on the CPO opportunity later this month. We plan to continue tracking this space closely as it evolves and aim to offer best-in-class execution, particularly given that many of our top winners over the past 12 to 18 months have come from AI networking. 

AI Monetization is Heating Up 

If you and I were in an elevator and I had only a few seconds to explain my view on the AI market, I’d say this: the biggest opportunity for AI stock returns still lies ahead, not behind us. 

A few months back, I wrote in a free newsletter that the greatest risk to an investor is not an AI bubble or the many other headlines that surface and weaken AI stocks, but rather the biggest risk in the current market is that an investor misses out on what be one of the strongest investing opportunities of our lifetime: what I’ve dubbed the AI Monetization Supercycle catalyzed by the inference phase. 

Recently, Reuters reported that OpenAI is seeing more than $25 billion in annualized revenue (although at what margin is still to be seen). OpenAI also recently stated enterprise now makes up 40% of revenue, and adoption on its consumer remains strong with 900 million weekly active users. 

Anthropic also revealed that its revenue run rate has now surpassed $30 billion in early April, more than doubling from February’s $14 billion and up $21 billion since the end of 2025. This helps investors with timing as it illustrates we are moving away from the experimental R&D phase. Notably, this is the fastest ramp in revenue we’ve seen in the tech industry. 

Tokens are the ‘currency’ of this monetization wave, and there’s ample evidence that the market is continuing to underestimate the sheer volume growth ahead in tokens (and thus revenue) as inference-based applications expand and new use cases pop up week after week. Dell’s COO Jeff Clarke provided a great visual on the growth point, explaining that his company had originally modeled inference driving 1 quadrillion tokens by 2028, but now that view has already risen 57X: 

“The demand for inference, long thinking, auto aggressive reasoning models is now requiring more computational intensity. At minimum, at a minimum, 100x, two orders of magnitude greater than we thought less than a year ago. More than two orders of magnitude more than we thought just a year ago. And while that shows up in, in the form of tokens, the measure and what do tokens need, tokens need computational capacity and capability to provide them. We thought as we model this, that inference would drive by 2028, 1 quadrillion, that's 15 zeros, 1 quadrillion tokens. Now it's 57 quadrillion, and I'm sure we're wrong.”  

To wrap your head around just how big 57 quadrillion, if you spent $1 million dollars each and every day, it would take more than 156 million years to reach that number.  

What’s more impressive here is that the token growth is already rapidly accelerating and well on the way to reach that 57 quadrillion estimate. Alphabet revealed in Q4’s earnings in early February that its first party models including Gemini were processing more than 10 billion tokens per minute via direct API use, up more than 40% QoQ from 7 billion per minute in Q3. To apply this to Dell’s framework, that would be more than 5 quadrillion tokens annualized. 

According to a note written by OpenAI’s CFO in early April, their APIs are processing 15 billion tokens per minute, up 2.5X from 6 billion tokens per minute just six months ago in October. This would represent nearly 7.9 quadrillion tokens annualized. 

Combined, Alphabet and OpenAI alone are processing nearly 13 quadrillion tokens annualized, while growing  >40% QoQ and 5X YoY. When taking into account Anthropic, AWS, Azure, OCI and all of the other R&D labs running models in the cloud, we may well be at the 57 quadrillion pace at some point this year.   

Perhaps more important than revenue is the capability of models with OpenAI’s GPT 5.4 “Thinking” model scoring 83% on the GDPVal benchmark, placing it at or above human experts on economically valuable tasks. In Morgan Stanley’s “Intelligence Factory” model, these breakthroughs “lead to an accelerating learning curve, with successive models rapidly outperforming their predecessors”  

In other words, because models are outperforming their predecessors, it makes for a case for more AI spend (not less), supported by the unprecedented revenue trajectories we are seeing from R&D labs now, but likely Big Tech soon too.  

In their last earnings report, Microsoft stated, “We are only at the beginning phases of AI diffusion and already Microsoft has built an AI business that is larger than some of our biggest franchises.”  

Alphabet stated they’ve sold more than 8 million paid Gemini Enterprise seats to more than 2,800 companies with customer interactions growing 65% year-over-year – all within four months. The management team made another comment about the impact AI is having on independent software vendors, stating: “Revenue from AI solutions built by our partners increased nearly 300% year-over-year, and commitments from our top 15 software partners grew more than 16x year-over-year.” 

Initial Signs the Agentic AI Market is Taking Off 

Model context protocol (MCP) is a standard that lets AI models connect to external tools, data and systems. The standard allows AI to have access to tools and to be able to take actions, which marks an important shift to agentic AI. It also will lead to skyrocketing inference demand as the protocol allows AI agents to query databases, make API calls and execute, leading to significantly higher token usage.   

Anthropic released MCP as an open standard in late 2024, with OpenAI adopting it in April of 2025, Microsoft adopted MCP in July of 2025, and AWS in November of 2025. Today, MCP-compatible tooling was donated to the Linux Foundation and is the default protocol for agentic AI with 97 million downloads. MCP is often compared to TCP/IP, or the internet protocol that is owned by nobody yet enabled extensive development across the broader web. Furthermore, without a standardized protocol, developers would have to build observability at an unsustainable level for every tool integration. 

MCP is what powers Claude Code’s workflows, allowing the LLM to use internal databases and external APIs. AI is transitioning toward “autonomous execution engines” by going beyond the terminal command on a computer to now accessing external tools, such as querying a database or sending Slack messages. Each MCP connection results in a bigger and better workflow, adding to the complexity that AI agents can handle autonomously.  

MCP adoption is essentially a leading indicator for the enterprise agentic AI market, as it’s the protocol that deploys AI agents that are capable of production at the system-level. The acceleration in SDKs from 2 million monthly downloads to 97 million across about 18 months is signaling the inference market is beginning to take off.  

Although I can’t promise timing will be exactly in 2026, a reasonable assumption is 2H 2026-2028 time frame. Gartner is more bullish on timing than even myself, stating 40% of enterprise applications will integrate task-specific agents by 2026, up from less than 5% in mid-2025. According to Deloitte, 93% of IT leaders plan to introduce autonomous agents in the next two years, while nearly half have already implemented them. 

Going back to my introduction on execution, and pointing to where I plan to hit the ball, the inference opportunity resembles more of a zero-sum game. We can see evidence of the market agreeing as the software trade has been hit hard lately. The I/O Fund team is cautiously optimistic for a time when software dominates our portfolio again; but that time is not right now.  

AI Energy: Last but Certainly Not Least 

Although I originally thought compute would mark the largest supply-demand imbalance in my career, remarkably, there is another imbalance that carries far more importance. Which brings me to our last thematic trend for Q2: AI Energy. 

Two years ago, the I/O Fund began setting the stage to remain competitive on portfolio returns by introducing energy to our Research Members when we stated: “Big Tech is spending tens of billions quarterly on AI accelerators, which has led to an exponential increase in power consumption. Over the past few months, multiple forecasts and data points reveal soaring data center electricity demand, and surging power consumption. The rise of generative AI and surging GPU shipments is causing data centers to scale from tens of thousands to 100,000-plus accelerators, shifting the emphasis to power as a mission-critical problem to solve.” 

At the time of publishing, there was not even a whisper on the Street about the incoming fundamental bottleneck to AI data center expansion. While those who only follow our free analysis think we’ve been stuck on the Nvidia thesis for years, the truth is Nvidia hasn’t been our leading allocation for quite a while.  

As you know by now, Bloom Energy was one of our biggest winners last year and I reiterated it was a Top Stock Pick for 2026. But it would be too narrow to focus on Bloom as the only destination for our energy allocation. We have many others we want to own when the timing is right.  

Consider this question from my point of view, which is, will there be a time when energy dominates our AI allocation, even above and beyond AI accelerators or networking? Yes, that time is approaching. We want to be fastidious in our analysis now given the I/O Fund tends to be about 2 years earlier than the Street – and well, we are coming up on our first coverage being 2 years ago, indicating the time for energy to lead could be nearer than you might think. 

Briefly Revisiting the AI Energy Thesis 

Up to this point, we’ve hammered on the increase in power requirements across Nvidia’s GPU systems. To keep the math simple, it looks like this: 

• Blackwell doubled the power consumption of the previous Hopper generation from 70 kW to 120 KW-140 kW.  
• Vera Rubin will increase 50% from 180 kW to 230 kW  
• Rubin Ultra racks with 576 GPUs will increase this by roughly 3X to 600 kW by late 2027. That’ll represent 5X in a two-year design timeframe. 

Note, these figures do not include networking, interconnects, cooling and other hardware, which will further boost power draw per rack. 

The numbers above reveal Nvidia’s aggressive product road map is set to release new GPUs every 1-2 years, yet grid infrastructure operates on a 5-10 year timeline. For example, in its largest configuration, the Vera Rubin NVL576, dubbed the ‘Kyber’ rack, could draw as much as 600kW or 5x that of the GB200 NVL72 in just a two-year design timeframe. 

Therein lies the compounding problem that brought us to Bloom Energy, which is that current data centers are fitted for 50kW racks and need to be retrofitted for 600kW racks. This will move to 1 MW racks as AI servers are expected to use over 1000 kW of power in the Feynman architecture due out in 2028, which represents 8X the power requirements of Blackwell in about 4 years' time. 

In the Q1 2026 Top 15 AI Stocks Report, I emphasized the timing issue the AI data center buildout faces specifically between 2026-2029: 

“For example, across the board, developers are expecting to have power delivered by late 2026 to early 2027 on average, with most regions seeing expectations as early as late 2025. This is likely driven by consistent strong demand for AI infrastructure services, as new capacity will allow hyperscalers to meet more demand and drive more revenue.  

Yet, utilities do not expect to be able to meet these delivery timelines in most of these primary and secondary markets, with many projecting late 2027 through 2028, with major hub Northern Virginia seeing one of the longest timelines at nearly 2029.” 

Since I wrote that in mid-January, the problem has only been exacerbated. The capex raise we saw in late January through early February of nearly 40% to $600 billion is bullish for key suppliers but also puts into question how to power an increase in compute that is into the hundreds of billions year-over-year.  

For example, following capex raises, this estimate from Congress.Gov on January 23rd is likely outdated, which states that U.S. data center energy use comprised 4.4% of the United States annual electricity consumption yet is expected to consume 12% by 2028 for a 3X increase. 

Evidence Mounts on the Incoming Energy Shortage 

Evidence of an energy shortage is growing each quarter. For example, we covered in a Discovery tier analysis that PJM clearing prices have surged to the tune of 11X over the past two years. This began in the 2025/26 auction, where clearing prices jumped 833% from $28.92/MW-day to $269.17/MW-day, reaching the annual cap. The 2026/27 auction saw prices once again hit the FERC-approved cap at $329.17/MW-day, a 22% YoY increase. 

Skyrocketing power prices and elevated risk of grid shortfalls from a fifth consecutive year of declining supply puts major emphasis on adding new capacity to the grid. What the Street (and the public) have not realized yet is the extent of the incoming capacity that is being claimed by data centers. PJM reported in August that it's long-term projected load growth from 2024 through 2030 would be 32GW, with 30GW of that coming from data centers, assuming many data center projects materialize on time.  

However, the problem here is that PJM’s forecasting has recently underestimated peak demand growth, even with significant upward revisions over the last few years. For example, realized peak demand is already approaching 160 GW, nearly two years ahead of current forecasts, and if data center builds progress at current (or accelerated) paces, peak load may continue to outpace forecasts through 2030. 

This helps explain why the market is not unilaterally rewarding Nvidia for capex raises in the same manner as during the Hopper architecture. Not only will a higher percentage of capex be allocated to energy; but it’s also a puzzle as to how this will be perfectly resolved. We had noted in a previous analysis the risk is that GPUs sit idle: 

“If a company like Microsoft buys tens of billions of Nvidia’s Blackwell GPUs, the longer the massive investment in GPUs waits for power, the more delayed that revenue and profits become. In turn, this plays into market share as competitors who can energize GPUs faster will have a critical head start over those that are waiting for power. This is simple in concept, yet the lack of power having vast consequences cannot be overstated if you combine the sheer size of investments being made in AI alongside fierce, heightened competition. 

AI is a spending race, but this means it is at the core, a power race. It does not matter if a hyperscaler spends tens of billions more on capex if it cannot secure the power to stand up new data center infrastructure to then deploy those GPUs immediately. The AI market is officially moving from being compute constrained to being power constrained, and this shift is important for I/O Fund members to prepare for.” –Why Power is Critical for Data CentersWhy Power is Critical for Data Centers 

Notably, most growth investors do not have experience generating returns in the energy sector. It has long been one of the most difficult industries to find alpha, given its lumpy cycles, commodity sensitivity, and heavy exposure to regulation. 

That is where the I/O Fund’s flexibility becomes a meaningful advantage. Our process is designed to be early to stocksearly to stocks (as opposed to a process that is designed to establish expertise in only one domain). This has allowed us to find winners year after year in every subsector of technology. You can fully expect us to use our flexible yet effective process in the energy sector over the coming years. 

Large-Scale Utility:

Large-scale utility is naturally the first subsector to think of to address the energy crisis. Utilities benefit when there is this much demand because one thing utilities do well is deliver reliably.  

As discussed, the PJM clearing prices saw prices surge 11X over the past two years when you combine the 2025/2026 prices jump 833% to $269.17/MW-day and then again 22% in the 2026/2027 auction prices to $333.44/MW-day. When you consider zones that are further constrained, such as BGE and Dominion, the prices increased even further to $465.35/MW-day and $444.26/MW-day, respectively, in the 2026/2027 auction. 

This represents a cap for the $333.44 MW-day as the auction pricing would have hit $530/MW-day without the cap. 

When we look at large-scale utility stocks such as Talen, Constellation Energy or Vistra, the high auction prices are leading to higher prices on their existing assets. Although that may seem intuitive, the key here is they don’t have to spend more on capex to report higher profits – which is a rarity right now. 

Utilities like Constellation Energy, Vistra and Talen earn capacity revenue in addition to energy revenue. The retainer payment correlates to the following capacities: 

• Constellation cleared 17,950 MW in the latest auction for roughly $2.2 billion in capacity revenue for 27/28. 
• Vista cleared 10,566 MW in the latest auction for roughly $1.3 billion in capacity revenue for the 27/28 year 
• Talen cleared 8,745 MW in the latest auction for about $1.1 billion. 

According to a recent report from Morningstar, energy prices increased from $33.74 MWh to $50.73 MWh in 2025, for an increase of 50.4%. If we break the report down further, we see that capacity costs are skyrocketing. 

• Energy is 59.6 percent of the cost of wholesale power and rose 51.2% 
• Capacity is 15.8 percent of the cost of wholesale power and rose 262.3% 
• Transmission is 22.4 percent of the cost of wholesale power and rose 4.5% 

Capacity is where the auction pricing surge is showing up as energy companies are paid more just for having capacity exist. When energy pricing goes up, it’s because energy costs more. However, capacity pricing is surging while Utility companies do very little to justify this increase as there were no new plants built or new upgrades. Conversely, Utilities benefit from not adding new plants as it creates scarcity pricing. 

As you can imagine, residential and commercial customers in data center regions aren’t too happy about having to absorb capacity pricing, which is allocated more broadly rather than tied to usage. For example, according to a recent report from IEEFA, the residential bill in Maryland is expected to go up $18/month and up $16/month in Ohio, as " ratepayers across the region will collectively be paying an additional $1.4 billion in capacity market costs, again driven largely by data center demand.” 

This is important to consider as investors typically want to see uncapped growth in a stock, yet growing concerns from consumers could lead to another capped PJM auction come June 2026. 

Talen is more of a pureplay for PJM auction pricing compared to Constellation which includes the recent Calpine acquisition, a retail business and is exposed to many markets outside of PJM like gas-heavy ERCOT and CAISO. Although Talen is not entirely PJM, it’s heavily centered in this market, which helps to strip out the revenue trajectory seen below. The main takeaway is the bulk of the growth is accounted for in the strong 2026 year, yet there is a leveling out in future years as it implies capped auction pricing.  

However, profits are expected to outpace revenue growth, which means despite Utilities not fitting the growth characteristics we typically seek at the I/O Fund, there could be times their defensibility is attractive. 

Behind-the-Meter: 

There exists a significant disconnect between when hyperscale and colocation developers expect to have site power, and when large-scale utilities expect to be able to deliver. This means even if you want to build a new facility for 600kW racks, getting power to it is a multi-year ordeal.  

Due to time constraints, connecting new data centers to the grid is not the most feasible option for hyperscalers looking to deploy gigawatts of capacity quickly, and instead, alternative power sources in the near term will be in higher demand. This is supported by research from TD Cowen regarding grid connection timelines for new data centers, which span anywhere from 36 months to 48 months in these markets. In 2024, Bloomberg reported that utility Dominion Energy said >100MW data centers in Virginia were facing up to seven year wait times for new connection hookups. 

This has led to firms like McKinsey predicting 25% to 33% of net new generation will come from behind-the-meter solutions by 2030. This is significant considering the market was effectively zero going into the AI boom. 

We've covered the advantages of behind-the-meter and on-site power generation for about two years. Briefly, it refers to bypassing the grid entirely by generating power on-site or nearby with solutions like natural gas turbines, small modulator nuclear reactors, fuel cells and solar/batteries. These solutions don’t require the grid and can be deployed in 1-2 years time or as quickly as 3 months in Bloom Energy’s case. 

In our September analysis, we began to answer the question of how many GWs correlate to capex spending. For 2025, we had projected Big Tech could bring on 7-9 GWs with 2025 capex, and then 9-12 GWs with 2026 projected capex, for a total of 16-21 GWs across the two years. We raised this another 4GWs to include Oracle for a total of 20-25GWs for 2025 and 2026, yet this still does not include xAI, CoreWeave or Nebius. 

Given the capex increases we saw in January, that 2026 number has nearly doubled, even without including the neoclouds, with our September framework projecting Big Tech and Oracle could bring on 16-20GW this year alone. Cumulatively, that could bring total capacity additions in 2025 and 2026 to 23-29GW.  

Put another way, any raise in capex spending over the next two years puts greater emphasis on behind-the-meter solutions.

The Case for the Miners 

The term “don’t throw the baby out with the bath water” could apply to Bitcoin Miners. Despite weak price action over the past few months, it’s hard to imagine a way forward without utilizing these brownfield sites.  

Applied Digital pointed out that less than two percent of data centers have racks with greater than 50kW. Where retrofitting is attractive is not necessarily the costs associated with construction but rather that the turnaround time for using an existing facility can be shortened from 4-7 years to 18 months to 30 months. 

Bitcoin mining is not behind-the-meter in a strict sense, but it is effectively behind-the-meter because miners secure direct, wholesale power through upfront contracts that are not renegotiated. They are also considered on-site power as there is minimal transmission dependency due to co-locating near a mix of power sources (near gas plants, augmented by wind, solar, water/hydro). In most cases, even if a utility meter exists, the power system is purpose-built for that site and is not shared retail infrastructure. This also helps equal the playing field as the biggest drawback for Bitcoin Miners is they need extensive retrofitting.  

You’ve heard me use the words “time to power” before to describe Bloom Energy’s investment thesis. Bitcoin Miners are similar – they offer time to power. Not only do Bitcoin Miners offer speed, but they also offer below-market rates to hyperscalers. 

Regarding the weak balance sheets, for the very best power sites, this will transform quickly as these companies are transitioning from volatility that is characteristic of Bitcoin to AI contracts that provide fixed, recurring revenue with 80% to 90% operating margins. What will matter is if the market is willing to re-rate these stocks based on Big Tech being the collateral backing them as we are seeing about $3 billion or more in convertible notes and long-term debt in some of these names with very little revenue (yet). 

The debt plays a big part as to why Bitcoin Miners go in and out of fashion, but underneath the debt is execution risk. We will often see large revenue targets offered by the management teams, yet the market is essentially saying – we need more evidence you can deliver what you say you will.  

In terms of how we plan to play this, not much has changed in terms of the risk profile of Miners and our overall investment strategy: 

“Right now, we prefer to stay as close to the hyperscaler deals as possible when evaluating Bitcoin Miners. The reason for this is that it solves the pain point of having a company with deep pockets back-stop the leases, which in turn, improves creditworthiness and credit terms. As many of you are aware, our ethos is to participate in the upside while protecting to the downside. We want the best of both worlds, and in a highly speculative momentum play like Bitcoin Miners pivoting to AI data center infrastructure, the primary goal is to reduce risk.” -September 2025 Discovery Analysis on Bitcoin MinerSeptember 2025 Discovery Analysis on Bitcoin Miner 

Join the Discovery tier for early access to stock ideas and to stay ahead of where the market is heading next. To subscribe to Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40 Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40 

Top 15 AI Stocks List 

Section 1: AI Accelerator Stocks 

Broadcom: Strong Contender for First Place 

Broadcom guided to $22 billion in FQ2 revenue up 47% YoY and adjusted EBITDA at 68% of revenue. Within that, management guided semiconductor revenue to $14.8 billion, up 76% YoY, and AI revenue to $10.7 billion, up 140% YoY, indicating an acceleration from Q1.   

The most explosive comment was this: “Today, in fact, we have line of sight to achieve AI revenue from chips, just chips in excess of $100 billion in 2027. We have also secured the supply chain required to achieve this.”   

Management characterized this demand as being driven by a small number of hyperscalers  and frontier model builders, with both training and inference contributing as those customers will soon productize their LLM platforms. Within discussing the impressive customer list, the CEO of Broadcom hinted toward 2027 being significantly higher than $100 billion – plus another analyst did math that would show a sharp inflection in 2028 due to OpenAI’s incoming GWs. 

In our last Quarterly write-up, it was stated that Morgan Stanley now expects 5 million TPUs to be shipped in 2027, a 67% rise from its prior estimate for 3 million; for 2028, the firm estimates shipments as high as 7 million, a 120% increase from its prior estimate. This would project YoY growth of 40% from 2027 to 2028, a substantial increase from 6% previously, and will represent more than 2X growth in two years. 

More recently, Hong Kong-based GF Securities stated that they now expect total TPU shipments to be 4.5 million/7.9 million units in 2026E/2027E, up from previous estimates of 4.5 million/6 million. The upward revision is primarily driven by external sales. For Broadcom, they expect its TPUs to be 4.1 million/5.8 million for 2026E/2027E. 

We had stated the estimates provided in the Thematic section were a bit aggressive, yet rather it lands at 6 million or higher for Broadcom in FY2027, the direction is firmly up. Even with MediaTek potentially taking some TPU business on the inference side, Broadcom is in pole position across many hyperscaler customers. 

Revenue: 

Broadcom’s FQ1 ending January 2026 revenue grew by 29.5% YoY and 7.2% QoQ to $19.3 billion, beating estimates by 0.9%. Revenue growth accelerated by 1.3 percentage points from 28.2% growth in the previous quarter.   

Management provided a strong FQ2 guide of $22 billion, implying a YoY growth of 46.6% and 13.9% QoQ, beating estimates by 7.8%.  

The expected strong growth is primarily driven by AI revenue, which is expected to grow 140% YoY and 27% QoQ to $10.7 billion. Analysts expect strong revenue growth to continue, with FQ3 revenue expected to grow 81.8% YoY $29 billion and 87.5% YoY to $33.8 billion in FQ4. 

AI Revenue: 

AI revenue was at $8.4 billion this quarter and is guided to $10.7 billion next quarter for a run rate of about $43 billion. On the surface, the guide doesn’t look like much and would imply a deceleration given AI is growing at a rate of 140% YoY and 27% QoQ – whereas this is effectively saying Broadcom will double in 7-8 quarters.   

However, the easy-to-miss details on the guide is that the $100 billion is only for silicon and does not include networking. The words “significantly in excess” were also added later to the guide in the following statement during the Q&A portion. 

According to the earnings call, networking is about 33% to 40% of AI revenue today: “AI networking revenue grew 60% year-on-year and represented 1/3 of total AI revenue. In Q2, we project AI networking to accelerate a lot more and grow to 40% of total AI revenue.”  

If we assume this mix continues on the low end for about 30% mix in AI networking of total AI revenue, then it’s reasonable to assume Broadcom’s AI revenue will be $143 billion with networking of $43 billion (or 30% of $143B). This represents a QoQ growth of 30% for 7-8 quarters – which is an excellent baseline to set. 

Earnings: 

FQ1 GAAP EPS grew by 31.6% YoY to $1.50. Adjusted EPS grew by 28.1% YoY to $2.05, beating estimates by 1.3%, primarily driven by operating leverage. 

Margins: 

The company’s adjusted EBITDA margins beat management guidance in FQ1, primarily driven by operating leverage.  

Gross profit margin improved by 10 basis points YoY and QoQ to 68.1%. Adjusted gross margin came at 77%, down 210 basis points YoY and 90 basis points QoQ and marginally beat the guidance by 10 basis points. 

Operating margin improved by 230 basis points YoY and 260 basis points QoQ to 44.3% primarily driven by operating leverage. The adjusted operating margin was 66.4%, compared to 65.9% in the same period last year and 66.2% in the previous quarter.  

FQ1 net income grew by 33.5% YoY to $7.35 billion with a net profit margin of 38.1% compared to 36.9% in the same period last year. 

FQ1 adjusted EBITDA grew by 30.2% YoY to $13.1 billion with an adjusted EBITDA margin of 68% and was better than the management guidance of 67%.  

Cash: 

Broadcom’s cash flows are improving, driven by higher profits.  

FQ1 operating cash flows grew by 35.1% YoY to $8.26 billion with an operating cash flow margin of 42.8% compared to 41% in the same period last year. 

FQ1 free cash flows grew by 33.2% YoY to $8.01 billion with a free cash flow margin of 41.5% compared to 40.3% in the same period last year.  

Cash was $14.2 billion at the end of FQ1 with debt of $66.1 billion compared to cash of $16.2 billion and debt of $65.1 billion at the end of FQ4. The company repurchased shares worth $7.85 billion and paid dividends of $3.1 billion in the recent quarter. 

Valuation: 

Broadcom trades at a forward P/S ratio of 16.9. The company traded at a minimum forward P/S ratio of 6.7 and the maximum of 28.8 in recent years. Broadcom is currently trading around the mid-range. On the bottom line, it is trading at a forward P/E ratio of 32.6. The company traded at a minimum forward P/E ratio of 17.3 and a maximum of 57.2. Broadcom is trading slightly lower than the mid-range on the forward P/E ratio. 

Notable Risks: 

Broadcom’s debt load has increased following past acquisitions, which adds balance-sheet risk. That said, the company has a strong track record of deleveraging and generates substantial cash flow, which helps offset this concern. 

Google is deliberately diversifying away from depending solely on Broadcom, giving itself more leverage on pricing and supply chain resilience. MediaTek is a clear winner in this shift, but Broadcom retains a meaningful role in the core TPU architecture for now, and Broadcom is also growing in importance with other hyperscalers such as Meta. 

Arm Stock Could Win as Agentic AI Shifts the Bottleneck to CPUs

Arm unveiled an AGI CPU last month to address one of AI’s biggest bottlenecks, which is orchestration. During the chatbot craze of 2023-2025, GPUs did most of the heavy lifting while CPUs had become an afterthought. Yet with agentic workloads, which is perhaps the single largest catalyst on the horizon for the AI trade in 2026 and beyond, the importance of CPUs is set to increase.  

In agentic workflows, the GPU still handles inference, but between each inference call, the CPU is doing the orchestration – which are best described as handling tool calls, API requests and memory tasks. AI agents are surfacing this new constraint, which is how to prevent latency and underutilized GPUs following the exponential growth of orchestration needs. 

For investors, what matters is that CPUs account for 50% to 90% of total latency in workflows, which means the CPU-to-GPU ratio in AI clusters will need to increase. Earlier this year, both AMD and Intel saw analyst upgrades based on the outstripped supply of CPUs leading to higher average sales prices of roughly 10% to 15%. Reuters also reported that Intel’s unfulfilled orders are reaching longer than six months while AMD delivery times are believed to be eight to 10 weeks. 

Regarding how Arm fits in, the company’s expertise in lowering power requirements could matter more than the market expects. After years of supplying the architecture IP behind other companies’ CPUs, Arm is preparing to directly compete with its customers and x86 CPU competitors by transitioning to a chip designer themselves. This comes during a time when CPU cores are expected to go up 4X from 30 million CPU cores per gigawatt to 120 million CPU cores per GW. 

Revenue: 

Q3 FY2026 ending December revenue was up 26.1% YoY and 9.4% QoQ to $1.242 billion, representing a record quarter for revenue and exceeding $1 billion for the fourth consecutive quarter. 

Key Advantages of Arm’s ‘AGI CPU’ for Agentic AI Workloads 

Arm also marked its long-awaited foray into physical chip development with its ‘AGI CPU’, launched at its Arm Everywhere event last month. The company’s pivot into physical CPU and rack development is one the AI industry will watch with great anticipation given Arm’s history of owning significant IP in the mobile space combined with the company setting out to solve agentic AI’s orchestration challenges.  

Leveraging Arm’s history of delivering high performance with low power requirements for mobile devices, the new AGI CPU is designed to offer a similar balance between high performance and low power consumption.   

The AGI CPU was co-developed with key partner Meta, the chip’s first customer, who revealed they turned to Arm almost two-and-a-half years ago to see if there was a CPU option that fit Meta’s needs: “put in a lot more cores per watt, but we do not want to compromise on the performance piece.” Meta had only been finding options satisfying one of the two criteria: meeting the performance but with too much power, or meeting the power but with too little performance. 

Margins: 

Arm has a profitable business model that constitutes licensing revenue and royalty revenue. The company reported a strong gross margin of 97.6% in Q3 FY2026 ending December.   

Arm reported a GAAP operating margin of 14.9% and an adjusted operating margin of 40.7% in the recent quarter. The difference between adjusted operating margin and GAAP operating margin is that the company is a recent IPO and has high stock-based compensation of $285 million or 23% of revenue. 

Cash: 

The company’s cash flows have been lumpy due to high working capital and high capex to support the long-term growth. However, with the expected strong future profit growth, the cash flows should improve. The company also has a strong balance sheet with cash & short-term investments of $3.54 billion and no debt. 

Valuation: 

Arm is currently trading at a P/S ratio of 37.1 and a forward P/S ratio of 29.1. The company is trading significantly higher than its other semiconductor peers like Broadcom’s forward P/S ratio of 16.9 and Nvidia’s forward P/S ratio of 12.4.  

The company’s revenue growth is expected to accelerate in the next five years compared to the previous period. The company’s revenue CAGR has been 19.3% from FY2021 to FY2026E. Analysts expect revenue to grow at a CAGR of 34% from FY2026E to FY2031E and will be even higher at 38.5% if we use the $25 billion management guidance. However, when looking at the AI segment of many semiconductor peers, the growth rate does not stand out, per se, to justify the high valuation. Rather, the consistency of licensing and royalties' revenue does stand out, and this recurring revenue will create a nice baseline when you combine higher growth from their merchant CPUs. 

Notable Risks: 

The company’s cash flows have been lumpy due to high working capital and high capex to support the long-term growth. 

AMD: Underestimated and Largely Misunderstood 

About 18 months ago, I spelled out AMD could outpace Nvidia’s returns by 2030 stating in a Real Vision video interview that the company’s opportunity is closely tied to the inference market.  

The overall thesis is that the data center GPU market desperately needs a second-place contender. Investors may appreciate Nvidia’s pricing power, but hyperscalers and companies like OpenAI do not; they’d like to see more competition and optionality including lower prices. That is why we are seeing Meta work alongside AMD to bring Helios to market and a recent 6GW deal from OpenAI.  

One key area where Helios stands out is memory — the platform offers roughly 50% more total memory capacity compared to Nvidia’s Vera Rubin rack architecture. AMD will offer 1.4 PB/s of memory bandwidth, slightly below Rubin’s 1.6 PB/s as Nvidia is said to be requiring pin speeds of 11 Gb/s, above the standard 8 Gb/s, driving the higher bandwidth despite lower HBM content. The HBM content and nearly comparable bandwidth will likely make AMD a compelling solution for inference workloads considering its price-advantage over Nvidia. 

Buried in the most recent earnings call was a rather strong statement for this otherwise-conservative management team that AMD is “well positioned” to grow data center revenue by more than 60% annually over a 3-5 year time frame:  

“With the launch of MI400 series and Helios representing a major inflection point for the business, as we deliver leadership performance and TCO at the chip compute tray and rack level. Based on the strength of our EPYC and Instinct road maps, we are well positioned to grow data center segment revenue by more than 60% annually over the next 3 to 5 years, and scale our AI business to tens of billions in annual revenue in 2027.”  

Given the strength of the comment, an analyst asked about the comment on the call and if the 60% applies to 2026 with management replying this is certainly possible:  

“We're not obviously guiding specifically by segment, but the long-term target of, let's call it, greater than 60% is certainly possible in 2026.” 

Per my last earnings writeup: “AMD Q3: The Catalyst is Expected in H2 2026," which stated, “AMD is a stock where I’ve been intentional about managing expectations. The upside is compelling — as the second place in data center GPUs is wide open. Yet for those who have followed our coverage, the timing has always been key: meaningful execution in AI accelerators is not expected to materialize until the second half of 2026. In other words, the long-term opportunity is substantial, but patience remains part of the thesis.” 

Revenue: 

AMD’s Q4 revenue grew by 34.1% YoY and 11.1% QoQ to $10.27 billion, beating estimates by 6.2%. However, the company’s revenue this quarter included approximately $390 million from MI308 sales to China and excluding this revenue since it was not included in the guidance would yield only a 2.2% beat, the smallest in the last four quarters. 

Management guided Q1 revenue of $9.8 billion at the midpoint, implying a YoY growth of 31.8% YoY and down (4.6%) QoQ and the guidance includes about $100 million of MI308 chip sales to China. 

AI Revenue: 

The company’s Data Center segment revenue grew by 39% YoY and 24% QoQ to a record $5.4 billion, led by accelerating Instinct MI350 Series GPU deployments and server share gains. However, it included MI308 chips sales to China, otherwise, would be only 29% YoY and 15% QoQ growth. MI450 ramp is expected in the second half of the year, particularly in Q4. The company remains on track to launch its MI500 chips in 2027.  

Earnings: 

The company’s Q4 adjusted EPS grew by 40.4% YoY to $1.53 primarily driven by operating leverage, beating estimates by 16%.  

Analysts expect Q1 adjusted EPS to grow 32.7% YoY to $1.27 and 195.7% YoY to $1.42 in Q2 2026. 

Margins: 

The company’s profits are growing. However, near term margins are negatively impacted by higher operating expenses to support strong future AI opportunities. Management expects margins to improve by the end of Q4 due to favorable product mix, particularly the ramp of MI450 chips.  

Q4 gross profits grew by 44% YoY and 17% QoQ to $5.58 billion. Adjusted gross profits grew by 41% YoY and 17% QoQ to $5.86 billion. Excluding the inventory reserve release and MI308 revenue from China, gross margin would have been 55%, up 100 basis points YoY and QoQ. Management has guided 55% adjusted gross margin in Q1. 

Operating margin improved by 600 basis points YoY and 300 basis points QoQ to 17%. Adjusted operating margin improved by 200 basis points YoY and 400 basis points QoQ to 28%. Management has guided an adjusted operating margin of 24% in Q1. The company’s near-term margins are negatively impacted by higher operating expenses to support strong future AI opportunities. 

Cash: 

Q4 operating cash flow grew by 77% YoY to $2.3 billion with an operating cash flow margin of 22%, up 500 basis points YoY and 300 basis points QoQ. 

Q4 free cash flow grew by 91% YoY to $2.1 billion with a free cash flow margin of 20%, up 600 basis points YoY and 300 basis points QoQ. 

The company has cash and short-term investments of $10.5 billion, up from $7.24 billion in Q3. While debt remained the same at $3.22 billion. 

Valuation: 

AMD is trading at a forward P/S ratio of 8.6. The company has traded at a minimum of 3.7 and a maximum of 13.3 in recent years. AMD is currently trading at mid-range. On the bottom-line, it is trading at a forward P/E ratio of 36.6. The company has traded at a minimum of 19.7 and a maximum of 66.4 in recent years. AMD is trading slightly lower than the mid-range on a forward P/E ratio. 

Notable Risks: 

AMD faces constraints around packaging capacity, particularly around CoWoS, where industry constraints can limit how quickly advanced AI accelerators are brought to market. There is also execution risk, as AMD must take on Nvidia. In addition, AMD’s AI mix may carry lower margins than investors prefer, especially as the company competes aggressively on price and invests to gain share.  

Lastly, Arm-based CPUs present a competitive risk in the server market, as hyperscalers continue exploring alternative architectures that could pressure x86 share over time. 

Nvidia: Seeking to Defend its Throne 

Inventories increased more than 8% QoQ to $21.4 billion, but more importantly, Nvidia’s supply related commitments surged. We highlighted this last quarter as a key sign that the strong data center QoQ revenue inflection would continue.   

In Q4, Nvidia’s supply-related commitments surged nearly 90% sequentially to $95.2 billion, a major step-up from the prior ~$28-30 billion range through late FY25 and the first half of FY26. Nvidia says it is strategically securing inventory and capacity to meet demand beyond the next several quarters, which we believe serves as a key sign that the current accelerated QoQ data center growth of ~$10 billion will likely persist as Blackwell Ultra continues ramping and as Vera Rubin ramps. 

While initially, this could be taken as evidence that Blackwell’s ramp is persisting; the more likely outcome now is that it signals a Rubin delay. If this is true, the risk is that it sits on the balance sheet until Rubin ships. However, the more likely scenario is that most of these commitments could be converted to Blackwell and Blackwell Ultra orders.  

TrendForce data supports this theory, stating that industry watchers expect Rubin to account for 22 percent of Nvidia’s high-end GPUs, down from 29 percent. As stated in our Thematic section, the reason stated is: “time required to validate the newer HBM4 memory used by the chips, challenges with the migration to Nvidia's faster ConnectX-9 NICs, the system's higher overall power consumption, and the more advanced liquid cooling requirements as contributing to the delays.”  

In the same article, the stated assumption is that Blackwell mix rises to 71% while Hopper is down to 7% from original expectations of 10% due to China tensions. 

According to additional checks, this is aligned with Keybanc, stating 2026 supply is expected to support "5.5M-6M Blackwell GPUs, 1.5M Rubin, and 1M Hopper GPUs.” KeyBanc’s estimates imply higher Hopper revenue – which is what could sting slightly – as these numbers would make up roughly 69% to 71% of Nvidia’s 2026 GPU output, while Rubin accounts for about 18% to 19% and Hopper about 12%. Keybanc also cut VR rack estimates by 50% to 6K, down from 12-14K.  

Overall Revenue Growth: 

Nvidia’s Q4 revenue accelerated to 73.2% YoY from 62.5% YoY in Q3, while QoQ growth moderated slightly to 20% QoQ from 22% in Q3, due to Nvidia’s increasing revenue base. Revenue for the quarter was $68.13 billion, beating estimates by 2.9%.  

For Q1, Nvidia guided revenue growth to accelerate further to 77% YoY, forecasting revenue to be $78 billion, +/- 2%, coming in well ahead of consensus for $72.03 billion. 

Sequential growth would again moderate to 14.5% QoQ at the midpoint of guidance, though again this is partly due to the law of large numbers as dollar growth is projected to be nearly $10 billion QoQ, versus $11.1 billion in Q4. 

AI Segment Growth: 

Data Center revenue in the quarter was $62.31 billion, with YoY growth accelerating nine points to 75% YoY although QoQ growth moderated 3 points to 22% QoQ, due to the larger revenue base. This compares to 25% QoQ growth last quarter, marking two strong back-to-back quarters from Blackwell Ultra shipping in volume.  

Within Data Center, Compute revenue rose 58% YoY and 19% QoQ to $51.33 billion, slowing from 27% QoQ in Q3 though YoY growth accelerated 2 points.  

Networking revenue was once again quite an outlier in Q4’s report, with growth accelerating sharply on both a YoY and QoQ basis in the quarter. Networking revenue was $10.98 billion in Q4, up 34% QoQ and 263% YoY – this represents more than a 100 point acceleration from 162% YoY growth in Q3, while QoQ growth accelerated 21 points. Nvidia said the strong growth stemmed from the introduction and ramp of NVLink compute fabric for both GB200 and GB300 systems, as well as growth in Ethernet, InfiniBand and Spectrum-X.  

On the call, it was stated that Nvidia is likely the largest Ethernet company in the world. 

Earnings: 

GAAP EPS saw a rather large beat in Q4, coming in at $1.76, up 98% YoY and beating estimates for $1.47 by more than 19%.  

Adjusted EPS was $1.62, up 82% YoY and beating estimates by just over 5%. Growth accelerated from 60.5% in Q3 and marked Nvidia’s fastest adjusted EPS growth in the last five quarters. 

Margins: 

Nvidia’s gross margins moved higher in Q4 to the 75% range, with GAAP operating margin following this expansion and moving back to the 65% level. To note, starting in Q1, Nvidia’s adjusted margin figures will include SBC.   

Q4 GAAP gross margin was 75%, slightly ahead of management’s guidance for 74.8% and expanding by 2 points YoY and 1.6 points QoQ, with the sequential expansion driven by Blackwell and better product mix. 

Q4 GAAP operating margin was 65%, also slightly ahead of guidance for 64.5%, and expanding 3.9 points YoY and 1.8 points QoQ, showing a hint of operating leverage. For Q1, Nvidia guided for operating margins to be flat at 65% and since the adjusted margin figures will include SBC, it will be lower sequentially at 65.4%.  

Q4 GAAP net margin was 63.1%, expanding 6.9 points YoY and 7.1 points QoQ, while adjusted net margin was 57.2%, up 1.1 points YoY and 1.5 points QoQ. 

Cash: 

Cash flows were robust in Q4, with cash flow margins improving significantly on both a YoY and QoQ basis. Operating cash flow was $36.2 billion in Q4 for a 53.1% margin, up 10.9 points YoY and 11.4 points QoQ.  

Free cash flow was $34.9 billion for a 51.2% margin, up 11.7 points YoY and 12.4 points QoQ. Cash and equivalents totaled $62.6 billion, while debt was $8.47 billion. 

Valuation: 

Nvidia trades at a forward P/S ratio of 12.4. The company has traded at a minimum forward P/S ratio of 10.8 and a maximum of 28.3 in recent years. Nvidia is currently trading significantly lower than mid-range. On the bottom line, it trades at a forward P/E ratio of 22.7. Nvidia has traded at a minimum forward P/E ratio of 19.9 and a maximum of 50.6 in recent years. Nvidia is currently trading significantly lower than mid-range.  

Notable Risks: 

Net-net on the Rubin delay: Given Blackwell backfill, revenue may not be heavily impacted yet the optics around the delay could lead to more diversification into custom programs and AMD GPUs. Rubin systems go for a higher average sales price, yet the bigger issue isn’t losing the markup in the near-term but rather: 1) is the delay truly only one quarter (we’ve been here before and the delay was longer) and 2) Nvidia’s product road map is not seen as invincible. If you recall, I stated the product road map is the second line of defense should the CUDA moat be breached.  

Our catalysts to the $20 trillion thesis remain – which is a strong product road map, analyst estimates being far too low in the 2028-2030 window, but even more importantly, my prediction is that Nvidia exits the decade as one of the largest AI software companies. We saw how quickly the company took over Broadcom as the largest Ethernet companies; something similar is what my $20 trillion thesis hinges on, but in robotics and automation.  

TSMC: The Importance of CoWoS Capacity 

The main challenge for AMD and its data center growth boils down to capacity at TSMC on the CoWoS side. Not only does CoWoS capacity remain tight, but Nvidia is locking in a majority of TSMC’s capacity, leaving AMD, Broadcom, Google, and others to fight for its scraps. For example, analysts from KeyBanc estimate that Nvidia has secured ~650K wafers in 2026, up 76% YoY, whereas AMD’s 2026 allocation is estimated to be 80K, up barely 14% YoY.  

Other reports suggest Nvidia’s allocation is around ~595K and AMD at 105K in total, with 80K at TSMC and the rest at other OSATs; regardless of the exact split, the fact is that AMD’s CoWoS allocations are a fraction of Nvidia’s this year.  

To put this in perspective with percentages, TSMC is expected to ramp CoWoS capacity from ~75-80K per month at the end of 2025 to ~130K per month by the end of 2026, so it’s likely that its current capacity is closing in on the ~100K per month level. Thus, Nvidia would be locking up more than 50% of current supply at 650K, with AMD getting less than 7%.  

This headwind may ease come 2027, with AMD’s allocation projected to rise as much as 70% YoY, per KeyBanc, taking its CoWoS wafers to ~136K, supporting higher GPU volumes and thus revenues. On the other hand, KeyBanc estimates Nvidia’s allocation to rise to 840K in 2027, still more than 6X AMD’s. 

Revenue: 

Q1 revenue grew by 40.6% YoY and 6.4% QoQ to $35.9 billion, beating the mid-point guidance by 2%, primarily driven by strong AI demand. 

Management guided Q2 revenue of $39 billion to $40.2 billion, implying a YoY growth of 31.7% and 10.3% QoQ. 

AI Revenue: 

HPC revenue increased 20% QoQ in NT$ to account for 61% of the Q1 revenue. 

Management mentioned that AI-related demand is robust and increased the company’s full year total revenue guidance to above 30% growth in U.S. dollar terms from the earlier close to 30% provided during Q4 earnings.

Margins: 

TSMC’s ability to generate exceptionally strong profits showcases that the company is one of the best-managed companies in the world. Despite the rising inflation, tariff concerns, technological advancement, trade wars, overseas fab expansion, and geopolitical tensions, TSMC has overcome these challenges by continuing to generate superior profits. Margins continue to expand due to cost controls, higher capacity utilization rates, economies of scale, and better price negotiation with customers and suppliers. 

Q1 gross margin was 66.2%, up 7.4 percentage points YoY and 3.9 percentage points QoQ primarily due to cost improvement efforts and better capacity utilization rate. 

Q1 operating margin improved by 9.6 percentage points YoY and 4.1 percentage points QoQ to 58.1% primarily due to operating leverage.  

Q1 net profit margin improved by 7.4 percentage points YoY and 2.2 percentage points QoQ to 50.5%. 

Earnings: 

Q1 GAAP EPS grew by 64.6% YoY to $3.49, beating estimates by 3.2%. 

Cash: 

Q1 operating cash flow was $22.1 billion or 61.6% of revenue compared to $19 billion or 74.5% of revenue in the same period last year. 

Q1 free cash flow was $11 billion or 30.7% of revenue compared to $9.0 billion or 35.1% of revenue in the same period last year.  

The company had cash & marketable securities of $105.5 billion and debt of $31.7 billion. 

Valuation: 

TSMC trades at a forward P/S ratio of 11.5. The company has traded at a minimum forward P/S ratio of 5.9 and a maximum of 13.4 in recent years. TSMC is currently trading slightly higher than mid-range. On the bottom line, it trades at a forward P/E ratio of 23.2. TSMC has traded at a minimum of 13.5 and a maximum of 29.6. TSMC is currently trading around the mid-range on the bottom line. 

Notable Risks: 

Geopolitical concerns.

Memory Stocks: 

Micron: Doors. Blown. Off. 

As someone who looks at hundreds of earnings reports a year, there are times an earnings report shatters expectations like an Olympian breaking a record or an athlete leaving no doubt who is the best in the game. Micron did this by dropping an earnings report so strong on fundamentals that I cannot recollect seeing one quite like this.  

Micron blew the doors off with revenue growth of 196.3% YoY and up 75% QoQ for a beat of 22.3% on a massive revenue base of about $24 billion a quarter. The forward fiscal Q3 growth is eye-watering at 260.2% YoY and 40.4% QoQ. This is nearly 100 points higher than what analysts had slated for fiscal Q3 with consensus at 150.2% growth YoY.  

The $10.2 billion sequential increase is nearly unprecedented outside of Nvidia’s most recent quarter posting $11 billion QoQ growth – yet, let’s not forget that Nvidia is the world’s most valuable company.   

Here is what management stated about this record-breaking quarter:   

“Quarterly revenue nearly tripled versus one year ago, and revenue for DRAM, NAND, HBM (high-bandwidth memory) and each business unit reached new highs. Our fiscal Q3 single quarter revenue guidance exceeds the full year revenue for every year in our company’s history through fiscal 2024. For fiscal Q3, we anticipate exceptional records across revenue, gross margin, EPS and free cash flow.”  

To also help illustrate just how impressive this earnings report was, consider that Micron was not supposed to see $33 billion in a single quarter until FQ1 2028 (November of 2027) yet following a second quarter of $10B sequential growth, will now see this revenue in the quarter ending in May of 2026.  

As incredible as the revenue growth is; the margins are arguably even more incredible at an 81% gross margin and 76% operating margin guide for the next quarter.   

So, why would the market sell the report after hours? Well, to be prudent, Micron was reporting a steep, negative gross margin of (9%) in FY2023 with one quarter as low as (32.7%) in FY23. Thus, the question of whether we are seeing a cyclical top or a structural shift in memory is a very valuable question to answer, and the importance of this broader question is only further reinforced by the results we saw this past quarter. 

The difference between the AI cycle and the cyclical peaks in the past is that Micron is combining record fundamentals with improving visibility through multi-year customer agreements and a strengthening product roadmap (HBM4/HBM4E, 1γ DRAM, Gen6 SSDs).  

Management repeatedly stated the market is supply constrained beyond 2026, with new fabs coming online in 2028. Meanwhile, longer context windows, reasoning, and agentic workloads will keep HBM and DRAM demand elevated. NAND is proving it's no longer an afterthought with historic pricing surges that are causing heavyweight customers to seek more stability with SCA agreements.   

While I do believe SCAs could be the reason for softer price action, it’s my conclusion at this time that memory remains an important strategic asset that results in Micron being in the driver’s seat during a sustained upward trend, albeit with the occasional lumpiness inherent to supply chains. In that sense, I foresee Micron becoming more secular than the market has historically treated it.  

Revenue: 

Micron’s Q2 FY2026 ending February revenue grew by an impressive 196.3% YoY and 74.9% QoQ to a record $23.9 billion, beating estimates by a solid 22.3%. Revenue growth accelerated by nearly 140 percentage points from 56.7% YoY and 20.6% QoQ growth in the previous quarter. The $10.2 billion sequential increase was the largest in the company’s history and was primarily driven by strong AI memory demand.  

Management also provided a strong FQ3 revenue guidance of $33.5 billion, implying a YoY growth of 260.2% and 40.4% QoQ. The revenue guidance beat consensus estimates by a stellar 44%. 

AI Revenue: 

Combined CMBU and CDBU FQ2 revenue QoQ growth was 75% and calculating using the similar mix in FQ2 implies 40% QoQ guide in FQ3.  

Micron’s Cloud Memory Business Unit (CMBU) FQ2 revenue grew by 163% YoY and 47% QoQ to a record $7.75 billion. Revenue growth accelerated by 63 percentage points from 100% YoY growth and 16% QoQ growth in the previous quarter. The strong sequential growth was primarily driven by an increase in prices and favorable mix.  

Core Data Center Business Unit (CDBU) FQ2 revenue grew by 211% YoY and 139% QoQ to a record $5.69 billion. Revenue growth accelerated sharply from 4% YoY and 51% QoQ growth in the previous quarter. The strong sequential growth was primarily driven by higher pricing and growth in bit shipments. 

Earnings: 

Micron’s FQ2 GAAP EPS grew by 756% YoY to $12.07, beating estimates by 36.3%. Adjusted EPS grew by 682.1% YoY to $12.20, beating estimates by 36%, primarily driven by higher memory prices, cost controls, favorable revenue mix, and operating leverage.  

Management also provided a strong guide for the next quarter. GAAP EPS guide is $18.90, implying a YoY growth of 1025%. While the adjusted EPS guide is $19.15, implying a YoY growth of 902.6% YoY, beating estimates by 77.8 

Margins: 

Micron’s margins are gravity-defying. 

FQ2 gross profits grew by 499.2% YoY to $17.76 billion. Gross profit margin was 74.4%, an improvement of 37.6 percentage points YoY and up 18.4 percentage points sequentially. It beat the management guidance of 67%. The adjusted gross margin improved by 37 percentage points YoY and 18.1 percentage points sequentially to 74.9%. The strong gross margin was driven primarily by higher pricing, favorable mix, and cost controls.  

Management has guided further improvement of gross margin to 81% in the next quarter.  

FQ2 operating profits grew by 810% YoY to $16.14 billion. Operating margin came at 67.6%, an improvement of 45.6 percentage points YoY and 22.6 percentage points sequentially. It beat the management guidance of 58.7%.  

The adjusted operating margin improved by 44.1 percentage points YoY and 22 percentage points sequentially to 69% driven by operating leverage. Management has guided further improvement of operating margin to 76.2% and adjusted operating margin to 76.8% in the next quarter.  

FQ2 net income was $13.79 billion or 57.8% of revenue compared to $1.58 billion or 19.7% of revenue in the same period last year. Adjusted net income was $14.02 billion or 58.8% of revenue compared to $1.78 billion or 22.1% of revenue in the same period last year. 

Cash: 

Micron’s strong profits are leading to higher cash flows.  

FQ2 operating cash flows grew by 202% YoY to $11.9 billion with an operating cash flow margin of 49.9% compared to 49% in the same period last year.  

FQ2 adjusted free cash flows grew by 705% YoY to $6.9 billion with an adjusted free cash flow margin of 28.9% compared to 10.6% in the same period last year.  

Capex grew by 61.3% YoY to $5.0 billion. For FQ3, management has guided a capex of $7.0 billion and expects adjusted free cash flows to roughly double sequentially.  

Cash and investments were $16.6 billion and debt of $10.14 billion compared to $12.02 billion and $11.76 billion in the previous quarter. Micron repurchased shares worth $350 million and also reduced debt by $1.6 billion in the recent quarter. 

Valuation: 

Micron trades at a forward P/S ratio of 4.3. The company has traded at a minimum forward P/S ratio of 1.2 and a maximum of 6.8 in recent years. Micron is currently trading slightly above the mid-range. On the bottom-line, the company is trading at a reasonable forward P/E ratio of 7.2 due to the strong margin expansion and expected adjusted EPS growth of 597% to $57.8 for FY2026. 

Notable Risks: 

Memory can be stubbornly cyclical, and the market appears to be discounting the familiar pattern Micron has faced in past cycles where peak shipments were followed by a sharp reversal ahead of pricing and demand normalizing.  

Management commentary supports Micron being in a sustained uptrend as 2026 is supply-constrained and greatly limited by DRAM and NAND supply. However, despite the outsized demand and strong product road map, Micron will likely see peak sales before Nvidia’s Vera Rubin sees peak sales given HBM and data center DRAM sits earlier in the supply chain. Therefore, there can be air pockets tied to Nvidia’s GPUs shipping in volume even when the overall trend remains intact.   

Micron is also exposed to PC/consumer and traditional server revenue.   

SanDisk: A Thing in Motion … 

SanDisk’s second quarter report was a blowout on all accords, with the company reporting an impressive 31% QoQ growth for revenue to $3.03 billion and a tremendous 408% QoQ growth to $6.20 in adjusted EPS, capitalizing on strong demand and strong pricing from undersupply dynamics.   

However, the guide was even more impressive, with SanDisk forecasting $4.4 to $4.8 billion in revenue, up 52% QoQ at midpoint, and adjusted EPS more than doubling QoQ to $12 to $14, roughly 200% above consensus at midpoint.   

To put in perspective just how large of a beat this was, SanDisk was not expected to see this level of revenue or EPS at the end of 2027 – consensus for the Dec 2027 quarter was $4.19 billion and $9.29 in EPS heading into this report.  

There were a handful of important comments from management in the call regarding the NAND market, that it will be even more undersupplied in fiscal Q3, while data center growth forecasts raised yet again.   

SanDisk noted that it was unable to fulfill customer demand in Q2, yet management added that it anticipates “the market to be more undersupplied [in Q3] than it was in the second quarter” with bit growth down mid-single digits QoQ compared to a mid-single digit increase QoQ in Q2.   

Management also added that they expect “customer demand well above supply beyond calendar year 2026, which requires careful allocation planning and alignment with our customers.”   

This is rather important as analysts were currently expecting NAND pricing to peak in the calendar Q1 quarter on a QoQ basis, yet ASP growth may end up higher for longer considering the supply-demand imbalance is widening.   

For example, analysts were projecting NAND ASPs to accelerate to the low-20s to low-30s QoQ in calendar Q1 and then slow to the mid-teens in calendar Q2, yet a widening imbalance could potentially push prices up to 40% QoQ and 20% QoQ, respectively.   

This accelerating forecast for data center exabyte growth ties into NAND’s increasing role in AI infrastructure, as we had recently outlined with KV cache requirements and Nvidia’s new inference memory platform. On this exact topic of Nvidia’s KV cache discussion and TB of content per GPU, management explained that “none of that demand is in the numbers we're talking about, demand numbers at this point,” but “our initial looks at it when we look at, let's say, '27 demand, we think that's roughly maybe 75 to 100 additional exabytes. And then a year after that, you can double that. So it is a significant amount of demand.”   

For context, 75-100 EB of demand in 2026 would account for roughly 6-8% of the entire flash market, while doubling that to 150-200EB in 2027 would correspond to 10-13% of the market – a significant new demand driver.  

As a result of the increasing role of NAND and enterprise SSDs in AI inference applications, and expectations for a “meaningful increase in NAND content per deployment,” management expects data center revenue to “grow meaningfully in both the near and long term.”   

However, there is another near and medium-term risk to the SSD story related to the KV cache-optimized tier Nvidia recently proposed with its ICMS platform – Google’s TurboQuant.  

As we had discussed in our first SanDisk analysis, SanDisk: Shares Up 559% In 2025 On NAND Flash, Enterprise SSD Tailwinds, the KV cache essentially serves as a model’s long-term memory that is reused and extended throughout many steps or requests. KV cache capacity is a known pain point when working to balance long-context reasoning and memory capacity in inference workloads, as it can consume ~30% of GPU memory during deployment. While TurboQuant directly addresses this pain point by compressing the vectors to reduce KV cache memory size by up to 6X, it likely will not fundamentally alter the architectural needs for NAND and SSDs for storing, caching and retrieving training and inference data. Instead, it will likely help drive KV cache usage lower and enable longer context windows, more concurrent requests and open the door for previously-infeasible memory-constrained workloads to arise.  

Revenue: 

SanDisk reported $3.03 billion in revenue in Q2, beating estimates by ~12.5%, with SanDisk attributing the growth to higher prices across its three segments with prices strengthening through the quarter.  

Revenue growth accelerated more than 38 points to 61.2% YoY, while sequential growth accelerated nearly 10 points from 21.4% QoQ in Q1 to 31.1% QoQ in Q2.   

Q3’s guide was a blowout versus consensus, with SanDisk forecasting $4.4 to $4.8 billion in revenue, more than 58% ahead of consensus for just $2.91 billion. This also points to a significant 110 point acceleration to 171.3% YoY at midpoint and 21 points to 52% QoQ.  

Estimates for fiscal Q4 points to 219.6% YoY growth to $6.1 billion. For the full year, current consensus points to 116.1% YoY growth to $15.89 billion. 

AI Revenue: 

SanDisk’s data center revenue growth was robust in Q2 with the company reporting growth of 76% YoY and 64% QoQ to $440 million, accelerating 86 and 38 points respectively. Data center still accounts for a smaller portion of overall revenue at almost 15% in the quarter, though this is up from 12% last quarter.  

Management said they are seeing strong adoption of data center products from cloud hyperscalers, enterprise and edge data centers, and system integrators. SanDisk completed qualification of its PCIe Gen5 high-performance TLC SSDs at a second hyperscaler in the quarter, while two major hyperscalers are advancing with qualifications for its BiCS8 QLC ‘Stargate’ products, set to begin shipping in the next several quarters, providing another tailwind for growth.    

Earnings: 

SanDisk reported a large beat on EPS in Q2, though arguably the Q3 guide could be one of the largest beats in tech, with management forecast Q3 adjusted EPS 200% above consensus estimates.  

GAAP EPS was $5.15 in Q2, up 587% QoQ and 615% YoY, and nearly $2 ahead of consensus estimates for $3.20. Adjusted EPS was $6.20, beating the $3.78 estimate by 64% and representing 408% QoQ and 404% YoY growth. 

For Q3, management guided for $12 to $14 in adjusted EPS, up 110% QoQ, and coming in 200% above consensus estimates for $4.33 at midpoint.  

Margins: 

SanDisk saw strong gross margin expansion in Q2 stemming from higher prices, while unit cost reductions served as an operating margin tailwind.   

Q2 GAAP gross margin was 50.9%, up 21.1 points QoQ and 18.6 points YoY, while adjusted gross margin was very similar at 51.1%, up 21.2 points QoQ and 18.6 points YoY.  

GAAP operating margin was 35.2%, up 27.6 points QoQ and 24.8 points YoY, while adjusted operating margin was 37.5%, up 26.9 points QoQ and 25.1 points YoY.   

GAAP net margin was 26.5%, up 21.6 points QoQ and 21 points YoY, and adjusted net margin was 32%, up 24.2 points QoQ and 20.5 points YoY.  

For Q3, SanDisk projected margins to expand further, guiding GAAP gross margin to be 64.9% to 66.9%, up 15 points QoQ and 43.4 points YoY at midpoint, while GAAP operating margin was implied to be 54.7% at midpoint, up 19.5 points QoQ.  

Adjusted gross margin was guided to be 65% to 67%, with adjusted operating margin guided to be 56% at midpoint.   

Cash: 

Operating cash flow of $1.02 billion, up ~973% YoY, for a 33.7% margin, up 12.6 points QoQ and 28.6 points YoY; FCF of $980 million and adj FCF of $843 million for a 27.9% margin, up 8.5 points QoQ and 23 points YoY; Cash of $1.539 billion and debt of $603 million 

Valuation: 

SanDisk’s valuation is somewhat hard to pin down given the company’s limited history on the public markets after its February 2025 spinoff, and its 2,720% rally in the past one year. On the top line, SanDisk is trading at 8.4 forward P/S ratio, having traded as low as 0.6 last August and with an average multiple of 2.1 for its limited public history.   

On the bottom line, SanDisk is trading at a 21.5 forward P/E ratio, having traded as low as 1.0 last August with an average of around 9.7. 

Notable Risks: 

For a stock with this level of top-line and bottom-line growth, the risk is deceleration. While adjusted EPS is expected to surge 1307% in FY2026 and 142.6% in FY2027, expectations call for a 12.8% decline in FY2028. 

Quick Note on HDD Stocks: 

We’ve covered HDD stocks recently on our Discovery tier.  

Hard disk drives (HDDs) offer the lowest-cost per terabyte, which makes HDDs ideal for “big data” storage, backups and large AI datasets. Compare this to solid state drives (SSDs) which store data on flash memory chips and are far faster and lower-latency. SSDs cost more per terabyte, thus leveraging a mix of HDDs and SDDs is a popular choice.   

As it stands today, SDDs have illustrated significant pricing power compared to HDDs. Therefore, because HDDs are considered more commoditized in the current market dynamics compared to SDDs, these stocks are being overlooked. 

As inference requires more exabytes to be stored, HDDs offer an advantage in that storage tier. In fact, a leading HDD management team sees a CAGR of 25%+ over the next 5 years with HDD representing “80% of the storage media that deployed within a hyperscale environment.” Multimodal datasets are among the largest drivers of incremental storage demand. Video is another data hog and even modest growth here from multimodal AI can create what’s called “data exhaust.”   

There are two things to note when looking at HDD stocks. The first is these are bottom-line stories in the current market with growth of 67%+ and even 270%+ last quarter on EPS. The second is that while pricing is fixed for 2026, it comes up for re-negotiation in 2027.  

Subscribe to Discovery to unlock the full Top 15 AI Stocks report and get additional insights into the next phase of the AI trade. To subscribe to Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40Discovery with 40% off, click here to email usclick here to email us or email premium@io-fund.com and mention code DISCOVERY40

AI Networking Stocks 

We covered quite a bit of the shifts in AI networking in our last quarterly report, found here. 

Lumentum: Capacity Constrained (and Loving It) 

Lumentum is a leading allocation as of January and is up 127% YTD. The company (and cleary the stock too) is benefiting from outsized demand for its EML lasers, reaching a quarterly company record in EML laser shipments. While EMLs are largely spoken for with InP wafer fab capacity fully allocated with long-term agreements, the company is expanding its capacity with additional supply expected to come online in the second half of this calendar year.   

The transition to 1.6T is moving faster than management originally anticipated, which contributed to the beat/raise with management stating: “We achieved another quarterly company record in EML laser shipments led by 100 gig line speeds and bolstered by a ramp in 200 gig devices. Simultaneously, we expanded our footprint in next-generation architectures shipping CW lasers for 800 gig manufacturers and increased volumes of ultra-high-power laser shipments for CPO applications.”  

There are additional growth levers for Lumentum as we look further out, driven by optical circuit switches and co-packaged optics. Optical circuit switches are beginning to move the needle now with a $400 million backlog, although currently at around $10 million in revenue. In 2027, co-packaged optics (CPOs) will represent another important market for Lumentum alongside UHP chips and ELS modules will help expand the company’s serviceable addressable market.   

Management emphasized that indium phosphide wafer fab capacity is fully allocated, with the company indicating they have already delivered half of their expansion target over one quarter alone due to strong customer demand necessitating they pull forward delivery. Thus, the natural question for an investor is whether Lumentum can add more capacity. The company stated they foresee more capacity coming in the second half of the calendar year:  

“We are scaling rapidly through precision tool optimization and yield gains. This execution will help to ensure that additional capacity comes online as planned over the next two quarters and beyond. While not able to size it, we now have line of sight to a significant block of additional capacity starting in the second half of 2026 both recurrent activities in Sagamihara and better utilization of our Caswell, United Kingdom and Takao, Japan fabs.” 

Although minimal right now, 200G is ramping faster than expected, representing 5% of unit volume yet represents 10% of laser chip revenue. According to management, demand for 200G EMLs is about a quarter faster than they originally anticipated with the goal of ending the year with 25% of unit volume from this new product mix – with these seeing higher average sales prices than the 100-gig.  

Management stated the following: “Our 200-gig line speed, as we said, is actually doing a little bit better than we expected. I think on the last call, we had said that the 5% revenue of — 5% of mix would be this quarter. It was a quarter earlier than we had expected, and that's primarily because 1.6T is coming on, I think, faster than we initially anticipated, and that is heavily being driven by 200-gig EMLs.”  

This was discussed further in the call with management stating 1.6T was stronger than it was 90 days ago. 

Revenue: 

Lumentum delivered Q2 revenue at the upper end of its guided range, yet its guidance stands out as it not only points to YoY growth accelerating almost 24 points to 89.3% YoY, but it also was a larger magnitude beat in dollar terms versus last quarter.   

Q2 revenue was $665.5 million, a modest 2% beat to estimates and in the upper end of Lumentum’s guidance for $630-$670 million. Revenue growth accelerated 7.1 points to 65.5% YoY, while sequential growth was robust at 24.7% QoQ, its fastest growth in eight years and accelerating 13.7 points.   

For Q3, Lumentum guided for revenue between $780 million and $830 million, accelerating 23.8 points to 89.3% YoY at midpoint. Sequential growth will remain strong with guidance pointing to growth of 21% QoQ at midpoint.  

What’s impressive here is that Lumentum’s guidance beat consensus by a larger margin than it did last quarter – at the $805 million midpoint, this would be nearly $99 million ahead of the $706.4 million estimate, whereas Q2’s guide for $650 million at midpoint beat by ~$88 million.   

Following the report, we had stated: “Considering the scope of this raise for Q3, it’s likely that estimates for Q4, which currently are pegged at just $770.4 million, are revised much higher in the coming days/weeks. As a result, it’s likely that consensus estimates for FY26, currently at $2.64 billion, move ~8-10% higher.” 

At time of writing in April, FQ4 estimates were revised up 19.1% since the earnings report for 90.9% YoY growth to $917.8 million. FY2026 revenue estimates are $2.92 billion, up 77.8% YoY and have been revised up by 10.6%. 

AI Revenue: 

Components revenue was $443.7 million in Q2, up 68.3% YoY and 17% QoQ. YoY growth accelerated 3.4 points while QoQ growth decelerated 1.4 points from last quarter. Components accounted for 66.7% of revenue in Q2, down from 71% in Q2. 

Driving this growth were EML shipments, with Lumentum saying both 100G and 200G EMLs reached new company records. Systems revenue rose 43.5% QoQ and 60.1% YoY to $221.8 million, a sharp acceleration from a (3.6%) QoQ decline and 46.5% YoY increase in Q1. This was driven by record cloud transceiver shipments. 

EPS: 

Q2 GAAP EPS was $0.89, up from just $0.05 in Q1 and beating the $0.50 consensus estimate by 78%. Adjusted EPS was $1.67, up 51.8% QoQ and 297.6% YoY, and beating the $1.40 estimate by 18.4%. 

For Q3, Lumentum guided for $2.15 to $2.35 in adjusted EPS, pointing to YoY growth of 294.7% and QoQ growth of 34.7%. At midpoint, this represented a 40.6% beat to the consensus estimate of $1.60. 

Margins: 

Lumentum reported solid expansion in gross margins in Q2, with GAAP gross margin up 2.1 points QoQ and 11.3 points YoY to 36.1%, and adjusted gross margin up 3.1 points QoQ and 10.2 points YoY to 42.5%.  

GAAP operating margin in Q2 was 9.7%, up 8.4 points QoQ and 22.5 points YoY, while adjusted operating margin was 25.2%, up 6.5 points QoQ and 17.3 points YoY (and ahead of guidance for 20-22%).  

Lumentum forecast this operating margin to continue at a similar rate, projecting adjusted operating margin of 30-31% in Q3, up 5.3 points QoQ and 19.7 points YoY. 

GAAP net margin expanded 11 points QoQ and 26.9 points YoY to 11.8%. Adjusted net margin expanded 5.4 points QoQ and 14.1 points YoY to 21.6%. 

Cash: 

Operating cash flow of $126.7M for a 19% margin in Q2, up from 6% in the year ago quarter and 10.9% in Q1 

FCF of $43.1M for a 6.5% margin, up from (4%) in the year ago quarter and (3.4%) in Q1 

Cash and equivalents increased slightly to $1.16 billion while debt was $3.29 billion at the end of Q1. The company recently entered into privately negotiated exchange agreements with certain holders of its 0.50% Convertible Senior Notes due 2026 and 1.50% Convertible Senior Notes due 2029 totaling $474.6 million. The company will issue about 6.3 million of shares in exchange for these notes and will not receive any cash proceeds. 

Valuation: 

Lumentum trades at a forward P/S ratio of 21. The company traded at a minimum forward P/S ratio of 1.7 and a maximum of 21.9 this month. On the bottom-line, it trades at a forward P/E ratio of 110.5. Lumentum traded at a minimum forward P/E ratio of 11.6 and a maximum of 115.3 this month.  

Notable Risks: 

Lumentum’s valuation has become more demanding following the stock’s strong move higher, which creates the risk of multiple compression if growth falls short of elevated expectations. The company also carries a relatively high debt load, which adds financial risk and reduces flexibility compared to a cleaner balance sheet.  

AAOI: Buckle Up 

AAOI grew to become our top position as that’s what a quick 350% return does to a portfolio. We might need to trim back for a more intentional allocation but that does not deter from the carefully placed thesis we put into place months ago. 

As you’ll recall last quarter, AOI (Nasdaq: AAOI) missed earnings due to orders getting pushed out to Q4. Therefore, it was quite important that AOI meet expectations following the delay. The company’s revenue grew 34% YoY and 13% QoQ and guided to grow 58% YoY and 17% QoQ for Q1. Of this, data center inflected with growth of 69% YoY and 70% QoQ. Suffice to say, AOI met the bar set for the company with momentum headed into 2026.  

During the call, management focused on detailing the ramp for 800G and 1.6T with targets shared through mid-2027. The forecast implies that AOI’s optical attach per unit of compute is rising as network sizes increase to include more lanes and more links. For example, the 800G era is widely expected to require record ports, resulting in higher revenue for optical networking companies. The industry is shifting from 400G to 800G with 1.6T on the roadmap as throughput becomes more critical with the incoming inference phase.   

Management offered a forecast for mid-2027 of $378 million per month with the framework of 800G being the bulk of the revenue, 1.6T contributing and some 100G/400G content contributing yet the lowest of the mix. Importantly, management framed this discussion as being capacity-constrained rather than demand-constrained. In the more near-term, management guided for $1 billion in 2026 revenue with $120 million in adjusted operating profit.  

The following was stated in the opening remarks:  

“Given the recent surge in customer inquiries and apparent rising demand, we believe that by mid-2027, 100G and 400G revenue will be approximately $90 million. 800G revenue will be approximately $217 million and 1.6 terabit revenue will be approximately $71 million monthly. Altogether, this represents $378 million in monthly revenue for transceiver products.”  

The company also stated they expect $1 billion in revenue this year compared to analyst estimates that are just shy of $764 million: “Looking more broadly at 2026. While it's still early in the year, we expect to generate over $1 billion in revenue this year, with a non-GAAP operating profit of over $120 million. This revenue level is limited by our production capacity and supply chain, not market demand, which we believe is much larger.” 

Analyst estimates greatly missed the mark at $521M per quarter in September of 2027, assuming the $378 million per month plays out ($1.14B per quarter). 

Revenue:  

Applied Optoelectronics (AOI) Q4 revenue grew by 33.9% YoY and 13.2% QoQ to $134.3 million, beating estimates by 4.7%. The revenue growth was primarily driven by strong data center revenue which grew by 69.2% YoY and 70.4% QoQ to $74.9 million.  

During the quarter, the company also announced that they received the fourth 800G volume order from one of their major hyperscale customer to support its AI data center growth, which is likely to be Amazon. AOI has begun ramping up production of this 800G module in anticipation of a strong volume ramp starting in Q2. Management also mentioned in the earnings call that they are in discussion with a new hyperscale customer about qualifying for 800G and 1.6T products and sounded confident about the growth trajectory in both these products with multiple customers.  

Management also provided a strong Q1 revenue guide of $150 million to $165 million, implying a YoY growth of 57.7% and 17.3% QoQ at the midpoint. The strong Q1 revenue growth is led by sequential revenue growth in both CATV and data center revenue.   

The company’s 2025 revenue grew by a solid 82.8% YoY to $455.7 million. Management expects strong revenue growth to continue in the coming years and guided 2026 revenue of over $1 billion, implying a 119% YoY growth, beating estimates by 31%. 

AI Revenue: 

The company’s Q4 data center revenue grew by 69.2% YoY and 70.4% QoQ to $74.9 million. The revenue growth sharply accelerated from 7.3% YoY and decline of (1.9%) QoQ in Q3. Revenue of 100G products grew by 54% YoY and 400G products grew by 141% YoY. 100G products accounted for 51% of data center revenue, 200G and 400G transceiver products accounted for 41%, and 8% was from 10G and 40G transceiver products. 

EPS: 

The company’s Q4 GAAP EPS came at ($0.03), beating estimates by $0.12. While the adjusted EPS came at ($0.01), beating estimates by 91%. 

Margins: 

The company witnessed a turnaround in margins and expects to be sustainable profitable on an adjusted basis from Q2 driven by the shift to higher margin revenue, operational efficiencies, and leverage.  

The company’s Q4 gross profits grew by 46% YoY to $41.95 million. Gross profit margin improved by 250 basis points YoY and 320 basis points QoQ to 31.2%. Adjusted gross margins improved by 250 basis points YoY and 40 basis points sequentially to 31.4%, beating the guidance of 30%. The improvement in gross margins was primarily due to the favorable product mix and cost reduction efforts.  

Q4 operating margin was (8.6%) compared to (6.5%) in the same period last year and (15.3%) in the previous quarter. Adjusted operating margin was (5.3%) compared to (2.5%) in the same period last year and (8.7%) in the previous quarter. 

Cash: 

The company’s cash flows have been weak. However, with improved profitability expected in the coming quarters we could expect cash flows to improve.  

Q4 operating cash outflow was ($29.6 million) or (22%) of revenue compared to (24.6%) in the same period last year.  

Q4 free cash outflow was ($113.6 million) or (84.6%) of revenue compared to ($53.1 million) or (53%) of revenue in the same period last year. To support the strong expected growth capex grew by 227% YoY to $84 million in Q4.  

Cash and short-term investments were $216 million and debt of $197.2 million at the end of Q4 2025. The company also announced an equity offering of $250 million after the announcement of Q4 results. 

Valuation: 

The company is trading at peak multiples on the top line and the bottom line. The company is trading at a forward P/S ratio of 12.4 and traded at a minimum level of 0.2 in May 2023. On the bottom line, it trades at a forward P/E ratio of 188 and we have limited data here since the company will be profitable on an adjusted basis from Q2 this year. 

Notable Risks: 

Valuation remains a key risk, as a higher multiple leaves the stock with less room for error if growth or guidance falls short of expectations. The company is also generating negative cash flow, which adds financial risk and can weigh on investor sentiment if profitability takes longer to materialize. Another important risk is that networking supplier dynamics can shift quickly, particularly in AI infrastructure where customer preferences, architectures, and share positions may change faster than expected.  

Coherent: Slow and Steady 

Coherent is a stock that will test investors as the company has near-perfect positioning, yet the timing is taking longer than what growth investors typically look for. If I had to describe this earnings report, I would use the word “visibility” as the headline numbers will fail to impress, yet I believe the stock price will march upward as the equation of what Coherent offers + where the demand is = will eventually materialize (in 2026).   

The data center and communications segment revenue grew by 33% YoY and 11% QoQ in FQ2, accelerating from 26% YoY growth and 7% QoQ growth in FQ1 driven by strong AI demand. The Communications segment grew 44% YoY and 9% QoQ, although this was down from 11% QoQ growth and 55% YoY reported last quarter. However, the data center segment accelerated meaningfully to 14% QoQ and 36% YoY, up from 4% QoQ growth and 23% YoY last quarter. As of this quarter, data center and communications segment represents 70% of revenue.  

The company offered strong visibility metrics, such as stating book-to-bill ratio is 4X, meaning they are booking orders 4X faster than they can ship. Much of Coherent’s timing hinges on indium phosphide capacity as the company has been working to increase this capacity by moving from 3-inch wafers to 6-inch wafers, which will produce 4X the amount of chips at half the cost. The words “second half" came up frequently with management emphasizing an incoming inflection: “We expect 1.6T to ramp significantly over the coming quarters, with the early phase of the ramp driven by our EML and silicon photonics-based transceivers, followed by our 200G VCSEL-based 1.6T transceivers ramping in the second half of this calendar year.”  

In addition to the transition toward 1.6T being a catalyst, optical circuit switches (OCS) and co-packaged optics (CPO) represent additional catalysts as we move look into 2027. Although in the future, an area where Coherent could stand out is CW lasers for the incoming CPO wave in AI networking. According to management, they secured a large order from a hyperscaler. Management also emphasized their non-mechanical liquid crystal technology for OCS provides an edge, with an update on the call they currently have 10 customers in their pipeline. 

Revenue: 

Coherent’s FQ2 ending December 2025 revenue grew by 17.5% YoY and 6.6% QoQ to $1.69 billion, beating estimates by 2.7%. On a pro forma basis, excluding revenue from the divested Aerospace and Defense business, which the company sold in FQ1, revenue grew by 9% QoQ and 22% YoY primarily driven by AI Datacenter & Communications demand. 

Management guided FQ3 revenue of $1.70 billion to $1.84 billion, implying a YoY growth of 18.2% and 5% QoQ at the midpoint, beating estimates by 3.5%. As per our internal proforma estimate, it implies a YoY growth of 23.8% and 6.3% QoQ in FQ3 after excluding Aerospace and Defense business revenue from the prior year quarter and also the recently sold product division based in Munich. The product business in Munich had averaged $25 million quarterly revenue and had a gross margin well below the company’s corporate gross margin. 

Management expects continued strong growth in the second half of fiscal year 2026 and throughout fiscal year 2027 based on strong datacenter and communications demand and the continued production capacity expansion along with improving demand in the Industrial segment. 

AI Revenue: 

FQ2 data Center segment revenue grew by 36% YoY and 14% QoQ, accelerating from 23% YoY and 4% QoQ growth reported in FQ1. The FQ2 data center revenue growth was driven by growth in both 800 gig and 1.6T transceivers. The company is witnessing very strong AI demand and is also rapidly expanding capacity, and management expects double-digit sequential growth in data center segment in both FQ3 and FQ4.   

Management expects revenue growth in the current quarter to be driven by a combination of growth in both 1.6T and 800 gig transceivers as well as growth in the OCS systems. Coherent is witnessing strong demand for the 1.6T transceivers across multiple customers and continue to expect both 800 gig and 1.6T to grow significantly in calendar 2026.  

Coherent expects OCS revenue to grow sequentially in the coming quarters as they ramp production capacity as fast as possible to meet the rapidly growing demand. Management estimates over $2 billion of addressable OCS market in the coming years. 

EPS: 

FQ2 GAAP EPS grew by 72.7% YoY to $0.76, beating estimates by 10.1%. Adjusted EPS grew by 35.8% YoY to $1.29, beating estimates by 7%. 

Management has guided adjusted EPS of $1.28 to $1.48 for FQ3, implying a YoY growth of 51.6% at the midpoint and beating estimates by 4.5%. 

Margins: 

The company’s margins are improving driven by reductions in product costs, manufacturing efficiency gains, and operating leverage.  

FQ2 gross profits grew by 22.3% YoY to $622.8 million. Adjusted gross profits grew by 20% YoY to $657.4 million with an adjusted gross margin of 39%, up 80 basis points YoY and 30 basis points sequentially and was in-line with the guide. The improvement in gross margin was driven by reductions in product input costs, efficiency gains from improved cycle times in the manufacturing process, as well as yield improvements. Pricing optimization also continued to contribute meaningfully to the gross margin expansion. The management FQ3 guide is 39.5%.  

FQ2 operating income grew by 34.3% YoY to $184 million. Adjusted operating income grew by 26.8% YoY to $336 million with an adjusted operating margin of 19.9%, up 140 basis points YoY and up 40 basis points QoQ and was in-line with the guide. The operating margin improvement was due to operating leverage and operational efficiencies. The management FQ3 guide is 20.9%. 

Cash: 

Coherent’s balance sheet is beginning to improve, with the company using proceeds from the divestment to pay down debt, though debt to cash remains upside down. Operating cash flow margins were also thin and free cash outflows increased due to high capex to support the strong AI demand. 

FQ2 operating cash flow was $57.9 million or 3.4% of revenue, down from $187.4 million in the same period last year and up from $46 million in the previous quarter. 

FQ2 free cash outflow was ($95.7 million) or (5.7% of revenue), down from $81.7 million or 5.7% of revenue in the same period last year. FQ2 capex grew by 45.3% YoY to $154 million to support the strong AI demand.  

The company had debt of $3.35 billion and cash of $863.7 million at the end of the December quarter. 

Valuation: 

Coherent trades at a forward P/S ratio of 8.3. The company has traded at a minimum forward P/S ratio of 0.9 in September 2023 and is currently trading at its peak levels. On the bottom line, it trades at a forward P/E ratio of 57.5. Coherent has traded at a minimum of 14.5 in April 2025 and is currently trading at its peak levels on the bottom line too.  

Notable Risks: 

Coherent carries a relatively high debt load, which adds balance-sheet risk and can pressure the stock if growth or margins fall short of expectations. While the company is beginning to improve its financial profile by using divestment proceeds to pay down debt, leverage remains an important watchpoint. Valuation is another risk, as a stronger stock price leaves less room for error. 

Astera Labs: Bouncing off the Lows 

Astera Labs has seen a tremendous comeback this year, and although flat YTD, our buy in February (up 40%) puts us in the positive this year. 

Astera Labs delivered a solid Q4 beat with revenue up another 17.4% QoQ, though the one point to nitpick from this report was Q1’s softer margin guidance, as it would imply a step down to below the 20% GAAP operating margin level sustained for the last three quarters. In addition, the hypergrowth company is not able to keep up with high comps given sequential growth is expected to be 7.7% QoQ following many quarters of double-digit QoQ growth with some quarters as high as 20%+ sequentially.   

There were clues in the call as to when Astera is most likely to see a second wind with Scorpio-X as the catalyst. Overall, Astera has a longer runway than the market is communicating given there is an element of vendor lock-in to their products. Additionally, Ethernet is optimized for reach, whereas Astera specializes in PCIe, which is optimized for something quite different – GPU-to-GPU communication and memory-level workloads inside the rack.  

Astera also announced that it entered into a warrant agreement with Amazon, allowing the tech giant to purchase up to 3.26 million shares at $142.82 through February 2033. The warrants will vest in tranches of payments made by Amazon for the purchase of up to $6.5 billion worth of Astera’s smart fabric switch, signal conditioning and optical engine products. The vote of confidence from one of Astera’s major customers is certainly welcomed. 

Revenue: 

Astera reported Q4 revenue of $270.6 million, topping estimates for $249.6 million by 8.4%. Growth continued to decelerate on both a YoY and QoQ basis, with YoY growth decelerating more than 12 points to 91.8% and QoQ growth by 2.7 points to 17.4%.   

For Q1, Astera guided for revenue between $286 to $297 million, more than 12% ahead of estimates for $260.1 million. However, this guidance points to YoY and QoQ growth continuing to decelerate, to 82.9% YoY and 7.7% QoQ. This would represent Astera’s slowest QoQ growth in its public history. As we had covered in detail last quarter, Astera’s higher-ASP Scorpio X-Series product now entered initial production in late January, likely becoming a greater tailwind to growth as its ramp progresses throughout the year.  

While there was no specific guidance for 2026, current estimates for $1.36 billion, up 59.1% YoY, and revised higher from $1.18 billion when the company reported its Q4 earnings in February.   

AI Revenue: 

Scorpio-P contributed 15% of revenue this quarter and it was stated previously that Scorpio-P and Scorpio-X will reach more than 50% of revenue by 2026. The X-Series is highly anticipated as it’s expected to be a much higher ASP product than the P-Series. Management in the past has called the X-Series an “anchor socket” which means it will secure vendor lock-in for Astera and they will be able to add more products, such as modules and silicon level products. Last quarter, management stated: “we expect our overall dollar content opportunity per AI accelerator to significantly increase, representing another step-up from a baseline revenue standpoint.”  

The update this quarter is that the X-Series will “incrementally grow revenue in the first half of 2026, followed by a transition to high-volume production in the second half. We continue to make excellent progress with additional engagements looking to leverage PCIe for scale-up networking. As previously communicated, we are engaged with 10-plus customers for Scorpio X family. And our current expectation is that we will ship initial quantities of Scorpio X series to support new customer platforms in the second half of 2026 with volume ramp set for 2027.” 

Accounts receivable surged nearly 94% QoQ to $83.2 million, while inventories rose more than 14% QoQ to almost $59 million, both positive signals that revenue growth is likely to remain strong considering the state of demand and hyperscaler capex plans. 

EPS: 

Astera reported its smallest EPS beat since going public, with its $0.58 in adjusted EPS in Q4 beating the $0.51 estimate by just 13.7%; for comparison, its second-smallest beat was in Q2 2024 at 18.9%, while the prior two quarters saw beats of >25% each. Adjusted EPS growth was 56.8%, decelerating from 113% in Q3.  

GAAP EPS was $0.25 in Q4, missing estimates for $0.30, likely due to the sharp net margin contraction related to the income tax provision. GAAP EPS growth was 78.6%.  

For Q1, Astera guided for adjusted EPS to be $0.53 to $0.54 and GAAP EPS to be $0.36 to $0.38, both figures barely ahead of estimates for $0.52 and $0.34 respectively. This would point to adjusted EPS growth accelerating slightly to 62.1%, and GAAP EPS growth accelerating to 105.6%. 

Margins: 

Scorpio-X transitions Astera from selling high-margin silicon with retimers to fabric switches, which could see lower margins in the initial stages until the product scales.  

Cash: 

Operating cash flow was $95.3 million in Q4 for a 35.2% margin, up 7.1 points YoY and 1.3 points QoQ. For 2025, operating cash flow was $319.3 million for a 37.5% margin, expanding 3 points YoY.  

Free cash flow was $76.6 million for a 28.3% margin, up 11.1 points YoY but flat QoQ. For the year, free cash flow was $281.8 million for a 33.1% margin, up 7.3 points YoY.  

Cash and equivalents totaled $1.19 billion while debt remained zero. 

Valuation: 

Astera Labs trades at a forward P/S ratio of 20.9. The company has traded at a minimum forward P/S ratio of 10.5 and a maximum of 60.4 in recent years. Astera Labs is currently trading significantly lower than mid-range. On the bottom line, it trades at a forward P/E ratio of 67.1. The company has traded at a minimum of 29.9 and the highest of 202.2. Astera Labs is currently trading significantly lower than mid-range on the bottom line too.  

Notable Risks: 

Astera Labs may see near-term margin pressure as hardware becomes a larger part of the revenue mix, which can dilute profitability relative to lighter, higher-margin revenue streams from retimers. In addition, the company is likely to maintain elevated operating expenses as it invests aggressively to support growth and expand its position in AI infrastructure with Scorpio and other product lines. As a result, strong top-line growth may not translate as cleanly into bottom-line upside in the near term. 

AI Ethernet Switches and Broadcom Partner 

On our Discovery tier, we recently covered a leading supplier in back-end networking with over 41% market share of the 200G switch market and 55% share of the custom switch market, up from 40% in 2024.

The back-end networking positioning is important for this stock as it means the company is exposed to the faster-growing segment of Ethernet switching – the back-end TAM is forecast to grow at a 56% CAGR through 2029 on scale-out, and potentially soon, scale-up demand, whereas front-end (user-facing) is forecast to grow at a 20% CAGR.   

Per management, the back-end also sees a much faster refresh rate of every 18-24 months versus >5 years for front-end deployments, and  adopts the newest and fastest bandwidths (800G and soon 1.6T) due to the greater performance and reliability requirements of XPU-to-XPU and rack-to-rack communications.   

The company is a lead supplier to Broadcom for 800G switches and 1.6T for Tomahawk6. Both 1.6T switches are optimized for AI back-end networking (scale-out and scale-up), as well as large-scale AI fabrics for AI training and inference for frontier model sizes. Management expects the 1.6T upgrade cycle to emerge in late 2026 but primarily land in 2027, with one customer giving visibility to a back-half 2026 ramp and multiple other ramps occurring through 2027. 

Cooling Technologies 

Vertiv: Facility-Level Cooling is in High Demand 

Nvidia’s future design lineup shows continual increases in power consumption, with Vera Rubin expected to boost thermal design power (TDP) by 50% over Blackwell at up to 180 kW to potentially 230kW per rack, with the Rubin Ultra boosting this to 600kW by late 2027. These advancing power requirements place much more emphasis on liquid cooling, fluid management, and related thermal management technology.  

Goldman Sachs’ Mark Delaney question about cooling product mix evolution and opportunity per MW, noting that “there was some discussion that Rubin raised racks may not need chillers, and conversely post-Supercompute last fall, there was a proposal from a competitor about stainless steel chillers maybe displacing CDUs.”   

Vertiv CEO Giordiano Albertazzi said that this topic is not theirs to confirm, but even if CDUs are reduced, Vertiv stands to benefit as its portfolio spans the entire thermal management chain. He also emphasized that CDUs are likely to persist into the foreseeable future as other cooling tech remains too niche:   

“All in all, we see that design continues to be mixed. If anything, this complicates the thermal chain and this complexity is something that we like as someone who has got the entire portfolio, we certainly are perfectly positioned to support our customers. And again, going back to what we're saying enable the right choice for our customers.  

Cooling chips directly in other ways than through CDU in this moment is not something that we see. Simply because it would — in most of the cases, it will be niche applications probably, but in most of the cases, that would be too dangerous. Blast radius is a little bit too big, et cetera.” 

Regarding CDUs, Vertiv acquired CoolTera in late 2023, a specialist in CDUs and data center liquid cooling. This acquisition expanded Vertiv’s IP, patents and engineering expertise ahead of Vera Rubin, which is primarily liquid cooled. 

Revenue: 

Vertiv delivered a strong Q4 with exceptional strength across key metrics, with backlog more than doubling YoY, orders more than doubling sequentially, and a significant step-up in book-to-bill ratio. Supported by these strong key metrics and ordering patterns, Vertiv guided for revenue growth to accelerate to 32% YoY in FY26, a more than four point YOY acceleration.  

Vertiv reported a solid Q4 with revenue up 22.8% YoY (19% organic) and 7.6% QoQ to $2.88 billion, decelerating from 29% YoY in Q3. This revenue growth was driven entirely by strength in the Americas with revenue up 50.2% YoY, as Europe and APAC both registered YoY declines. 

For Q1, Vertiv guided revenue to be $2.5 billion to $2.7 billion, marking a reacceleration to 27.7% YoY and 22% organic growth at midpoint; however, this will mark a QoQ decline of (9.7%) at the midpoint of this forecast, following typical first quarter seasonality (though slightly better compared to Q1 2025’s (13.2%) QoQ decline). As noted above, growth is expected to accelerate towards the 38% range by Q4, supported by orders growth, backlog and book-to-bill.  

Looking ahead to FY26, Vertiv laid out initial guidance for revenue to be between $13.25 billion to $13.75 billion, accelerating to 32% YoY from FY25’s 27.7% growth; organic growth is projected to be 27-29% YoY, a slight acceleration from 26%. This also marked a significant beat over consensus estimates for $12.39 billion.    

AI Revenue: 

Vertiv reported 109% YoY and 57% QoQ growth to $15 billion in Q4, accelerating sharply from 28% YoY and 12% QoQ in Q3. On a dollar basis, Vertiv added $5.5 billion to its backlog sequentially.  

This backlog growth was out of the ordinary for a few reasons – over the last two years, Vertiv’s fastest QoQ backlog growth up until this point was 15%, yet now growth was ~57%.   

It also created an entirely new dynamic for backlog-to-revenue ratios. Vertiv has seen its backlog to forward revenue (full year guidance from Q4) ratio hover between 72% to 78% over the last three years, yet now this ratio stands at ~111%, suggesting much more elevated revenue visibility through next year with the majority being firm orders.   

Additionally, Vertiv’s conversion time for this backlog has been pushed out, from its typical 9 months to roughly 15 months, with management stating it expects the backlog to be shipped in the next 12 to 18 months. 

Aiding the backlog growth was an increase in organic orders in Q4, with Vertiv reporting organic orders up 252% YoY (though against a flat YoY comp).   

This marked a substantial 192 point acceleration from 60% YoY growth, while QoQ growth accelerated from 20% in Q3. This strong Q4 order intake drove TTM order growth up to 81% YoY, from 21% YoY in Q3. Despite this surge, management emphasized that the pipeline continues to grow across all regions and has not depleted, with the order growth simply reflecting the level of demand in the market with no abnormalities in purchasing.  

Despite this strength, Vertiv's management emphasized that orders are lumpy, and in fact, management plans to drop this metric from future reporting. 

Driven by Q4’s order growth, Vertiv’s book-to-bill ratio jumped to 2.9X, up from 1.4X in Q3. As is the case with orders, book-to-bill has seen some lumpiness quarter to quarter, though there are some key parallels that we can draw here given the simultaneous strength in orders and backlog.   

EPS: 

While adjusted EPS decelerated 26 points in Q4 to 37% YoY, Vertiv forecast a sharp rebound in Q1 to 53%, with FY26’s guide implying that growth will persist at a similar rate through the year.   

Adjusted EPS was $1.36 in Q4, up 37% YoY but decelerating from 63% in Q3, coming in 4.9% ahead of estimates. GAAP EPS growth was exceptionally strong, up 200% YoY to $1.14, though growth was off a smaller base.   

For Q1, Vertiv guided for adjusted EPS to be $0.95 to $1.01, up 53% YoY at midpoint. Estimates point to ~50% growth being maintained in Q2 before a step lower towards the 40-45% range in the second half of the year.   

FY26 adjusted EPS was guided to be $5.97 to $6.07, up 43% YoY and decelerating only slightly from 47% growth in FY25. 

Margins: 
Vertiv saw slight gross and operating margin expansion in Q4, though in line with seasonal trends. Q1 margins are projected to take a step down QoQ but remain higher YoY; however, management added that they expect “to have materially offset unfavorable margin impact from tariffs as of the first quarter of this year,” providing more room for upside beginning in Q2.  

GAAP gross margin was 38.9% in Q4, up 1.1 points QoQ and 1.8 points YoY.  

GAAP operating margin was 20.1% coming in below management’s guidance for 20.7%.  

Adjusted operating margin was 23.2% (versus guidance for 22.4%). Looking ahead to Q1, GAAP operating margin was guided to be 16.3%, down 3.8 points QoQ but up 2 points YoY, while adjusted operating margin was guided to be 19%, down 4.3 points QoQ but up 2.5 points YoY.  

GAAP net margin was 15.5%, flat QoQ and up 9.2 points YoY, as the year-ago quarter recorded a $180 million negative impact related to warrant liabilities. Adjusted net margin was 18.5%, up 0.4 points QoQ and 2.1 points YoY. 

Vertiv guided for solid margin expansion for FY26, suggesting that Q2 through Q4 will see much stronger margins to offset Q1’s softness. GAAP operating margin was guided to be 20.5%, up 2.6 points YoY, while adjusted operating margin was guided to be 22.5%, up 2.1 points QoQ. This will flow through to net margin, with GAAP net margin guided at 15.4%, up 2.4 points, and adjusted net margin guided at 17.5%, up 1.5 points YoY. 

Cash: 

Driven by the surge in orders and larger advanced payments, Vertiv reported robust cash flows in Q4.   

Operating cash flow in Q4 was $1.01 billion for a 34.9% margin, up 15.9 points QoQ and 16.8 points YoY; for the full year, operating cash flow was $2.11 billion (with Q4 accounting for nearly half of that) for a 20.7% margin, up 4.1 point YoY.   

Adjusted free cash flow was $910 million, up 151% YoY, representing a 31.6% margin, up 14.3 points QoQ and 16.2 points YoY. For FY25, adjusted FCF was $1.89 billion for an 18.4%, up 4.2 points YoY.   

Cash and equivalents were $1.83 billion, while debt was $2.91 billion; however, Vertiv’s net leverage ratio remained at 0.5X.   

Inventories increased marginally in Q4, up ~1.8% QoQ to $1.46 billion, while accounts receivable showed a larger jump at 10.6% QoQ to $3.11 billion.   

In accordance with the order surge, deferred revenue jumped more than 60% QoQ to more than $1.81 billion, with management noting that order mix and order type are the two drivers, with mix possibly having a larger influence in Q4.   

Valuation: 

Vertiv trades at a peak forward P/S ratio of 8.4 and traded at a minimum forward P/S ratio of 2.2 in April 2025. On the bottom line, it trades at a forward P/E ratio of 48.7. Vertiv has traded at a minimum of 14.4 and a maximum of 52.5 in recent years.  

Notable Risks: 

As long as hyperscalers continue building capacity, demand for Vertiv’s power and cooling infrastructure should remain supported, particularly since Blackwell systems already require advanced thermal solutions. However, if analysts are modeling a step-function increase in revenue per rack from Rubin’s higher power density and more demanding liquid cooling requirements, that uplift could be pushed out depending on when the Rubin delay resolves. 

Dell: Margin Story; then Revenue 

Dell is not a stock we would own indefinitely, but given the strong recent performance, there’s a chance the stock is in play right now. Above and beyond revenue, Dell’s stock depends on its margins. 

Dell reported some of the strongest AI revenue numbers in the industry in Q4 with AI server revenue up 342% YoY to $9.0 billion, orders up 1,906% YoY to a record $34.1 billion and backlog up 177% QoQ to a record $43 billion.  

Margins: 

The market was growing concerned that rapidly rising memory costs would squeeze on Dell’s margins (“you're supposed to miss numbers, by the way, when memory prices go up”), yet Dell’s margins are among the highest they’ve been since we began tracking the stock. This is quite impressive given the AI server and memory headwinds, with storage being a key piece of this margin strength despite being a much smaller portion of revenue at $4.8 billion this quarter.    

Q4 gross profits were $6.7 billion or 20.2% of revenue compared to $5.7 billion or 23.7% in the same period last year. The lower margins reflect higher proportion of AI revenue mix.    

Q4 operating income grew by 43.2% YoY to $3.1 billion primarily driven by operating leverage. Operating margin was 9.3% compared to 9% in the same period last year.     

Q4 net income was $2.3 billion or 6.8% of revenue compared to $1.5 billion or 6.4% of revenue in the same period last year.   

Revenue: 

Dell’s Q4 revenue grew by 39.5% YoY and 23.6% QoQ to $33.4 billion driven primarily by outperformance in AI servers. Revenue growth accelerated by 28.7 percentage points from 10.8% YoY growth in the previous quarter and significant improvement from the (9.3%) QoQ decline in the previous quarter.  

Management also provided strong Q1 guidance of $34.7 billion to $35.7 billion, implying YoY growth of 50.6% and 5.5% QoQ at the midpoint. 

Cash: 

Similar to margins, Dell’s cash flows were equally as strong, with operating cash flow margin expanding by double digits and free cash flow following. Cash flow margins were also around the highest they’ve been over the last three years. 

Q4 operating cash flow was $4.7 billion or 14% of revenue compared to $585 million or 2.4% of revenue in the same period last year.  

Q4 adjusted free cash flow was $5.1 billion or 15.2% of revenue compared to $474 million or 2% of revenue in the same period last year. 

The company had a high debt of $31.5 billion compared to cash & investments of $13.3 billion at the end of Q4. The company repurchased shares worth $1.85 billion and paid dividends of $346 million in Q4. 

Valuation: 

Dell trades at a forward P/S ratio of 0.9. The company has traded at a minimum forward P/S ratio of 0.4 and a maximum of 1.3 in recent years. The company is currently trading at the mid-range. On the bottom line, it currently trades at a forward P/E ratio of 15.8. The company has traded at a minimum forward P/E ratio of 7.2 and a maximum of 22.3. The company is currently trading at the mid-range on the bottom line too.  

Notable Risks: 

The company had a high debt of $31.5 billion compared to cash & investments of $13.3 billion at the end of Q4. 

AI Software: 

Meta: Tied for the Best Mag 7 Stock 

Meta is an “eyeballs” company, and thus, an important lever to growth is increasing user engagement. In the most recent quarter, the company drove incremental engagement from ranking and product improvements. Primarily, the company optimized their systems to consider longer interaction histories to better identify a person’s interests. This led to the highest lift in feed views that the company has seen in two years: “The optimizations we made in Q4 drove a 7% lift in views of organic feed and video posts on Facebook, resulting in the largest quarterly revenue impact from Facebook product launches in the past two years.”  

Moving forward, Meta’s goal this year is to scale their training data to offer more personalized recommendations. By moving away from algorithms driving the feeds to LLMs, Meta can make the systems more responsive to real-time interest.  

This may seem like a subtle shift, but it’s actually not subtle at all – Meta is proposing a complete overhaul in how their systems surface content. Moving forward, LLMs will offer reasoning for a level of personalization not possible in the current approach, which is more pattern recognition based. Think of how Spotify works – it surfaces music you’ve already listened to. Facebook feeds are similar. However, moving forward, Meta can offer a personalized agent approach to where AI optimizes a feed to suggest content that does not require a direct signal.   

Here is what was stated on the call:  

“We're seeing in our early testing that personalized responses drive higher levels of engagement, and we expect to significantly advance the personalization of Meta AI this year. This dovetails with our investments in content understanding, which will enable our systems to develop a deeper understanding of each person's interests and preferences while also identifying the most relevant content across our platform to pull into responses.”  

Although Meta uses AI in its recommendations, the current systems are based on pattern and behavior-driven algorithms. For 2026, Meta will offer content that goes beyond the bounds of what you’ve already searched for/engaged with AI agents that can more intelligently infer your interests.   

The result will be more time spent on the platform and with higher engagement. Even incremental gains here will lead to more advertising dollars. 

The second area that Meta is making “big bets” by increasing monetization efficiency. Last quarter alone, the company doubled the number of GPUs used to train their GEM model for ads ranking. Similar to what was stated above, part of the improvements is using longer sequences of user behavior to inform the feed plus which ads are placed and when: “This new sequence learning architecture is significantly more efficient than our prior architectures which should enable us to further scale up the data, complexity and compute we use in our future ranking models to deliver performance gains.”  

Meta’s main approach to increasing the effectiveness of ad placements remains user targeting, but just smarter user targeting. This results in 4X better results than using AI to increase overall ad load: “In fact, in the second half of 2025, our initiatives on Facebook to redistribute ads across users and sessions delivered a nearly 4x larger revenue impact than Facebook ad load increases.”  

As you’ll see below in the Financials section, these improvements are making a material difference with Q4 revenue growing 17% QoQ and with a forward guide that implies the highest YoY growth rate for Meta since Covid-fueled 2021. 

Revenue: 

Q4 revenue grew by 23.8% YoY and 16.9% QoQ to $59.9 billion, beating estimates by 2.4%. Although strong sequential growth in Q4 is seasonal and Meta posted a 19.2% QoQ increase in Q4 2024, the current sequential growth is being achieved on a substantially higher revenue base of $51.2 billion versus $40.6 billion in the prior-year period. The strong revenue growth was primarily driven by robust demand stemming from AI advancements in ad recommendations, monetization, and user engagement.   

Management issued strong revenue guide of $53.5 billion to $56.5 billion, implying a 30% YoY growth and a sequential decline of (8.2%) at the midpoint. While the QoQ contraction reflects normal seasonality, the implied 30% YoY growth represents the fastest pace in the last 4.5 years. 

The company’s 2025 revenue grew by 22.2% YoY to $200.97 billion. Looking ahead, revenue growth is expected to accelerate 2.7 percentage points to 24.9% YoY growth to $250.97 billion in 2026 and will moderate to 17.8% YoY to $295.7 billion in 2027. 

AI Revenue: 

Meta is already seeing tailwinds from AI recommendation models driving higher ROI for advertisers following increased time spent across its family of apps.   

Q4 advertising revenue grew by 24.3% YoY to a record $58.1 billion. Notably, absolute advertising revenue growth reached $11.3 billion in the quarter, surpassing the $10.2 billion increase recorded in Q3. 

Perhaps the most important metric for Meta’s ad monetization is Family ARPP (average revenue per person). It reached a record $16.56 in Q4 2025, highlighting that Meta’s AI-driven ad performance improvements and monetization efforts are bearing fruit. While the 16.2% YoY growth in Q4 reflects a deceleration from the 17.7% seen in Q3, such a trend is common on a higher base and less of a concern given Meta is guiding for a modest acceleration this fiscal year. Notably, Q4 ARPP outpaced the 15.6% growth recorded in the prior-year period. 

EPS: 

Q4 GAAP EPS grew by 10.7% YoY to $8.88, beating estimates by 8%, driven primarily by higher revenue from stronger AI monetization. Analysts expect EPS to grow 2.9% YoY to $6.6 in Q1 2026. 

Looking ahead, GAAP EPS is expected to grow 25.8% YoY to $29.5 in 2026 and 15.9% YoY to $34.2 in 2027. 

Margins: 

Q4 gross margin was 81.8%, up 10 basis points YoY and down 20 basis points sequentially.  

Q4 operating income grew by 5.9% YoY to $24.7 billion with an operating margin of 41.3%, up 130 basis points sequentially and down 700 basis points YoY primarily due to higher AI-related operating expenses.  

Q4 net income grew by 9.3% YoY to $22.77 billion with a net profit margin of 38% compared to 43.1% in the same period last year. 
Cash: 

Meta’s cash flows improved in Q4, driven by higher profits.   

Q4 operating cash flow grew by 29.4% YoY to $36.2 billion with an operating cash flow margin of 60.5% compared to 57.8% in the same period last year and 58.5% in Q3. 

Q4 free cash flow grew by 7% YoY to $14.1 billion with a free cash flow margin of 23.5% compared to 27.2% in the same period last year, and 20.7% in Q3.  

The company had cash & marketable securities of $81.6 billion and debt of $58.7 billion. 

Valuation: 

Meta trades at a forward P/S ratio of 6.4. The company has traded at a minimum forward P/S ratio of 5.3 and a maximum of 9.9 in recent years. Meta is currently trading slightly lower than mid-range. On the bottom line, it trades a forward P/E ratio of 21. Meta has traded at a minimum of 15.2 and a maximum of 27.9. Meta is currently trading around the mid-range on the bottom-line. 

Notable Risks: 

Elevated capex increases financial risk by requiring substantial upfront investment before returns are fully realized. If monetization lags spending, margins and sentiment could come under pressure. 

Google: Tied for First Place Among Mag 7 

Revenue: 

Google delivered Q4 revenue of $113.83 billion, up 18.2% YoY, accelerating from 16.2% YoY in Q3 and marking the fastest growth since Q1 2022, driven by the accelerations in both Search and Cloud revenues.  

AI Revenue: 

Of the Big Three, Google reported the strongest AI-driven cloud acceleration this quarter, coupled with strong AI metrics and backlog growth that support this acceleration continuing through 2026.    

Google Cloud growth accelerated each quarter this year, though Q4 recorded the sharpest acceleration at 14 points to 48% YoY, with revenue coming in at $17.66 billion. Notably, this marked the segment surpassing a $70 billion annualized run rate, up from less than $50 billion annualized at the start of 2025. This would also mark its fastest revenue growth in more than four years. For Q1, Google expects strong growth to continue despite having tight accelerator supply.   

While the sharp acceleration is certainly impressive, sequential growth figures show a strong underlying trend within Cloud – for three quarters in a row, Cloud has delivered >$1 billion in QoQ growth, with each quarter larger than the last and Q4 increasing more than $2.5 billion versus Q3. Putting this in perspective to highlight Google Cloud’s strong AI-driven momentum, this was nearly as large as a QoQ increase as AWS, which rose $2.57 billion sequentially despite being double the size of Google Cloud.   

In percentage terms, Cloud growth accelerated from ~11% QoQ in Q2 and Q3 to 16.5% QoQ in Q4; this compares to 7.8% QoQ for AWS in Q4 and likely <2% QoQ for Azure.   

Google also provided a handful of stats accentuating AI’s impacts to growth. Revenue from products built on Google’s own genAI models increased nearly 400% YoY. Revenue from third-parties building AI applications rose 300% YoY. In total, Google Cloud has 14 product lines spanning infrastructure, platform and high-margin AI products and services exceeding $1 billion in annual revenue.    

It’s important to note that growth is currently off of a small base, thus the market will likely look toward overall AI revenue to justify the capex increase Alphabet is guiding for. While there were no specific updates to Cloud’s AI revenue or contribution, assuming that AI contributed roughly half of the quarter’s 48% YoY growth, this would place AI’s run rate at more than $11 billion.   

EPS: 

Google reported EPS of $2.82 in Q4, beating estimates by 6.8% and representing a growth of 31.1% YoY.   

Margins: 

Google reported a gross margin of 59.8% in Q4, up 1.1 points YoY. 

Operating margin was 31.6%, down 0.5 points YoY but up 1.1 points QoQ. 

Net margin was 30.3% in Q4, up 2.8 points YoY but down 3.8 points QoQ (as Google recorded a $10.7 billion gain on equity investments in Q3 that impacted the bottom line).   

Cash: 

Google’s cash flows remained rather resilient in 2025, with free cash flow margin declining marginally in the face of a 74% increase in capex. However, FCF must be tracked closely as the capex surge could easily bring free cash flow margin to the single-digits. Google has guided capex of $175-$185 billion in 2026, up 96.7% YoY at the midpoint. 

In Q4, Google reported operating cash flow of $52.4 billion for a 46% margin, up from a 40.6% margin in the year ago quarter but down from a 47.2% margin in Q3. For the full year, Google reported OCF of $164.7 billion for a 40.9% margin, up from 35.8% in 2024.  

Q4 free cash flow was $24.55 billion for a 21.6% margin, contracting on both a YoY and QoQ basis, from 25.8% in the year ago quarter and 23.9% in Q3. For 2025, free cash flow was $73.3 billion for an 18.2% margin, down from 20.8% in 2024.  

Looking ahead to 2026, analysts project Google to generate operating cash flow of $195.9 billion, though this would leave just $15.9 billion in FCF at the midpoint of capex guidance. Based on current revenue estimates for $471.4 billion, this would roughly project FCF margin to be 3.4%.  

Google’s balance sheet remains healthy with cash and marketable securities of $126.8 billion, while debt was $46.5 billion, up from $21.6 billion in Q3 as Google issued more than $26.5 billion in debt in the quarter. Debt is likely to rise sharply again in Q1 as Google’s recent bond sale reportedly took in over $30 billion. 

Valuation: 

Google trades at a forward P/S ratio of 8.6. The company has traded at a minimum forward P/S ratio of 4.4 and a maximum of 9.7 in recent years. Google is currently trading slightly higher than mid-range. On the bottom line, it trades at a forward P/E ratio of 28.9. Google has traded at a minimum of 13.7 and a maximum of 30.5. Google is currently trading slightly higher than the mid-range on the bottom line too. 

Notable Risks: 

The high capex would put pressure on the company’s cash flows.  

Reddit: The Scarce Asset in an AI-Generated Internet 

Reddit reported revenue of $725.6M for 70% YoY growth and 24.1% QoQ growth, which reflects seasonality from the holiday quarter. When comparing to last year's Q4, the company reported 130 basis points higher growth on a QoQ basis – no small feat given the tough comps the company is lapping with six quarters of 60%+ growth.   

The bottom-line shines with this stock as adjusted EBITDA was 45.1%, up from 36.1% in the year ago quarter. The GAAP operating margin of 31.9% has expanded sizably from the 12.4% margin reported last year for operating income of $232M. The free cash flow margin is 36.3%, leading the company to announce $1 billion in share repurchases.   

Although many investors consider Reddit niche compared to larger sites like Facebook or Google, the key metrics steadily move up on this audience of roughly 500 million monthly users and 120 million daily users. Global average revenue per user (ARPU) grew 42% YoY, up from 23% YoY growth in Q4 of last year. Advertising revenue also accelerated to 75% growth compared to 60% last year.  

Despite the strong report, the stock price has been slightly volatile. Management guided for a deceleration to 52.9% YoY growth, which leaves the market wondering if there is a catalyst in Reddit’s future. On the call, management pointed out they’ve guided conservatively for a few quarters now and discussed a new initiative to onboard advertisers at the bottom of the funnel with their AI-powered MAX platform. Another reason is that the company will no longer report logged-in users separately from logged-out users. This has been a point of contention for the Street for some time, which we covered in our previous analysis.  

That said, stocks with unwavering fundamentals with 50%-60% growth on the top line and 100%+ growth on the bottom line have a way of being mispriced quickly during periods of uncertainty. Consider that Reddit offers a Rule of 40 (revenue growth plus adjusted EBITDA margin) of 115 compared to Palantir’s Rule of 40 (revenue growth plus adjusted operating margin) of 127. Reddit’s rule of 40 is up 7 percentage points sequentially and 8 percentage points YoY.  

Reddit has always monetized through advertising, but Reddit Max marks a shift from primarily brand and contextual ads toward AI-driven, automated performance advertising that can increase the number of advertisers that Reddit onboards.   

Although early, this could put Reddit on the map for using its personalized data to compete for ad dollars in performance advertising. Should it prove successful, this would also be a strong motivating factor for Reddit to drop the logged-in/logged-out user metric given users will see the performance ads regardless of logged-in status. Most importantly, these ads monetize at a higher rate than brand ads. 

Revenue: 

Reddit once again reported stellar revenue growth of 69.7% YoY and 24.1% QoQ to $725.6 million. Revenue growth has been more than 60% for the sixth consecutive quarter. The company’s revenue beat estimates by a solid 8.8% and was better than last quarter’s beat of 6.4%. The strong revenue growth was primarily driven by 75% YoY growth in the advertising revenue to $690 million. While its other revenue, which includes licensing deals with Google and OpenAI, rose by a modest 8% YoY to $36 million. Regionally, U.S. revenue grew 68% and international revenue grew 78% YoY.  

Management guided Q1 revenue of $595M to $605M, implying a YoY growth of 52.9% YoY and down (17.3%) QoQ. The company’s Q1 guide beat the analysts estimates by 4% and was also stronger than last quarter’s beat of 3.5%. Analysts expect Q2 revenue to grow 42.8% YoY and Q3 revenue to grow 40% YoY to $818.8 million.   

Full year 2025 revenue grew by 69.4% YoY to $2.20 billion. Looking ahead, analysts expect 2026 revenue to grow by 42.7% YoY to $3.14 billion and 2027 revenue to grow by 30.2% YoY to $4.1 billion. 

AI Revenue: 

Q4 advertising revenue grew by 75% YoY to $690 million, accelerating from 74% growth in the previous quarter. Management attributed to impression growth as the main driver of revenue growth as the company’s AI investments are driving efficiency for advertisers delivering more outcomes and lower cost per action. Since last year, enhancements to the shopping ad ML models delivered over 75% improvement in advertisers return on investment.  

In Q4, click volume in the mid-funnel grew over 60% and lower funnel conversion volume doubled YoY. The company’s active advertisers grew by 75% YoY in Q4 and Reddit added new customers across its channels, including large, mid-market and SMBs.  

The company’s average revenue per user (ARPU) grew by 42% YoY and 19% QoQ to $5.98. ARPU growth accelerated from 41% YoY and 11% sequential growth in the previous quarter. 

The US ARPU grew by 53% YoY to $10.79. Although it slightly decelerated from 54% YoY growth in Q3, on a sequential basis it accelerated to 19% growth from 15% QoQ in the previous quarter. 

EPS: 

Q4 GAAP EPS grew by 244.4% YoY and 55% QoQ to $1.24, beating estimates by a solid 33.1%.  

Analysts expect EPS to grow by 286.6% YoY to $0.50 in Q1 and 86% YoY to $0.84 in Q2.  

Looking ahead, analysts expect 2026 EPS to grow by 56.9% YoY to $4.11 and 39.5% YoY to $5.74 in 2027. 

Margins: 

The company is experiencing strong profit growth, primarily driven by operating leverage.  

Q4 gross profits grew by 68.5% YoY to $666.9 million with a gross margin of 91.9%. The company reported its sixth consecutive quarter of above 90% gross margins.  

Operating margin improved by 19.5 percentage points YoY and 8.2 percentage points sequentially to 31.9% primarily driven by strong operating leverage.  

Net profit margin improved by 18.1 percentage points YoY and 6.9 percentage points sequentially to 34.7%.  

Q4 adjusted EBITDA grew by 112% YoY to $327 million with an adjusted EBITDA margin of 45.1%, beating the management guidance of 42.4%. Adjusted EBITDA margin improved by 9 percentage points YoY and 4.8 percentage points sequentially.  

Management has guided Q1 adjusted EBITDA margin of 35.8%, down 9.3 percentage points sequentially and up 6.4 percentage points YoY. 

Cash: 

Reddit reported strong cash flows primarily driven by record profits. The company’s balance sheet is robust, providing financial flexibility to invest in future growth and support share repurchases.  

Q4 operating cash flows grew by 196.5% YoY to $266.8 million with an operating cash flow margin of 36.8%, up 15.8 percentage points YoY.  

Q4 free cash flows grew by 195.7% YoY to $263.6 million with a free cash flow margin of 36.3%, up 15.5 percentage points YoY.  

The company has cash and marketable securities of $2.48 billion with no debt and cash increased by $250 million sequentially. 

Valuation: 

Reddit trades at a forward P/S ratio of 8.9. The company has traded at a minimum forward P/S ratio of 4.2 and a maximum of 24.4 in recent years. Reddit is currently trading significantly lower than mid-range. On the bottom line, it trades at a forward P/E ratio of 22.4. Reddit traded at a minimum of 18.2 and a maximum of 95.8. The company is trading significantly lower than mid-range on the bottom line too. 

Notable Risks: 

Reddit’s decision to stop reporting logged-in and logged-out user metrics in the second half of 2026 may lead to a transparency risk for investors. Those metrics help the market assess engagement quality, traffic mix, and monetization potential across the platform. Where the site ranks in terms of web traffic could change at any time as it’s entirely dependent on the Google partnership. 

Palantir: Commercial Surges, Yet Software Stocks Will be Tested 

Palantir reported another very strong quarter in Q4, with revenue accelerating to 70%, an impressive 57 point acceleration over the last ten quarters, while guiding for revenue to accelerate further to 73.6% in Q1.   

US commercial momentum remained unabated, with revenue accelerating 16 points sequentially to 137% YoY, surpassing the $500 million mark in the quarter. When looking at the strength of both QoQ and YoY growth, it’s likely Palantir represents the highest AI segment growth across the AI universe.   

On top of that, Palantir initially guided for fiscal 2026 revenue to accelerate from 56.1% to nearly 61% YoY, driven by US commercial revenue accelerating six points to >115% YoY. Driving such an acceleration at these growth rates is undeniably difficult, yet there are hints that Palantir could go above and beyond these figures by this time next year.   

If Palantir can outperform to a similar degree as 2025, such as 45-50 points above the first guidance, the revenue projection for US commercial would look much different. This scenario would need around a 10 to 12 point raise each quarter, and could project revenue as ~160% YoY, a 51 point acceleration. In dollar terms, this would project $3.82 billion, or ~$680 million above guidance.    

The main takeaway here is that even a modest outperformance and guidance raises of a few points each quarter could easily drive US commercial revenue growth to a double-digit acceleration from 2025’s 109% growth.   

Revenue: 

Palantir reported revenue of $1.41 billion in Q4, accelerating to 70% YoY while QoQ growth ticked 1.5 points higher to 19.1%, and marking a 50 point acceleration over the last two years. This also is Palantir’s highest revenue growth in their history as a public company.  

More impressively, Palantir guided for this revenue acceleration to continue into Q1 and for 2026, suggesting that the AI-driven growth engine that propelled shares higher through 2024 and 2025 is still intact, and potentially strengthening.  

Q1 revenue was guided to be $1.532 billion to $1.536 billion, accelerating 3.6 points to 73.6% YoY at midpoint (and what would be a fresh record growth rate), though QoQ growth would be just 9%.   

For 2026, Palantir offered an initial guide for $7.182 billion to $7.198 billion, up 60.7% YoY at midpoint, or $900 million ahead of consensus for $6.29 billion for 42.8% growth. This would also mark a 4.6 point acceleration, a significant feat considering the swift acceleration the company saw through the back half of 2025.   

AI Revenue: 

Palantir’s AIP-driven US commercial segment remains the company’s core revenue driver, with growth accelerating once again in Q4 to the fastest rate in four years. What’s more impressive is that Palantir not only has guided for US commercial revenue to more than double in 2026, but that it was guided to accelerate from 2025’s already-rapid 109% growth.  

US commercial revenue rose 137% YoY and 29% QoQ to $507 million in Q4, surpassing a $2 billion annualized run rate in the quarter, up from a $1 billion run rate at the start of 2025. QoQ growth accelerated only one point from 28% in Q3, though accelerating sequentially at this pace is difficult. 

On a YoY view, US commercial continued to accelerate, with the 137% growth in Q4 marking a 16 point acceleration from 121% YoY in Q3. Since the start of the year, US commercial revenue growth has accelerated a tremendous 66 points.   

RPO saw a meaningful step up in Q4, rising 62% QoQ to $4.21 billion, with YoY growth accelerating from 65.6% in Q3 to 143.4% in Q4. This also represented the company’s strongest RPO growth since the start of 2023 on both a YoY and QoQ basis.   

EPS: 

Palantir reported $0.25 in adjusted EPS in Q4, up 78.6% YoY, with GAAP EPS coming in at $0.24, up 700% YoY and beating estimates by 8.7% and 33.3% respectively.  

Palantir did not provide guidance for Q1, though consensus estimates currently call for adjusted EPS of $0.28, up 114.5% YoY, and GAAP EPS of $0.24, up 200% YoY.   

For the full year, Palantir delivered adjusted EPS of $0.75, up 82.9% YoY, and GAAP EPS of $0.63, up 231.6% YoY. Again, Palantir did not provide guidance for the forward fiscal year, though current consensus points to adjusted EPS up 76.2% to $1.32 and GAAP EPS up 79.4% to $1.13. 

Margins: 

While its revenue growth and acceleration are second-to-none in AI software, so are Palantir’s margins, with the company showcasing an impressive ability to drive margin expansion of >10 points while simultaneously accelerating revenue.  

For example, Palantir’s adjusted operating margin in Q4 was a record 57.4%, well ahead of its guidance for 52.4% and expanding 12 points YoY. This is a remarkable feat as it highlights Palantir’s ability to maintain its cost profile despite meaningfully accelerating revenue quarter after quarter.  

Adjusted EBITDA margin also showed strong expansion, coming in at 57%, up 6 points QoQ and 11 points YoY.   

Looking down the line, gross margins expanded nicely in Q4, with GAAP gross margin at 85%, up 6 points YoY and 3 points QoQ. Adjusted gross margin also expanded but at a smaller degree, up 3 points YoY and 2 points QoQ to 85%.  

The operating margin expansion was where Palantir shined. GAAP operating margin was 41% in Q4, up 40 points YoY (coming against a low comp due to the one-time stock appreciation rights (SARs) expense) and 8 points QoQ.  

As noted above, adjusted operating margin was 57.4%, up 12 points YoY and 6 points QoQ. Palantir guided for adjusted operating margin to remain strong in Q1 to 56.8% at midpoint, up 13 points YoY and down marginally QoQ. 

Cash: 

Palantir’s cash flows were robust in Q4, and management guided for adjusted FCF margin to expand in 2026 from an already strong 51% in 2025.  

Operating cash flow was $777.3 million in Q4 for a 55% margin, down slightly from a 56% margin in the year ago quarter but rebounding solidly from a 43% margin in Q3. For the year, Palantir delivered operating cash flow of $2.13 billion, or a 48% margin, up from 40% in 2024.  

Adjusted free cash flow was $791.4 million in Q4 for a 56% margin, down from a 63% margin a year ago but up from 46% in Q3. For 2025, Palantir generated $2.27 billion in adjusted FCF for a 51% margin, up from 44% in 2024.  

For 2026, Palantir guided for a step up in adjusted FCF, projecting it to increase more than 77% YoY to $3.925-$4.125 billion. This would represent an adjusted FCF margin of 56%, a five point expansion from 2025.  

Palantir’s balance sheet remained extremely healthy with cash of $7.18 billion and zero debt. 

Valuation: 

Palantir trades at a forward P/S ratio of 44.8. The company has traded at a minimum forward P/S ratio of 12.5 and a maximum of 112.3 in recent years. Palantir is currently trading significantly lower than the mid-range. However, forward P/S ratio > 30 is considered high. On the bottom line, it trades at a forward P/E ratio of 103.5. Palantir has traded at a minimum of 41.7 and a maximum of 285.8. Palantir is currently trading significantly lower than the mid-range. 

Notable Risks: 

Investors should be prepared for even well-insulated software names like Palantir and Cloudflare to face valuation pressure — not necessarily because their businesses are deteriorating, but because the pace of iteration from Anthropic, OpenAI, and a growing cohort of well-funded private startups continually resets the market's assumptions about who captures value in the AI stack and how quickly incumbents can be commoditized. 

Cloudflare: Strong Positioning, Timing is the Main Question 

While Big Tech witnessed weak price action following capex estimates for 2026, Cloudflare’s earnings report was being met with enthusiasm. Rather than competing with hyperscalers head-on, the company is taking a different route by offering an edge network where latency, global reach and lower costs matter more than compute and scale.   

Key metrics are suggesting an important inflection is underway, which was a theme from our coverage last quarter. Cloudflare reported the strongest revenue growth since Q1 2023. The company’s Q4 revenue grew by 33.6% YoY and 9.3% QoQ to $614.5 million, beating estimates by a solid 3.9%. The company’s Q1 revenue guide also beat estimates by 1%.    

The company reported a record new annual contract value (ACV) in Q4, which grew by nearly 50% YoY and was the fastest growth rate since 2021. Q4 remaining performance obligations (RPO) grew by 48% YoY and was the fastest growth rate since June 2022. Similarly, paying customers grew by 40% YoY and accelerated by 7 percentage points from 33% growth in the previous quarter. Notably, active developers on the Workers platform grew by 50% YoY to 4.5 million.   

Cloudflare’s CEO buried the lead a bit in the opening remarks, finally stating what is perhaps the most important element to his quarter’s beat: AI Agents. Although the key metric was provided for January, it’s clear that Cloudflare is seeing a strong inflection: “Over the month of January alone, the number of weekly requests generated by AI agents more than doubled across the Cloudflare network. This is driving increased demand for our whole platform.”  

According to management, this creates a “virtuous flywheel” as more agents drive more code execution on their Workers Platform, which in turn, drives more demand for their security products and networking services.  

AI agents also drive sheer infrastructure consumption for Cloudflare as agents look at many more sites and are always-on – which leads to more overall usage.   

Here was some commentary from the previous earnings call:  

“You've got a bunch of the agents of the world that are interacting with the Internet and they're interacting with it at a volume that we've just never seen before. And that's just driving more need for what are classically Cloudflare’s services. So the fact that more than 20% of the Internet sits behind us means that the agents have to interact with us, which means we have a seat at the table in defining exactly what the rules and the rails and the guardrails of the future of agentic commerce is going to look like; and be, and we are sitting in the middle of that.” 

Revenue: 

Cloudflare reported the strongest revenue growth since Q1 2023. The company’s Q4 revenue grew by 33.6% YoY and 9.3% QoQ to $614.5 million, beating estimates by a solid 3.9%. Revenue growth accelerated 2.9 percentage points from 30.7% growth in Q3 and was primarily driven by strong AI demand for its services, particularly from its enterprise customers. The company guided Q1 revenue of $620 million to $621 million, implying a YoY growth of 29.5% YoY and 1% QoQ and beating estimates by 1%.   

The company 2025 revenue grew by 29.8% YoY to $2.17 billion. Management provided a strong 2026 revenue guide of $2.785 billion to $2.795 billion, implying a YoY growth of 28.7% and beating estimates by 1.8%.  

AI Revenue: 

Note that Cloudflare does not have enough AI revenue to breakout into a standalone segment. However, many of the companies key metrics are benefitting from the overall increased internet traffic from AI agents with the CEO stating: “If you look at the last 30-plus years of the Internet and software ecosystem, they were built for human consumption, people in seats and clicks. Now the agentic Internet is emerging, and we can already see its trends. If humans looked at 5 sites when they were making a decision, agents might look at 5,000.” 

Over time, Cloudflare will see more revenue from edge inference, but right now, it’s mainly visible in internet usage. Here are some examples: 

Q4 remaining performance obligations (RPO) grew by 48% YoY and 16% QoQ to $2.496 billion, accelerating from 43% YoY and 8% QoQ growth in Q3. It was the fastest growth rate since June 2022. Current RPO was 63% of total RPO and grew 34% YoY. 

Cloudflare reported a record new annual contract value (ACV) in Q4. Matthew Prince said in the earnings call, “We blew away our previous record for new ACV in the quarter, with strong year-over-year and quarter-over-quarter acceleration. In Q4, new ACV book grew nearly 50% year-over-year, making it not only a record quarter in absolute ACV dollars but also the fastest growth rate we've delivered since 2021.” 

However, what has us on alert is the company’s billings grew by 27% YoY and 11% QoQ to $694.9 million. Among the key metrics, billings growth was blemish as it decelerated from 40% YoY and 12% QoQ growth in Q3. Billings represents real-time demand, and thus, RPO could be less meaningful if it’s signaling multi-year contracts. 

EPS: 

The company’s Q4 adjusted EPS grew by 47.4% YoY to $0.28, beating estimates by 3.2%. GAAP EPS was in line with estimates of ($0.03) compared to ($0.04) in the same period last year.  

Management Q1 adjusted EPS guide of $0.23 was lower than the estimates of $0.25. However, it implies a YoY growth of 43.8%. 

Margins: 

Q4 gross profits grew by 28.8% YoY to $452.5 million. Adjusted gross profits grew by 29% YoY to $460.18 million with an adjusted gross margin of 74.9%, down 270 basis points YoY and 40 basis points sequentially due to higher network expenses from the increase of paid customer traffic.   

Q4 operating loss was ($49.2 million) compared to ($34.7 million) in the same period last year. Adjusted operating income grew by 33.3% YoY to $89.6 million with an adjusted operating margin of 14.6%, which was flat YoY and down 70 basis points sequentially and beat the guidance by 40 basis points. Management Q1 guide is 11.4%. The company reported $132.4 million in stock-based compensation in Q4, which explains the difference between GAAP and non-GAAP operating income. 

Q4 net loss was ($12.1 million) compared to ($12.8 million) in the same period last year. Q4 adjusted net profit grew by 55.2% YoY to $106.8 million or 17.4% of revenue compared to 15% in the same period last year. 

Cash: 

Cloudflare’s Q4 operating cash flow grew by 49.6% YoY to $190.4 million with an operating cash flow margin of 31%, up 3 percentage points YoY and 1% QoQ. Similarly, free cash flows grew by 108% YoY to $99.4 million with a free cash flow margin of 16%, up 6 percentage points YoY and 3 percentage points QoQ. 

The company had cash and available-for-sale securities of $4.1 billion, while convertible senior notes outstanding were $3.27 billion at the end of Q4 2025. 

Valuation: 

Cloudflare trades at a forward P/S ratio of 23.1. The company has traded at a minimum forward P/S ratio of 13.8 and a maximum of 41.4 in recent years. Cloudflare is currently trading slightly lower than mid-range. On the bottom line, it trades at a forward P/E ratio of 164.2. Cloudflare has traded at a minimum of 90.6 and a maximum of 277.3. Cloudflare is currently trading slightly lower than the mid-range on the bottom line too. 

Notable Risks: 

The slowdown in billings is an important data point, as it may point to softer demand trends and reduced momentum in future revenue growth. At the same time, the company remains unprofitable on a GAAP basis, which suggests the path to sustainable profitability may be longer than investors typically want from a software company.  

Investors should be prepared for even well-insulated software names like Palantir and Cloudflare to face valuation pressure — not necessarily because their businesses are deteriorating, but because the pace of iteration from Anthropic, OpenAI, and a growing cohort of well-funded private startups continually resets the market's assumptions about who captures value in the AI stack and how quickly incumbents can be commoditized. 

Energy Stocks 

Bloom Energy 

The most important piece of information from Bloom’s Q4 earnings report was the company announcing its total current backlog at $20 billion, including $6 billion in product backlog, up 2.5X, and $14 billion in service backlog, up 1.5X.    

The backlog was driven by “half a dozen” hyperscale and neocloud customers compared to one customer a year ago.   

Bloom says the product backlog is attributable to its existing contractual commitments for purchases by a financier or customer in the future, including expected product revenue and anticipated ITC/tax incentives.   

Product backlog grew 140% year-over-year. Service backlog includes revenue for contracted operation and maintenance services for past and future product sales, in terms ranging from five to 20 years, meaning this backlog will take much longer to convert.   

Revenue: 

Bloom Energy once again delivered revenue more than 20% above analysts' expectations, with Q4 revenue of $777.68 million beating the consensus estimate for $643.5 million by 20.5%. This represented 35.9% YoY growth, decelerating from 57.1% YoY growth in Q3; however, sequential growth was very strong at 49.8% QoQ, accelerating from 29.4% QoQ in Q3 – this is because Q4 is typically Bloom’s seasonally strongest quarter. 

For the full year, Bloom reported record revenue of $2.02 billion, driven by significant AI data center growth and demand from commercial and industrial sectors. This represented 37.9% YoY growth. 

For 2026, Bloom guided for a sharp acceleration to 58% YoY at the midpoint of its guide for $3.1 to $3.3 billion, supported by its capacity expansion towards 2GW. This is a notable 24% beat over the consensus estimate and also would represent just 16% of its total $20 billion backlog. 

Product revenue was $638.5 million in Q4, up 35.4% YoY and 66.1% QoQ, though YoY growth did decelerate from 64.4% as Q4 faced a much tougher, seasonally strong comp. FY25 product revenue increased 41.1% YoY to $1.53 billion.   

Installation revenue was $67.3 million in Q4, up 86.4% YoY, though this did decelerate from 105.2% growth in Q3. FY25 installation revenue increased 66.9% YoY to $204.1 million.  

Service revenue was $61.7 million, up 14.7% YoY, decelerating slightly from 15.5% in Q3. FY25 service revenue increased 6.9% YoY to $228.3 million.  

Electricity revenue did reaccelerate in Q4 but growth continued to decline. Q4 revenue declined (5.3%) YoY to $10.2 million, improving from Q3’s (25.1%) decline. FY25 electricity revenue was $60.3 million, up 14.2%. 

Margins:   

Bloom’s margins showed a sharp sequential rebound in Q4 but remained lower on a YoY basis.  

Bloom reported GAAP gross margin of 30.9% in Q4, down 7.4 points YoY but up 1.7 points QoQ. Adjusted gross margin was 31.9%, also down 7.4 points YoY but up 1.5 points QoQ.   

GAAP operating margin was 11.3% in Q4, down 7 points YoY but up 9.8 points QoQ. Adjusted operating margin was 17.1%, down 6.2 points YoY but up 8.2 points QoQ. Bloom noted that it continues to focus on reducing product cost and driving operating leverage, which will likely be much more visible in 2026 based on its current guide.  

GAAP net margin was 0.1% in Q4, down 18.2 points YoY but up 4.5 points QoQ – to note, Bloom incurred a $66.2 million debt conversion expense charge that negatively impacted GAAP income this quarter. Adjusted net margin was 17.2%, down 3.5 points YoY but up 10.4 points QoQ. 

Earnings:   

Bloom reported GAAP EPS of $0.00 in the quarter, though adjusted EPS saw a large 50% beat, coming in at $0.45 versus the $0.30 estimate.   

Cash: 

Q4 is seasonally Bloom’s largest quarter for cash flows, with operating and free cash flow margins in excess of 50% this quarter, though this was much lower than the >80% margins it reported in Q4 2024. However, these large margins simply offset weak cash flows in the rest of the year, with full-year margins in the single-digit range.   

Operating cash flow was $418.1 million in Q4 for a 53.8% margin, down from an 84.6% margin in the year ago quarter.  

Free cash flow was $395.1 million in Q4 for a 50.8% margin, down from an 82.7% margin in the year ago quarter.  

Bloom reported $2.45 billion in cash, though debt rose to $2.61 billion, as Bloom raised $2.5 billion in convertible notes while also paying down $975 million in existing debt in the quarter.   

Valuation: 

Bloom Energy is currently trading at a peak forward P/S ratio of 19.1. The company has traded at a minimum of 1.4 in February 2024. On the bottom line, the company trades at a forward P/E ratio of 157.9. The company has traded at a minimum forward P/E ratio of 28.8 and a maximum of 462.4 in recent years.  

Notable Risks: 

Due to the strong stock outperformance of 1,180% in the past year. Bloom Energy is currently trading at a peak forward P/S ratio of 19.1 and leaves with less room for error if growth or guidance falls short of expectations. 

GE Vernova 

GE Vernova exited 2025 with one of the strongest AI demand and backlog profiles in the energy industry. In Q4, management emphasized accelerating slot reservations, rising pricing, improving backlog margins and multi-year visibility extending into the end of the decade.   

The company signed 6GW of incremental gas contracts in the final three weeks of December, bringing total Q425 contracts to about 24GW. As a result, the Gas Power backlog plus slot reservation agreements (SRAs) expanded from 62GW to 83GW sequentially.  

Management now expects to reach 100GW under contract in 2026, an upward revision from the 60GW discussed in mid-2025. Notably, the current 83GW under contract is heavily allocated toward 2029 delivery. By the time that 100GW is reached, both 2029 and 2030 capacity will be sold out.   

Revenue: 

GE Vernova Q4 revenue grew by 3.8% YoY to $10.96 billion, beating estimates by 7.1%, driven by rising AI energy demand. Organic revenue grew by 2% YoY to $10.8 billion. The company is a major beneficiary of the increasing energy requirements from the global AI infrastructure build-out, positioning the company as a key beneficiary of this secular trend. The continued slowdown in the Wind segment was offset by the growth in power and electrification segments that are benefitting from rising electricity consumption driven by data centers and artificial intelligence demand.   

AI Revenue: 

Q4 power orders increased 77% YoY to $11.7 billion, driven primarily by a sharp acceleration in gas power equipment orders, which more than tripled on higher volumes and favorable pricing. Gas turbine orders rose 71% YoY to 41 units, while power services orders grew 15%, reflecting continued customer investment in existing fleets.   

Q4 power segment revenue grew organically by 5% YoY to $5.7 billion. Management expects high single-digit organic growth in Q1. 

Electrification orders were 2.5x revenue and were up 50% YoY to $7.4 billion primarily due to growing grid equipment demand, particularly for synchronous condensers, substations partially to support data center growth and switchgear. The company also witnessed strong equipment orders growth in the Middle East, which increased over $1 billion and in North America, which more than doubled YoY.   

Q4 organic electrification revenue grew by 32% YoY to $2.9 billion primarily driven by strong growth in switchgear and High-Voltage Direct Current (HVDC) equipment. Management expects a similar revenue as Q4 in the next quarter, which will also include Prolec GE.   

Due to a sudden surge in AI-related electricity demand, the company’s turbine orders are vastly outpacing demand, and the company’s order book is sold out through 2028. 

Margins: 

The company’s Q4 adjusted EBITDA grew by 7.3% YoY to $1.16 billion with an adjusted EBITDA margin of 10.6%, an improvement of 250 basis points sequentially and 40 basis points YoY. Organic adjusted EBITDA margin improved 10 basis points YoY to 10.7%. 

Q4 net income was $3.7 billion or 33.5% of revenue compared to $484 million or 4.6% of revenue in the same period last year. The Q4 net income included a one-time tax benefit of $2.9 billion. 

Earnings: 

Q4 GAAP EPS was $13.39, up from $1.73 in the prior-year period, reflecting a one-time tax benefit of $10.58. Excluding this benefit, GAAP EPS would have been $2.81, below the consensus estimate of $3.13, primarily due to losses in the Wind segment. 

Cash: 

The company’s cash flows are improving driven by growth in profits and also improvement in working capital.   

Q4 operating cash flows grew by 169% YoY to $2.48 billion with an operating cash flow margin of 22.6% compared to 8.7% in the same period last year. The company benefitted from down payments on higher orders and slot reservations at Power as well as higher orders at Electrification.  

Q4 free cash flow grew by 214.7% YoY to $1.8 billion with a free cash flow margin of 16.5% compared to 5.4% in the same period last year. 

The company had cash of $8.85 billion and no debt at the end of Q4. In February, the company issued $2.6 billion of debt and completed the previously announced acquisition of the remaining 50% ownership stake of Prolec GE. 

Valuation: 

GE Vernova is currently trading at a peak forward P/S ratio of 6.0. The company has traded at a minimum forward P/S ratio of 0.96 in April 2024. On the bottom line, the company is trading at a forward P/E ratio of 69.4. The company has traded at a minimum of 38.5 and a maximum of 136.7 in recent years. GE Vernova is currently trading slightly lower than the mid-range on the bottom line.  

Notable Risks: 

The ongoing weakness in the wind segment is to be watched. That said, management expects a meaningful recovery in the wind business to materialize in the second half of 2026. 

NextEra Energy 

NextEra has traditionally been known as a regulated utility with a renewables development arm, yet is pivoting to become one of the few companies in the United States that can build power infrastructure at large scale across renewables, storage, gas, transmission and potentially nuclear. As you’re well aware, data center demand is insatiable, and NextEra’s ability to work across two growth engines is poised to benefit: Florida Power and Light provides the large, regulated utility platform while the Energy Resources solutions (NEER) provide renewables and storage. This can help break NextEra out of the bucket of being a passive beneficiary of load growth and into a builder that is enabling critical data center growth. In other words, NEE is pivoting toward being one of a handful of credible, large-scale solutions for the power demands of AI. 

Revenue: 

NextEra Energy’s (NEE) Q4 2025 revenue grew by 20.7% YoY and down (18.4%) QoQ to $6.5 billion. Revenue growth accelerated by 15.4 percentage points from 5.3% YoY growth in the previous quarter. The company is a beneficiary of AI data center energy demand. The Q4 sequential decline was seasonal as the company’s Q4 2024 revenue was down (21.7%) YoY and (28.8%) QoQ. 

During the Investor Day in December, management said they expect to develop data center hubs totaling 15 GW to 30 GW by 2035, and they reiterated this guidance during the Q4 earnings call. They have already identified 20 potential hubs and expect to identify 40 by the end of 2026. 

Margins: 

The company’s Q4 operating margin improved YoY, primarily driven by operating leverage.   

Q4 operating income was $1.59 billion or 24.4% of revenue compared to $941 million or 17.5% of revenue in the same period last year.   

Q4 net income was $1.54 billion or 23.6% of revenue compared to $1.2 billion or 22.3% of revenue in Q4 2024.  

Q4 adjusted net income was $1.13 billion or 17.4% of revenue compared to $1.1 billion or 20.3% of revenue in the same period last year.   

Earnings: 

The company’s Q4 adjusted EPS grew by 1.9% YoY to $0.54. Analysts expect Q1 adjusted EPS to be down (2.3%) YoY to $0.97 and expect adjusted EPS growth to accelerate to 5.6% and 12.4% in the subsequent two quarters.    

Looking ahead, analysts expect adjusted EPS to grow by 8.2% YoY to $4.01 in 2026 and 9% YoY to $4.37 in 2027. During the Q4 earnings, management reiterated its guidance set at Investor Day in December to grow adjusted EPS at a CAGR of 8% from 2025 to 2032 and at the same rate from 2032 to 2035. 

Cash: 

The company has steady operating cash flows. However, due to high capex, there is a wide difference between operating cash flow margin and free cash flow margin.  

Q4 operating cash flow was $2.5 billion or 38.4% of revenue compared to $1.98 billion or 36.8% of revenue in the same period last year.  

Q4 free cash flow was $519 million or 8% of revenue compared to $204 million or 3.8% of revenue in Q4 2024. 

The company had a high debt of $95.6 billion compared to cash of $2.8 billion at the end of Q4 2025. The company recently priced a $2.3 billion offering of equity units on March 3. The hybrid security will consist of a contract to purchase the common stock in about three years and undivided beneficial ownership interests in two series of debentures issued by NextEra Energy Capital Holdings. It provides the company with immediate liquidity while deferring common equity dilution for approximately three years. 

Valuation: 

NextEra Energy is currently trading at a forward P/S ratio of 6.1. The company has traded at a minimum forward P/S ratio of 4.2 and a maximum of 6.5 in recent years. NEE is currently trading slightly higher than the mid-range. On the bottom line, it trades at a forward P/E ratio of 23.1. The company has traded at a minimum forward P/E ratio of 16.1 and a maximum of 25. NEE is currently trading slightly higher than the mid-range on the bottom line too.  

Notable Risks: 

The company has high debt of $95.6 billion compared to cash of $2.8 billion at the end of Q4 2025. 

Conclusion: 

The 70-page report is not meant to explain the AI market or what AI companies do. Plenty of commentary already does that. Rather, the report and our I/O Fund Research site are designed to help our members act before the rest of the market catches up. What we offer is execution; not merely information. 

The AI trade is evolving, but the opportunity is far from over. If anything, the next phase will prove even more important as leadership broadens and the market becomes more selective. I can’t think of a better team to take on this challenge. 

Outsized returns will come not from following the crowd, but from being positioned ahead of it. That requires more than information. It requires judgment, discipline, and the willingness to act before consensus fully forms. Previous Quarterly Top 15 reports identified Bloom, Lumentum and AAOI early in their cycles, and the same discipline that found those names is driving this report. 

Our earnings season officially kicks off on Wednesday – Let’s go!

Royston Roche and Damien Robbins, Equity Analysts at I/O Fund contributed to this analysis.

I/O Fund Members Get 40% off Discovery 

Discovery members recently received an analysis on a stock poised to benefit as Nvidia tackles the context memory bottleneck and extends KV cache memory with its new Inference Context Memory Storage platform. To subscribe to Discovery with 40% off, click here to email usclick here to email us or email premium@io-fund.compremium@io-fund.com and mention code DISCOVERY40DISCOVERY40 

I/O Fund Pro Members Get 30% Off Advanced Market Signals 

Unlock real-time trade alerts, access to the I/O Fund’s momentum stock list, and weekly webinars every Thursday at 4:30 p.m. ET. To join Advanced Market Signals with 30% off, click here to email usclick here to email us or email premium@io-fund.compremium@io-fund.com and mention code ADVANCED30ADVANCED30

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• The I/O Fund’s Top 15 Stocks for Q1 2026
• The I/O Fund’s Top 15 AI Stocks for Q4 2025
• Micron Fiscal Q2: Record-Breaking Fundamentals
• Nvidia Q4: Stellar Report; Stock Remains Range Bound