Arm’s earnings call resulted in a sharp reversal as the stock was originally up 7%+ after hours yet settled down 6.4%. Overall, management’s commentary raised doubts in two areas – the first is whether Arm can secure the supply to meet AGI CPU demand, and the second is whether Arm has immediate inroads into Big Tech/Big Silicon given these larger players already license Arm’s IP and design their own custom CPUs. If the latter is true, then Arm’s near-term merchant silicon market is enterprises rather than hyperscalers.

Regarding supply, management stated the following on the earnings call: “As Rene noted or as Rene mentioned, customer demand for the ARM AGI CP was very strong. We now have line of sight to more than $2 billion of demand across fiscal '27 and '28. However, we are maintaining our outlook of $1 billion while we pursue supply chain capacity, and we still expect the first revenues from production ship sales to land in the fourth quarter of this fiscal year.”

In other words, regardless of demand, Arm's near-term AGI CPU revenue is supply-capped. The very catalyst that drove the stock's run into the print, anticipated demand for the new CPU, was effectively walked back in the near term.

AGI CPU Demand Doubles; But Won’t See Meaningful Revenue for 1-2 Years

We covered the launch of Arm’s AGI CPU and increasing CPU requirements being driven by agentic AI in the free newsletter, Arm Stock Could Win as Agentic AI Shifts the Bottleneck to CPUs.

Here is what we stated in our previous write-up:

“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.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.”

This long-term agentic AI-driven growth underpins Arm’s shift to chip design with the AGI CPU, with management emphasizing in FQ4 that demand for the AGI CPU is accelerating, with visibility into $2 billion through FY27 and FY28, double what was stated just six weeks ago at the CPU reveal in late March.

Despite this sharp increase in demand visibility, Arm is not yet ready to move the needle for AGI CPU revenue contribution, opting to maintain its FY27-FY28 target at $1 billion as it works to secure more supply. To secure this supply, Arm called out the following: “So the number that we talked about at the end of March was supply in place to support $1 billion of demand. And that includes memory that includes wafers, that includes packaging, that includes access to test equipment. So for the $2 billion, we are now in the process of securing supply to support that.”

To further complicate the supply-capped thesis, TSM recently exited its Arm position.

AGI CPU to Report $1B in Revenue in 2027-2028

Despite the excitement around Arm bringing to market a CPU with 2x the performance of x86 CPUs, the reality is that it will take time to secure memory wafers and ship in volume. The CFO indicated it would be FY28 (aligned to mid-CY2027 to mid-CY2028) before the $1 billion is recognized:

“The revenue split for '27, '28, something like [$90 to $100] million for Q4 '27 and then $910 million or whatever for '28. That's kind of what we laid out 5 or 6 weeks ago. And as said, we have demand above that. But for right now, let's just assume that's the number until we work through some of the wafer memory shortage issues.”

This is a rather long ramp for the AGI CPU, with Arm stating the first revenue will not come  until Q4 FY27 (next March) and contribute roughly $100 million – if you wanted to put this in perspective, Nvidia delivered nearly $700 million in data center revenue daily last quarter, and will scale much larger by this time next year.

However, for Arm, the new AGI CPU represents a rather lucrative revenue stream layering in on top of its existing IP business, despite operating far below the scale of Nvidia and AMD. Based on current consensus estimates for $7.61 billion in revenue in FY28, Arm’s projection for ~$910 million in revenue contribution from the AGI CPU would represent nearly 12% of revenue.

Assuming the full $2 billion of demand materializes and 80% converts to revenue in FY28, given Arm has >4 quarters to smooth out the supply chain and secure supply, this may present an ~$800 million uplift to consensus, or >10% on top of the $7.61 billion.

Arm IP Continues to be the Main Growth Driver

Data center royalty more than doubled YoY and is expected to double again in FY27, driven by large customers such as Amazon’s Graviton, Google’s Axion, and Nvidia’s Vera. Arm stated they have over 50% share with top hyperscalers and 100% share in DPUs/SmartNICs: “Royalty revenue grew 11% to $671 million with growth across Edge AI, physical AI and cloud AI, where our data center royalty has more than doubled year-over-year.”

Due to the Arm powering custom CPU programs, the CEO stated he foresees Arm being the largest CPU architecture by the end of the decade: “So we think it's a market that we can play in, in a very large way. And I think even indicators of AWS selling Graviton to outside partners — it's kind of an indication that there's just huge, huge demand for ARM-based capacity. So we think we're going to play alongside our partners in this space. And we also think the opportunity is very, very large for both. And I'm actually confident that by the end of the decade, I believe the largest market share by CPU type will be ARM.”

AMD Flexes Muscle for 50% Market Share with $100-120B TAM; Arm Offers a Rebuttal

AMD set the stage for strong server CPU growth earlier this week as it doubled its long-term industry growth forecast from 18% over the next three to five years to 35%, driven by increasing CPU requirements for agentic AI. This updated forecast now projects the server CPU TAM to reach over $120 billion by 2030, notably 20% higher than the $100 billion TAM Arm forecasted during its AGI CPU launch.

While discussing their new accelerated TAM, AMD’s management mentioned that they are confident in growing to >50% market share, implying a goal of capturing as much as $60 billion of the server CPU market.

In sharp contrast, Arm has stuck to its $15 billion AGI CPU revenue target by FY31, essentially implying a ~12% share based on AMD’s updated $120 billion TAM. Put differently, AMD is aiming to be 4X larger than Arm in AI-driven server CPU revenue by the turn of the decade, presenting stiff competition in server CPUs (Intel isn’t to be forgotten either).

However, when asked on the call about x86, Arm’s CEO offered a controversial take, which is that Arm CPUs will see nearly a 100% attach rate. The original statement was: “Those all connect to Arm. And increasingly, they are going to be 100% Arm. So we feel very, very good about the market share there.”

Here was the question on the call that offered a sharp rebuttal to AMD’s x86 bullish forecast:

“Timm Schulze-Melander   Rothschild & Co Redburn

So Rene, maybe just to start with you and to key off that CPU TAM commentary you just made there. I just want to check that I heard you right that you anticipate 100% attach rate of Arm CPU with those accelerators you mentioned? And then maybe just looking forward from an OpEx perspective, as you get into that merchant market, as your products attach to some of your partners' products, do you have any undertakings in terms of operating expenses in terms of in-market customer support? And then I had a quick follow-up for Jason.

Rene Haas   CEO & Director

Yes. Thank you for the question, Timm. Yes, so to clarify my comment, my expectation is that for the training platform over time, TPUs over time and NVIDIA's accelerated over time, I believe that the vast majority of the market share there will be Arm. NVIDIA is there essentially, and we are starting to see that happen with Graviton already over the last number of quarters and the announcement that Google made at Google Next with the TPU 8t and 8i, the training and inference chips. So that trend is well underway. And the reason for it, as stated, is that by getting much better performance in the same power envelope, the overall performance of the platform has greatly improved.

Google is talking about an 80% improvement in terms of the overall performance. So it's really numbers like that and the advantages that customers see in terms of embracing the platform that gives us very, very high confidence that, that trend should continue […]”

Technically, statements from both management teams offer an element of truth as there are many x86-hosted AI accelerator servers shipping today (hence Intel’s and AMD’s strong reports). While many hyperscalers deploy Arm-based servers internally, those same hyperscalers still run substantial x86 capacity for customer workloads.

Net-net, AI servers are primarily x86-hosted whereas mobile is entirely Arm-hosted. Arm is betting on a massive shift in the coming years, whereas AMD is offering the incumbent’s view.

In an important exchange with Vivek Arya, Arm CEO Rene Haas admitted the numbers don’t add up when taking management projections at face value for year-end CPU market share from AMD, Intel and Arm: “As far as the market share numbers, AMD has 50, Intel has 50 and we have 50. So you add up to some crazy number.”

Haas also had a quick comment about the AGI CPU’s initial customers that carries quite an important readthrough. Launch partners like Cloudflare, SAP or SK Telecom are adopting the chip because they do not have the capex budgets and/or engineering expertise to design and deploy custom Arm-IP based CPUs at scale – this will likely remain at the hyperscaler level with chips such as Amazon’s Graviton or Google’s Axion.

The main readthrough from Arm’s answer here is that will primarily be serving the enterprise market with the AGI CPU, having to compete with AMD and others for customers wanting internal CPU capacity, while also having to make a compelling argument for customers to adopt the chip instead of simply using Arm-based CPUs like Graviton in the cloud. It also hints that there may not be much of a runway for the AGI CPU at the hyperscalers who do indeed have the budgets and have already successfully deployed Arm-based custom CPUs at scale.  

The truth is that – nobody knows how this will play out exactly. AMD’s management team doubled their TAM very quickly in a way that suggests they were caught off guard by the demand signals. Therefore, long-term forecasts are hard to predict in this space.

Financials:

Q4 Revenue Grew by 20%

Arm’s Q4 FY26 revenue grew by 20% YoY and 20% QoQ to a record $1.49 billion, beating the midpoint of management’s guidance ($1.470 billion) by 1.36%.

Royalty revenue decelerated from 27% YoY in Q3 to 11% YoY in Q4 with revenue of $671 million; this also represented a (9%) QoQ decline off Q3’s strong $737 million. YoY growth was driven primarily Cloud AI with data center royalties more than doubling YoY. Arm also continues to benefit from an increasing mix shift to Armv9 and CSS, which carry meaningfully higher per-chip royalty rates than prior architectures.

License and other revenue grew 29% YoY and 62% QoQ to $819 million, driven by continued strong demand for Arm IP, the timing and size of multiple high-value license agreements and contributions from backlog.

Management guided Q1 FY27 revenue to $1.26 billion at the midpoint (+/- $50 million), implying YoY growth of 19.7% but down (15.4%) QoQ on the typical seasonality that follows a Q4 license catch-up. The Q1 guide is roughly in line with consensus of $1.25 billion. Both royalty revenue and license and other revenue were guided to be up around 20% YoY in Q1 FY27.

For the full year, FY26 revenue grew 23% YoY to a record $4.92 billion — the third consecutive year of more than 20% revenue growth since IPO — with royalty revenue up 21% YoY to $2.61 billion and license revenue up 25% YoY to $2.31 billion. Looking ahead, analysts expect FY27 revenue to decelerate slightly to 20.9% YoY to $5.92 billion, before reaccelerating to 28.5% YoY to $7.61 billion in FY28, the latter benefitting from initial contribution of the Arm AGI CPU silicon business.

ACV Growth Decelerates to 22% YoY

Annualized contract value (ACV), management’s preferred metric for normalized license and other revenue, grew 22% YoY and 2% QoQ to $1.66 billion; this marked a six point deceleration from 28% YoY growth maintained over the last three quarters.

Remaining performance obligations (RPO), however, declined (7%) YoY and (4%) QoQ to $2.07 billion, marking the third consecutive quarter of YoY RPO declines, which management attributed to improvements in the timing of revenue conversion (i.e. faster recognition rather than weakening demand).

Arm also signed two more CSS licenses in the quarter, one for smartphones and the other for data center networking chips. Arm Total Access licenses increased by 6 in the quarter to 56 (up 27% YoY), now including more than half of Arm’s top 30 customers, while Arm Flexible Access customers increased by 11 to 329 (up 5% YoY).

Margins

Gross margin remained near best-in-class IP-business levels, but operating margin compressed YoY (despite opex coming in below guidance) as Arm continued to invest in R&D for the AGI CPU and CSS roadmaps. Management has indicated that FY26 should mark the peak of opex growth, with non-GAAP opex CAGR decelerating from a 26% pace in FY24-FY26 to a mid-teens CAGR through FY31, which should drive operating leverage.

• Q4 GAAP gross margin was 97.9%, up slightly from 97.7% a year ago. Non-GAAP gross margin was 98.3%, essentially flat with 98.4% a year ago.
• Q4 GAAP operating margin was 29.4%, down from 33.0% in the prior year period. Q4 adjusted operating margin was 49.1%, down from 52.8% a year ago, as adjusted operating expenses grew ten points faster than revenue, up 30% YoY to $734 million
• Q4 GAAP net margin was 21.0%, up from 16.9% a year ago, while adjusted net margin of 43.0% declined from 47.1% a year ago.

Full-year FY26 GAAP and non-GAAP gross margins were 97.5% and 98.2%, respectively, both increasing roughly half a point YoY. However, operating margins felt some pressure, with FY26 GAAP operating margin contracting 2.4 points to 18.3%, and adjusted operating margin contracting 3.7 points to 43%. This was driven by strong opex growth, up 33% YoY to $2.72 billion, 13 points faster than revenue.

This operating margin contraction flowed through to the bottom line, with FY26 GAAP net margin of 18.4%, down 1.4 points, and adjusted net margin of 38.4%, down nearly 5 points.

Adjusted EPS Grew 9%

Q4 adjusted EPS was $0.60, up 9.1% YoY and beating estimates for $0.58. GAAP EPS in the quarter was $0.29, up 45% YoY but missing consensus of $0.37 by (20.7%) on the higher GAAP opex line. Management guided Q1 FY27 non-GAAP fully diluted EPS to $0.40 at the midpoint (+/- $0.04), implying 12.5% YoY growth.

For the full year, FY26 adjusted EPS was a record $1.77, up 8.6% YoY, while GAAP EPS increased 13.3% to $0.85. Analysts expect FY27 adjusted EPS to accelerate to 21.4% YoY to $2.14 (up 21.4% YoY), with this accelerating extending further into FY28, up 37.4% YoY to $2.94.

Cash Flow and Balance Sheet

Cash flow generation was strong on a full-year basis, although Q4 cash conversion was light.

• Q4 operating cash flow was $260 million, essentially flat with $258 million a year ago, for an operating cash flow margin of 17.4%, down from 20.8% a year ago as receivables grew. FY26 operating cash flow was $1.52 billion for a 31% margin, up sharply from FY25’s $397 million for a 9.9% margin.
• Q4 adjusted free cash flow was $152 million, down from $163 million a year ago, for an FCF margin of 10.2%, down from 13.1%. FY26 adjusted free cash flow was $882 million for a 17.9% margin, up from just $99 million in FY25 (a 2.5% margin).
• Cash and short-term investments totaled $3.60 billion at quarter-end, up from $3.54 billion in Q3, and the company continues to carry no debt.

Conclusion:

My view is that Arm remains a critical long-term player in the AI data center buildout, but this earnings report introduced more uncertainty around the near-term growth story. The issue is less about AMD or Intel’s current dominance in x86 and more about Arm’s ability to secure supply at a time when even the largest AI semiconductor companies are capacity constrained.

After years of appearing relatively uneventful compared to other AI semiconductor peers, Arm is now better positioned to compete as AI workloads expand from mobile and edge devices into the data center. The key unknown is valuation, especially because merchant CPU revenue will take time to scale, and investors may need to rely on Arm’s traditional IP engine as the primary growth lever for the next one to two years.

My takeaway is that the near-term AGI CPU narrative should be priced lower due to a capped growth trajectory, but Arm’s long-term strategic relevance remains intact.

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

Recommended Reading:

• Coherent FQ3: InP Capacity Doubling to Drive CY26 Inflection
• Lumentum FQ3: Firing on All Cylinders Despite Stiff Supply Constraints Across EMLs, Pump Lasers
• Astera Labs: Important QoQ Acceleration, Product Road Map is Loaded
• AMD Q1: Doubled CPU TAM, Helios Incoming for Q4

Astera Labs is navigating an important rite of passage that many post-IPO hypergrowth companies stumble through, which is to offer a consistent growth trajectory. Once the excitement of an IPO fades, most tech companies cannot sustain the growth the private sector primed the company for ahead of listing on the public markets.  

Astera is bucking this trend, as the company was expected to report 8% QoQ growth and instead reported 14% QoQ growth, which resulted in an official 1.6 percentage points acceleration on a YoY basis from 91.8% YoY growth last quarter to 93.4% growth this quarter. Looking ahead, Astera is offering a guide that indicates QoQ growth of 16.7% and YoY growth of 87.6% for revenue of $360 million at the midpoint. This handily beat forward expectations for revenue of $310.1 million next quarter.

As we look toward the second half of the year, management offered strong commentary that suggest growth will continue: “As we look to the second half of 2026, robust demand reflects secular AI infrastructure spending, deep customer partnerships and expansion towards higher-value solutions within our portfolio. […] As a result, we expect strong revenue growth to continue through 2026 and into 2027, driven by the proliferation of AI fabrics and the industry's transition to PCIe 6, 800 gig and 1.6T Ethernet connectivity.” 

Perhaps most notable is that Astera delivered a strong quarter even after the stock declined roughly 55% peak-to-trough from September through March, reflecting a disconnect between sentiment and fundamentals. 

Scorpio-X 320 Lane Smart Fabric Switch 

Positioning: 

In this evening’s print, Astera also announced the Scorpio X-Series 320 Lane Smart Fabric Switch, which is the largest open, memory semantic fabric switch on the market with 5.12 TB/s bidirectional bandwidth in a single ASIC. The 320 Lane variant offers 16 lanes per device and 20 accelerators per switch, which is roughly “2x the radix in a single hop,” which means twice the number of GPUs are connected on the same switch. With Scorpio-X, only one switch is needed for 320 GPUs, and fewer switch hops means lower latency.  

Astera differentiates itself from Broadcom’s Ethernet switch Tomahawk 6 and Nvidia’s NVSwitch by providing an open PCIe-based fabric for CPUs, NICs and storage with the P-Series and improving accelerator-to-accelerator performance specifically around memory sharing with the X-series. 

The X-Series is timed to the scale-up networking opportunity and the inference market. Mixture of Experts (MoE) inference is a steady stream of tasks, which requires very fast accelerator-to-accelerator communication. As discussed in the call, MoE requires frequent routing of tokens and data across expert models, which places more emphasis on the scale-up fabric. Astera Labs is uniquely positioned to enable GPUs and AI accelerators to communicate more efficiently across PCIe, especially when it comes to direct memory access. 

Here is what was stated in the opening remarks: 

“Scorpio X-Series portfolio now supports up to 320 lanes for high radix scale-up networking, and Scorpio P-Series PCIe 6 portfolio now spans 32 to 320 lanes for diverse system topologies, making it the broadest in the industry.  

Our new flagship Scorpio X-Series 320 lane has been purpose-built to maximize AI economics by leveraging hardware-accelerated hypercast and in-network compute engines to boost collective operations by up to 2x. In-network compute offloads critical accelerator to accelerator communication and computation directly onto the switch, dramatically reducing the networking overhead during large-scale training and inference.” 

This is a significant shift as it brings the math operations inside the switch instead of the GPUs, which Astera is referring to as “in-network compute.” Hypercast refers to handling operations inside the switch, which reduces the networking overhead associated with GPU-to-GPU coordination. The result for inference tasks is more tokens per dollar as Scorpio-X removes the need for GPUs to wait on other GPUs during MoE and agentic workloads.  

It's important to double-click on the memory-semantic piece. Astera's fabric lets accelerators access each other's memory directly like a single unified memory pool, eliminating the overhead of translating data into network packets. This is important for AI workloads, and especially MoE inference, which depend on constant sharing of weights, activations, KV cache, etc., across accelerators. Per the press release: “Its memory-semantic connectivity enables accelerators to access fabric resources through native load/store operations, eliminating software overhead and improving fabric efficiency at scale.” 

Astera’s X-Series offers communication across mixed architectures (both GPUs and ASICs) but also solves for memory sharing – both are key as we move into the inference market.  

Economics: 

We’ve covered the X-Series for about a year in our post-earnings analyses. For investors, some of the most important takeaways is that the Scorpio product is expected to increase from 15% of product mix at the end of CY25 to 50% of product mix by the end of CY26. Although the P-Series is driving the current growth, the X-Series will be the higher mix as we exit the year – which means this ramp is second-half weighted. 

“Given the size of the opportunity and the associated dollar content, we would expect to see that Scorpio will become our largest product line by the end of the year, which is strong performance for a product line that was only 15% of total company revenue last year. And as we go throughout the year, I would expect to see X-Series revenue exceeding P-Series.” 

Another point for investors is the average sales prices will increase from the X-Series. Here is what was stated on the call: 

“Yes. So in general, what I would say is the bigger the switch, the higher the ASP. That's the way industry works. But also, please keep in mind is that these switches are more like AI fabric class device, which are a lot more than just the number of lanes, right? […]So when it comes to ASP, obviously, it's a combination of how — what features are enabled and not just based on the port count. But we do see that our content continue to increase. And to that standpoint, we are expecting and going forward with the design wins we have, over $1,000 worth of content per accelerator.” 

Future Product Roadmap for 2027-2028 

Optical Opportunity: 

Astera’s optical roadmap is an extension of the company’s ability to offer end-to-end PCIe over optics for GPU clusters. As racks grow into larger pods, cable length and signal integrity become constraints. Astera has stated at a recent investor’s event that optical becomes necessary at higher data rates (which is also general consensus).  

Last October, Astera acquired a scale-up photonics company to offer optical scale-up interconnects. On the earnings call, it was shared that near-packaged optics will roll-out first following this acquisition in 2027, which is a bridge solution while co-packaged optics may take longer than the market cares to wait. 

Here is what was stated on the call: 

“For us, in terms of time line, what we believe is that the NPO-based opportunities, or the near package optics, would be the first one to ramp, and that will start happening in 2027. We will also be ramping our pluggable connector technologies for CPO, mostly for scale-out next year, 2027, with more of the mainstream deployments for CPO happening in the 2028 time frame.” 

NVLink Fusion Opportunity: 

Notably, Astera Labs offers connectivity solutions for hybrid AI racks. This widens Astera’s content opportunity beyond UALink as it provides an additional path to scale-up AI fabrics by offering a bridging solution for GPUs and custom silicon. In some cases, when NVLink is chosen, Astera will still be a key supplier for connectivity solutions. 

“Clearly, an area that we see tremendous opportunity for us going forward is the custom solutions under which we are developing the NVLink Fusion type of devices. And this actually is proving to be pretty interesting. We do have several opportunities. We're very deep in engagement for an initial design win in collaboration with NVIDIA and then a hyperscaler. So that project is going well. So we do expect that to start contributing revenue in 2027 as some of the GPUs that are designed for this kind of use case, which is called as a hybrid rack situation, where the GPU or the XPU still talks native protocols, which could be a protocol like PCIe or UALink and others. But then when they need to leverage and cross over and talk to an NVLink type of ecosystem, then they would need a product that's based on NVLink Fusion that we are developing.” 

CXL Opportunity: 

CXL is a longer-term opportunity for Astera Labs, and will extend Astera’s content opportunity (again) to include memory pooling and connectivity. This provides more direct exposure to the memory side of the AI buildout rather than only the accelerator interconnect. Here was the update for the call, including a newer customer win that could help with KV cache offload: “Finally, our LEO memory controller is on track for an early ramp of CXL attached memory with Microsoft Azure M-Series virtual machines. And during the quarter, we captured a new custom design win for a KV Cache offload application with shipments expected in 2027.” 

Note on UALink: 

We’ve written in the past that UALink as a scale-up fabric is expected to go head-to-head with Ethernet Scale-Up Networking (ESUN). In the past, when there are ESUN announcements, ALAB’s stock reacts negatively. However, that assumes a zero-sum outcome, whereas it’s more likely scale-up sees a mix of both UALink and ESUN. 

The quick refresher is that ESUN is attempting to make Ethernet work for scale-up whereas UALink was built from scratch for scale-up. The primary benefit ESUN offers is to move quicker than UALink (as discussed above, ALAB is saying it’ll be 2027 for UALink to be fully deployed). However, in the meantime, Astera’s PCIe solutions are in high demand and deployable now.  

Even if ESUN moves faster commercially, there is a performance gap that helps to ensure that Astera’s positioning with PCIe/CXL remains intact. That performance gap is best described as the low latency required for what are the most in-demand AI workloads today – those that require memory pooling and GPU-to-GPU communication.   

For more information, read our previous analysis here.previous analysis here. 

Financials 

By Royston Roche 

Revenue Accelerates to 93.4% YoY 

Astera Labs reported Q1 2026 revenue of $308.4 million, beating estimates by 5.5%. Growth continued at a robust pace on a YoY basis, with revenue up 93.4% YoY and accelerating 1.6 percentage points from 91.8% growth in the previous quarter. On a sequential basis, revenue grew 14.0% QoQ from $270.6 million in Q4 2025.  

Aries product revenue grew strongly in Q1 2026, with PCIe Gen 6 solutions for both scale-out and scale-up signal conditioning driving solid adoption. Management noted that PCIe Gen 6 revenue across AI fabric and signal conditioning contributed more than one-third of total revenue in the quarter — a significant milestone reflecting the accelerating industry transition to Gen 6. 

The Scorpio product family also performed well in Q1, driven by strong demand for PCIe Gen 6 switching applications and continued expansion of designs across various platforms. During the quarter, Scorpio X-Series products began shipping in initial production volumes. Management expects Scorpio X-Series shipments to increase in Q2, along with initial shipments of the new Scorpio X 320 lane product and then ramp to full volume production in the second half of 2026. 

Taurus product family continued to deliver solid results in Q1 2026, driven by broad adoption of Active Electrical Cable (AEC) to extend reach in both AI and general-purpose compute platforms. 

Leo's CXL memory expansion products continue to advance, with management highlighting an early production ramp of CXL-attached memory with Microsoft Azure M-Series virtual machines and a new custom design win for a KV Cache offload application with shipments expected in 2027. 

Management guided strong Q2 revenue guidance of $355 million to $365 million, implying a YoY growth of 87.6% and 16.7% QoQ at the midpoint, beating estimates by 16.1%. Aries revenue growth is expected to be driven by continued strong adoption of PCIe 6 across AI platforms, supporting both scale-up and scale-out connectivity. Taurus growth is expected to be driven by increased volumes for AI scale-out connectivity. And in AI fabric, management expects robust growth driven by the continued early-stage ramp of the Scorpio X-Series products for large-scale XPU clustering applications as well as continued growth in the P-Series solutions and customized GPU platforms. 

Margins Beat Guidance 

Astera Labs delivered impressive gross margin performance in Q1 2026, with GAAP gross margin coming in at 76.3%, comfortably ahead of the 74% guidance. This compares to 75.6% in Q4 2025 and 74.9% in the same period last year, a sequential expansion of 70 basis points and 140 basis points YoY, primarily due to favorable product mix. Adjusted gross margin improved 150 basis points YoY to 76.4%.  

Management has guided adjusted gross margin to be lower at 73% for the next quarter, primarily due to the estimated 200 basis point noncash impact related to a recently executed warrant agreement with one of its customers. 

GAAP operating margin improved 13 percentage points YoY to 20.1%. While adjusted operating margin improved 2.5 percentage points YoY to 36.2% primarily due to operating leverage and beat the guidance of 34.5%.  

Q1 2026 adjusted net income grew by 84.7% YoY to $110.7 million or 35.7% of revenue compared to 37.4% of revenue in the same period last year.  

Adjusted EPS grew by 84.8% 

Q1 adjusted EPS grew by 84.8% YoY to $0.61, beating estimates by 13.5% primarily due to operating leverage. GAAP EPS growth was even stronger as it grew by 144.4% YoY to $0.44 and beating estimates by 26.5%. 

Management also provided a strong EPS guide for the next quarter. GAAP EPS guide is $0.45 at the midpoint, up 55.2% YoY and beat estimates by 37.6%. Adjusted EPS guide is $0.69 at the midpoint, up 56.8% YoY and beat estimates by 25.5%. 

Cash Flow and Balance Sheet 

The company’s cash flows were strong primarily due to higher profits.  

• Q1 operating cash flow was $74.6 million or 24.2% of revenue compared to a mere $10.5 million or 6.6% of revenue in the same period last year. 
• Q1 free cash flow was $67 million or 21.7% of revenue compared to $5.97 million or 3.7% of revenue in the same period last year.  
• The company maintains a robust balance sheet with cash & marketable securities of $1.18 billion and no debt.  
• Inventories rose 2% QoQ to $60.2 million. 

Conclusion: 

Astera Labs is expanding their product road map well beyond selling PCIe retimers, and is now solving serious bottlenecks for the incoming AI inference market. Inference workloads are more memory-intensive and will see ongoing, exponential accelerator-to-accelerator communication, not to mention the critical importance of shared memory access.  

Astera’s role is becoming more strategic as the company has multiple paths to increase content through Scorpio-X scale-up switching, both UALink and NVLink Fusion content opportunities, CXL memory pooling, and they’re prepared for the optical transition – whew, that’s a lot. Near-term volatility could persist as the market debates protocol winners, but one thing is for certain – AI workloads are becoming more complex. Astera is on the front lines of solving that complexity.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in ALAB at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• The I/O Fund’s Top 15 Stocks for Q2 2026
• AMD Q1: Doubled CPU TAM, Helios Incoming for Q4
• Palantir Q1: Strong Headline Numbers; TCV to be Watched
• Reddit Q1: Bottom Line Expansion Will Face Off with Revenue Deceleration in Q2

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 that: “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. 

The majority of the Q&A was about the $100 billion comment, including surfacing why this estimate may prove to be far too low.  

$100 Billion for Chips Alone … and Counting 

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: 

“Now to clarify your first part, Blayne. When I say we forecast, we have a line of sight that our revenue in '27 will be significantly in excess of $100 billion, I'm focusing on the fact that these are pretty much all based on chips, whether they are XPUs, whether they are switch chips DSPs, these are silicon content we're talking about.” 

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. 

When you add the words “significantly in excess” of this amount, it spells a solid runway for Broadcom. 

The analysts in the Q&A session were not satisfied with this blanket comment and began poking around with one analyst doing the math on the customers plus what each GW is worth, and coming out closer to $180 billion to $200 billion in AI revenue. Here was the exchange – the CEO did not confirm the math but it still helps to know what the sell-side is thinking: 

“Stacy Rasgon Bernstein: 

I don't know if this is for Hock or Kirsten, but I wanted to dig in a little more to this substantially more than $100 billion next year. I'm trying to just count up the gigawatts. I counted, I don't know, 8 or 9, you have 3 from Anthropic, one from open AI, so that's 4. You said Meta was multiple, so east, that gets you to 6. Google, I figure should be bigger than Meta.so like at least 3, that's 9 and then you got a few others.  

I just thought that your content per gigawatt was sort of, call it, a $20 billion per gigawatt range. I guess what I'm asking, is my math around the gigawatts you plan to ship in 27 correct? And how do I think about your content per gigawatt as that ships — maybe we'll be "substantially" more than $100 billion. 

Hock Tan CEO: 

Stacy, you have a very interesting perspective, and I've got to [ remind ] you for that. But you're right. You can look at it a gigawatts, which is the right way to look at it instead of dollars because that's how we sell our chips. So you have to realize we — depending on our LLM customer, our 6 customers, sorry, not 56. The dollars per gigawatt is varies, sometimes quite dramatically. — it does vary. But you're right. It's so far from the dollars you are talking about. And if you look at it by gigawatt in '27, we are seeing getting close to 10 gigawatts.” 

OpenAI to Drive Strong Fiscal 2028  

During the opening remarks, the CEO stated that OpenAI represents 1 GW stating: “We expect OpenAI deploying in volume, their first-generation XPU in 2027 and over 1 gigawatt of compute capacity.” Later, an analyst drilled into the math to where OpenAI will eventually represent 10GWs, which could make for a strong inflection come 2028. 

“Joshua Buchalter, TD Cowen: 

Congrats on the results. I appreciate all the details on the expectations for deployment at specific customers. I was hoping you could just maybe reflect on how visibility has changed over the last 1 to 2 quarters that gave you the confidence to give us more details? And then on a specific one, you mentioned greater than 1 gigawatt for OpenAI in 2027, with that deal being for 10 gigawatts through 2029, that implied a pretty sharp inflection, I guess, in 2028. Is that the right way to think about it? And was that sort of always the plan? 

Hock Tan, CEO: 

Yes. Well, yes, this — as you've all seen, and you all know in this generative AI race that we are in now, and I shouldn't use the word race, let's call it, progression among the few players we see here, I mean, it's a competition. Each is trying to create an LLM better than the other and more tailored for specific purpose, be it enterprise, be it consumer, be it search. Each one is trying to create it more and more. And all of that requires not just training, which is important to keep improving your LLM models. But inference for productization and monetization of your LLMs.  

And we are going — and probably call it the fact that we've been engaged with some of them now for more than a couple of years. We're getting better and better visibility as they have more and more confidence that the XPUs they are working on with us is achieving what they are getting it as they get the sense that the XPUs they are working on with the software, with the algorithms they needed, they are having more confidence that this XPU silicon is what they need.  

And it gets better and better. And It's get better, we get more visibility as Charlie puts up perfectly because at the end of the day, we only have 6 guys to work on. And these 6 guys are all, as I said, look at XPUs and AI in a very strategic manner. They don't think 1 generation at a time. They think multiple generation multiple years.” 

Speaking of large customers, in the exchange above, Broadcom refutes the idea that Meta is facing significant hurdles for its MITA chips, stating in the opening remarks that: “Now contrary, the recent analyst reports Meta's custom accelerator MTIA roadmap is alive and well. We're shipping now. And in fact, for the next generation XPUs, we will scale to multiple [ gigawatts ] in '27 and beyond.” 

Financials 

By Royston Roche 

Revenue grew by 29.5%

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 growth to continue, with revenue expected to grow 71.7% YoY to $27.39 billion in FQ3 and 74% YoY to $31.35 billion in FQ4.

Key Segments 

Semiconductor Solutions 

FQ1 Semiconductor solutions revenue grew by 52% YoY and 13% QoQ to $12.5 billion and was better than the management guidance of $12.3 billion. Revenue growth accelerated by 17 percentage points from 35% growth in the previous quarter. Management expects semiconductor revenue to further accelerate to 76% YoY and 18% QoQ to $14.8 billion in the next quarter, driven by the surge in AI revenue.

FQ1 AI revenue grew by 106% YoY and 29% QoQ to $8.4 billion. Revenue growth accelerated from 74% YoY and 25% QoQ in the previous quarter. Management expects strong growth to continue in the next quarter and AI revenue is expected to grow 140% YoY and 27% QoQ to $10.7 billion.  

The company’s CEO, Hock Tan, said in the earnings call, “Now our custom accelerator business grew 140% year-on-year in Q1. This momentum continues in Q2. The ramp of custom AI accelerators across all our 5 customers is progressing very well. For Google, we continue our trajectory of growth in '26 with strong demand for the seventh-generation Ironwood TPU. In 2027 and beyond, we expect to see even stronger demand from next generations of TPU. For Anthropic, we are off to a very good start in 2026 for 1 gigawatt of TPU compute. And for '27, this demand is expected to surge in excess of 3 gigawatts of compute. Our XPU franchise, I should add, extends beyond TPUs.” 

Non-AI semiconductor FQ1 revenue was flat YoY at $4.1 billion, in line with guidance. Enterprise networking, broadband, server storage revenues were up YoY, offset by a seasonal decline in wireless. In FQ2, management expects non-AI semiconductor revenue to be $4.1 billion, up 4% YoY.

Infrastructure Software 

FQ1 Infrastructure Software revenue came at $6.8 billion, up 1% YoY and down (2%) QoQ and was in line with the guidance. VMware revenue grew 13% YoY. Management expects infrastructure software revenue to grow 9% YoY and 6% QoQ to $7.2 billion in the next quarter. 

Margins 

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

• FQ1 gross profits grew by 29.7% YoY to $13.16 billion. 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. Management has guided adjusted gross margin to be flat sequentially to 77% and down 240 basis points YoY. 
• FQ1 operating income grew by 36.8% YoY to $8.56 billion. 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. Adjusted net income grew by 30.2% YoY to $10.19 billion with an adjusted net profit margin of 52.7% compared to 52.4% 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%. Management has guided FQ2 adjusted EBITDA guidance to be flat sequentially and up 100 basis points YoY to 68%.

Adjusted EPS grew by 28% 

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. Analysts expect strong growth in the coming quarters and adjusted EPS is expected to grow 36.9% YoY to $2.16 in FQ2 and 69.5% YoY to $2.86 in FQ3.

Cash Flow and Balance Sheet 

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. Management authorized an additional $10 billion share repurchase program effective through the end of calendar year 2026. 
• Inventory increased by 30.4% QoQ to $2.96 billion to support the strong AI demand. 

Conclusion: 

Broadcom’s most consequential message this quarter was not the near-term guide, but the long-range visibility: management stated it has line of sight to AI-related revenue in FY2027 “significantly in excess of $100 billion.” 

If that framework proves accurate, the market will be forced to shift from debating whether AI spend can persist to modeling how Broadcom scales into that demand across XPUs and networking as customers expand from early deployments into multi-gigawatt clusters.  

Notably, management’s comments imply substantial silicon content per gigawatt—for example, Anthropic’s trajectory has been discussed in terms that can translate into roughly $20 billion per GW—and they also referenced a back-half weighted OpenAI ramp that appears more likely to contribute in 2028–2029.  

And that’s only what we can quantify today …

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in AVGO at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• Nvidia Q4: Stellar Report; Stock Remains Range Bound
• Alphabet Q4: Cloud Sees 14 Point Acceleration to 48% Growth, FY26 Capex to Nearly Double
• 2025 Full Year Audited Returns
• AppLovin: Strong Fundamentals; Sentiment Remains a Hurdle

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. 

UALink Versus ESUN Debate  

The simplest way to settle the UALink versus ESUN debate is that AI systems using PCIe today are upgrading to PCIe6, and those relying on Ethernet will remain with that protocol. Rather than a winner-takes-all market, the most likely outcome is a hybrid system as PCIe is superior inside the rack and Ethernet is superior for scale-out between racks. 

As covered last quarter in the Q3 Earnings write-up, Ethernet Scale-Up Networking (ESUN) is kicking up dust in the market. The market is concerned because ESUN is proposing an Ethernet solution for scale-up with the October press release stating: “ESUN is a new workstream collaboration designed as an open technical forum to advance Ethernet in the rapidly growing scale-up domain for AI systems.” 

Last quarter, I pointed out that latency is a differentiator as UALink operates in the 100s of nanoseconds versus microseconds for Ethernet (as it stands today). Bridging this gap requires a leap in product design and successful deployment, and until that occurs, ESUN is structurally disadvantaged on intra-rack scale-up workloads. 

There is a time to market issue for UALink, yet in the meantime, PCIe remains a strong choice for fast, scale-up systems. PCIe is deployable right now for scale-up pods and CXL is also a strong choice for memory pool connectivity (Astera participates in all of this). 

The reality is that PCIe wins out for tighter, low-latency scale-up. What PCIe offers is device-level interconnects for highly synchronized GPU-to-GPU communication. The strength of PCIe will only become more evident (not less) as large training jobs require more GPUs to communicate. To contrast, Ethernet is a networking protocol designed for maximum reach – which are strengths that are quite distinct from intra-rack connectivity. 

Memory access between devices is another distinction where PCIe excels as Ethernet Scale-Up does not natively address memory-level integrations. This is an area the I/O Fund is watching closely for Astera, because as memory capacity scales and speeds increase, the attach rate increases for a vendor like Astera with more PCIe lanes, more signal conditioning and fabric complexity. Overall, the memory boom should increase Astera’s content across both PCIe and CXL deployments. 

Which leads us to Scorpio-X … 

Scorpio to Become Largest Product by Year End 

For a refresher on Astera’s products, please reference our previous analysis hereour previous analysis here 

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.” 

There was an inquiry on the call as to timing and magnitude: 

Ross Seymore, Deutsche Bank AG 

I guess, Mike, as my follow-up on — to the Scorpio family. I believe you said it crossed 15% of sales in 2025. So I just want to clarify if that was true. But perhaps more importantly, sort of bogeys as far as the growth rate this year. I believe in the past, you talked about it would cross over and become your biggest product line at some point this year. Is that still the case? Any updates on those sorts of timing and magnitude? 

Michael Tate, CFO 

Yes. So yes, so we originally set up for a 10% bogey, we did cross above 15% for 2025. And again, that's all just P-Series, X is for scale up is a much bigger, larger TAM for us and that we're starting to shift initial volumes in the first half, but the more material step up in the back half. So the commonization of those 2 will put us on a trajectory for it to be our biggest product line. But Aries and Taurus and Leo are all growing as well. So it's hard to know exactly when we cross over.  

But definitely at some point, it will. It will — it's going to drive very good revenue growth for us.” 

Astera Hints at New Customer Design Wins for NVLink 

As stated in our Top 15 report, PCIe remains relevant during the Vera Rubin transition due to from Nvidia’s “Extreme Co-Design.” As Rubin brings multiple compute, networking, and memory components together into a single, tightly integrated platform, the need for rapid, low-latency data movement within the server increases. 

This extends further with Nvidia’s move to PCIe Gen6 alongside expanded CXL support on its Vera CPU, up from PCIe Gen5 on Grace. CXL enables low-latency, high-bandwidth memory and cache sharing between CPUs, GPUs, and attached memory devices, reinforcing PCIe’s role at the heart of the system architecture. PCIe fabric switches are also expected to remain critical for backend GPU-to-GPU communication and for linking CPUs, NICs, and storage at scale. 

That’s a bit of context for why one of the most important parts of the call was not related to ESUN but rather it was confirmation that Nvidia is working with Astera on NVLink integrations. This is a lengthier quote so bear with me, but it ties together the full picture of why I believe Astera has a longer runway than the market is communicating. 

Blayne Curtis, Jefferies: 

Congrats on the results and congrats, Mike, on the new role. I just want to ask you, obviously, this $6.5 billion is a huge number. I might already know the answer, but I wanted to ask you about what seems like one of the biggest debates still is the acceptance of UA Link for these next-gen designs. You mentioned two lead customers mentioning it. I'm just kind of curious as people think about your UA Link switch opportunity, particularly at your largest customer versus the custom connectivity and then maybe them using NVLink? 

I'm kind of curious with this deal, is there any better visibility you can kind of think about, you know, that mix between hybrid boxes and native UA Link for these Azure lead customers? 

Jitendra Mohan, CEO: 

Thanks, Blayne. Maybe let me start and then Mike can chime in on the warrant itself. So, yes, clearly, AWS announced at re:Invent that the Trainium 4, which is slated to ramp in 2027, will support UA Link, which was a very positive endorsement of UA Link, as well as support for NVLink fusion. Subsequently, AMD has also announced that their MI 500 series will also support UA Link, again in 2027. So these are two very good public announcements in support of UA Link, and there are several other discussions that are ongoing. The UA Link ecosystem is coming. We've got great availability of IP, a lot of vendor announcements, and so on. 

And so we will be ready with our UA Link solution to intercept the ramp that happens in 2027. Now for NVLink fusion, this also represents a meaningful opportunity for us. And before we jump into what the opportunities are, I do want to call out the fact that both Amazon, the hyperscaler, as well as NVIDIA have chosen Astera Labs as a partner. And that's a very important statement in terms of the trust that they place in Astera Labs. So the opportunity itself is to take the native protocol that the XPU or the ASIC speaks and translate that into NVLink. 

This is a sophisticated function, and we have a solution that we will deploy to address this. And given the fact that the solution attaches to the XPU on a one-to-one basis, we anticipate the overall revenues to be in line with the switch opportunity where we might be selling a UA Link switch. So all in all, the exact mix of how much NVLink fusion would be deployed versus a native solution would be deployed remains to be seen. But for us, the opportunity is roughly the same for both. 

My Takeaway: 

My goal is to always simplify things for my Research Members as much as I possibly can, which is why I had described about 8 months ago that Astera Labs is the best of both worlds as the company participates in both GPU workloads and custom silicon workloads from Amazon. That was echoed again this evening, which is that Astera will do well regardless of whether hyperscalers choose UALink (open standard mainly for custom silicon) or NVLink (Nvidia’s native scale-up for GPUs) 

Financials 

Q1 Revenue Guided to Increase 7.7% QoQ, Slowest in Public History 

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. 

For the full year, Astera reported revenue of $852.5 million, up 115.1% YoY, ahead of estimates for 108.9% growth. While there was no specific guidance for 2026, current estimates for $1.18 billion in revenue, up 42.4% YoY, are likely to be revised higher in the coming days considering Q1 beat by more than $30 million.  

Soft Q1 Guide for Margins 

The one piece to nitpick would be Astera’s softer gross and operating margin guidance, though this comes with good reason – management stated that they will be accelerating R&D investments, including investments in the Scorpio X-Series roadmap, and opening a new design center in Israel to focus on next-gen scale-up fabric and R&D to address memory bottlenecks.  

For the Scorpio X-Series, management said the decision to expand the product roadmap stemmed from discussions and initial deployments with hyperscalers revealing more opportunities in the scale-up switching market, estimated to be $20 billion by 2030. This expanded X-Series roadmap will focus on new capabilities “including support for increased radix, platform-specific protocols, in-network computing, Hypercast technology, and optical connectivity.”  

Breaking this down, increased radix will help the X-Series support varying cluster sizes, from small to large-scale configurations, with support for custom interconnect protocols and tech enhancements to improve utilization and reduce GPU-to-GPU communication overhead. Including photonic switch-to-accelerator links is expected to help enable multi-rack deployments and facilitate scaling to thousands of GPUs.  

Management stated: “You know, as we spoke on the call, the TAM is much bigger than we originally expected just, you know, when we measured it, just twelve, eighteen months ago. So we are increasing our investments to pursue these opportunities. Last quarter in Q4, we did close the XScale acquisition, so now we have a full quarter in Q1. And then just recently in this quarter, we closed another AQUI hire.” 

To put it briefly, the major takeaway here is that R&D expenditures moving through 2026 may create a more persistent operating margin headwind; for example, R&D expenses rose 18.9% QoQ in Q4 to $93.8 million, or 34.7% of revenue, and could continue to outpace revenue growth as these investments unfold.  

Turning to margins: 

GAAP gross margin was 75.6% in Q4, up 1.6 points YoY but down 0.7 points QoQ. Adjusted gross margin was 75.7%. For Q1, management guided for some gross margin contraction, forecasting GAAP and adjusted gross margin to be 74%, down 0.9 points YoY and 1.6-1.7 points QoQ.  

GAAP operating margin showed slight sequential improvement in Q4, coming in at 24.7%, up 24.6 points YoY and 0.7 points QoQ, beating guidance for 22.2%. Adjusted operating margin was 40.2%, up 5.9 points YoY but down 1.5 points QoQ. 

The main critique was Q1’s operating margin guidance, as it currently points to a more pronounced contraction, likely due to the R&D ramp — GAAP operating expenses were guided to increase nearly 15% QoQ, or almost double the guided revenue growth rate, while adjusted operating expenses were guided to increase nearly 20% QoQ.   

Thus, Q1 GAAP operating margin was guided to be 19.8% at midpoint, down 4.9 points QoQ (but still up 12.7 points YoY), while adjusted operating margin was guided at 34.5% at midpoint, down 5.7 points QoQ and up only 0.8 points YoY.  

Q4 GAAP net margin was 16.6%, down 0.9 points YoY and 22.9 points QoQ, due to a $33.9 million income tax provision, whereas the two comparable quarters above witnessed income tax benefits of $14 million and $24.2 million respectively. Adjusted net margin was 38.7%, down 8.4 points YoY but up 0.4 points QoQ.  

For the full-year, Astera reported substantial GAAP margin expansion down the line. GAAP gross margin contracted 0.7 points YoY to 75.7%, though operating margin improved 49.6 points, from (29.3%) in 2024 to 20.3% in 2025. GAAP net margin improved 46.8 points to 25.7%.  

Adjusted margins also expanded nicely, but not to the same degree – while adjusted gross margin contracted 0.8 points to 75.8%, adjusted operating margin increased 9 points to 39.2%. Adjusted net margin for 2025 was 38.8%, expanding just 2.6 points.  

EPS Beat of 13.7%, Smallest on Record 

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%.    

For 2025, Astera reported GAAP EPS of $1.22, up from a ($0.64) loss in 2024, while adjusted EPS was $1.84, up 119% YoY. Astera did not provide a full year guide for 2026, though considering the possibility for increased R&D to weigh some on margins and the marginal Q1 beat versus estimates, revisions are likely to be minimal from the current $2.37 for 33.1% growth.  

Cash Flows and Balance Sheet 

Cash flow margins remained strong in Q4, while Astera’s balance sheet remained healthy with zero debt and cash increasing slightly. Accounts receivable showed strong growth while inventories also rose, indicating that growth may remain strong through the initial part of 2026.  

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 down 0.2 points QoQ. For the year, free cash flow was $281.8 million for a 33.1% margin, up 7.3 points YoY.  

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.  

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

Conclusion: 

The I/O Fund considers many factors when determining how to position correctly. Frankly, the charts are picking up on market doubts from ESUN, yet the product analysis points to the setup for a second wind. Where Astera continues to stand out is that its products are qualified for both custom silicon for UALink and Nvidia deployments via NVLink, providing rare relevance across dueling platforms.  

Perhaps most importantly, Ethernet Scale-Up doesn’t weaken this position, rather the debate leads to a reinforcement on why PCIe is very difficult to displace. In a fluid and competitive space like networking, Astera breadth of customers and platforms integrations offer rare defensibility.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in ALAB at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• The I/O Fund’s Top 15 Stocks for Q1 2026
• Reddit Q4: Unwavering Fundamentals; Change in User Reporting Metrics
• Bloom Q4: $20B Backlog, Guides for 58% Revenue Growth
• Coherent Fiscal Q2: Strong Visibility for Back-Half of 2026 and Beyond
• AMD Q4 Earnings: 60%+ Data Center Growth for 3-5 Years

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. 

Book-to-Bill at 4X offers Important Visibility 

Coherent’s management team went to great strides to offer visibility, which helped the price stabilize after hours. In particular, the comment their data center bookings have a bill-to-book ratio of 4X was helpful: “In Q2, we experienced another step function increase in our data center bookings, with a book-to-bill ratio that exceeded 4x, as customer demand continues to increase and customers place orders further out in time, which provides us with strong visibility for the coming quarters.” 

This was asked about in the Q&A with management emphasizing again the line of sight they currently have: “I was really pleased with the acceleration of our sequential growth rate, 14% sequential growth. And then we also saw, as I mentioned in the prepared remarks, over 4x book-to-bill ratio. So just seeing incredibly strong demand, and we’re seeing bookings go further out in time than we would have in the past, which is great for us for visibility.” 

Coherent also noted that some large customers are booking 2-3 years out with long-term agreements, which guarantee customers a defined level of supply while offering a stronger growth outlook than optical networking companies have seen in the past.  

Per management: “Then the third thing I would mention with respect to visibility is, number of long-term supply agreements that we’ve either signed with customers or in the process of signing, where, you know, the LTA will provide, a guarantee to our customers for a certain amount of, supply, and in exchange, they give us a guarantee on a certain amount of demand. There’s often, some sort of financial commitment from our customers, like investment for CapEx, et cetera. I would say all those things combined, the visibility of the business is, the best it’s ever been, which gives us just kind of great confidence in terms of the go-forward growth that we’re seeing.” 

Pros/Cons of Internal Sourcing versus External Sourcing 

Coherent’s slower growth compared to peers is due to sourcing the substrates and wafers internally rather than rely heavily on external suppliers. Although this results in higher margins over time, it results in a slower near-term ramp. This manufacturing strategy could ultimately pay off given Coherent can yield more at fixed costs.  

Inevitably this is discussed at length on earnings calls given it’s a competitive differentiation versus other optical networking peers. In the opening remarks, Coherent explained they are on track to increase internal capacity by the end of the year: “For example, we significantly increased our indium phosphide production capacity in Q2, and we are executing on track to our plan to double our internal indium phosphide production capacity by the fourth quarter of this calendar year.” 

During the Q&A, management described the advantages of sourcing internally in the following way: “Another way to look at it is, any time the kind of market price of Indium Phosphide goes up, it makes our internally sourced Indium Phosphide that much more valuable, right, in terms of a differential. And then, you know, I would say in terms of our own pricing, you know, we continue to see, you know, the ability to continue to optimize pricing. I think Sherry mentioned in her prepared remarks that, some of our gross margin improvement last quarter was based on pricing optimization. We continue to see opportunity to optimize pricing, especially in a environment where, where we’re, where demand is very strong.” 

Financials 

By Royston Roche 

Organic Revenue Growth of 22% 

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. 

The company’s CEO and President, James Anderson said in the earnings call, “In particular, we expect continued strong sequential revenue growth in both our March and June quarters, and we expect our fiscal '27 revenue growth rate to exceed our fiscal '26 growth rate. The key growth drivers that we see over the coming quarters are growth in both 800 gig and 1.6T transceivers, growth from the ramps of new products such as OCS and CPO solutions and ongoing exceptionally strong demand in our products for DCI and scale across.” 

Segments 

Data Center and Communications Segment Revenue Growth of 33% 

The company’s data center and communications segment revenue grew by 33% YoY and 11% QoQ to $1.21 billion. Revenue growth accelerated from 26% YoY and 7% QoQ growth in FQ1 driven by strong AI demand. 

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. 

Communications segment FQ2 revenue grew by 9% QoQ and 44% YoY driven by growth in data center interconnects products and in traditional telecom applications. Management expects the communications business to grow sequentially in FQ3 and FQ4. 

Industrial segment revenue was down (10%) YoY and down (3%) QoQ. On a pro forma basis, excluding the divested aerospace and defense business revenue grew by grew 4% QoQ and was flat YoY. Sequential growth in FQ2 was driven by industrial lasers and engineered materials product lines. Management expects the Industrial segment to be roughly flat sequentially in FQ3 on a pro forma basis. Looking ahead, they expect improving demand as they witnessed significant increase in orders in FQ2 from the semi-cap customers, which they expect to translate into sequential growth for the industrial business in the June quarter and the remainder of this calendar year.

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%. 
• Net income grew by 41.9% YoY to $146.7 million with a net profit margin of 8.7% compared to 7.2% in the same period last year. Adjusted net income grew by 34.2% YoY to $248.2 million with an adjusted net profit margin of 14.7% compared to 12.9% in the same period last year. 

Adjusted EPS grew by 36% YoY 

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%. Analysts expect FQ4 adjusted EPS to grow 43.2% YoY to $1.43 and 31.3% YoY to $1.52 in FQ1. 

Cash Flow and Balance Sheet 

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. Management expects capex to increase in the coming quarters to support strong expected demand in the data center and communications segments.  
• The company had debt of $3.35 billion and cash of $863.7 million at the end of the December quarter. While the debt is high, the company has taken steps to streamline its portfolio, with the $400 million sale of its Aerospace and Defense unit in early September, which it used to pay down its debt. Thereby, reducing the debt leverage ratio from 2.4x in the September 2024 quarter to the current 1.7x. The company further plans to pay down its debt by using the proceeds from the recently sold Munich product division, which should also reduce interest expenses and lower the debt leverage ratio.  

Conclusion: 

Data center revenue is accelerating with a 4X book-to-bill ratio and the 6-inch wafer supply is already at 80% of target capacity. In addition, optical circuit switches are moving now and co-packaged optics are on the way – two solutions that Coherent maintains they have significant IP compared to its competitors. 

Coherent hurries for nobody, and that discipline is evident even as the emerging 1.6T cycle is arriving earlier than expected and will be margin accretive. The company clearly has a plan given its pivot to 6-inch wafers; that plan happens to be more gradual than the market prefers to see. However, when strong visibility intersects with an in-line earnings report; strong visibility tends to win out.

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:

• AMD Q4 Earnings: 60%+ Data Center Growth for 3-5 Years
• Meta Q4 Earnings: A New Era Driven by AI Agents
• Palantir Q4: Highest Growth as Public Company; US Commercial to Accelerate
• Lumentum Q2: Capacity Constrained (and Loving It)

AMD’s data center segment revenue increased 39% year-over-year to a record $5.4 billion, led by Instinct GPUs and EPYC CPUs. Of this, $390 million was from MI308 sales to China, which means the DC segment reported closer to $5.0 billion in revenue. That number is worrisome to the market because it would mean QoQ growth of 15% for Q4, down from 34% QoQ growth in Q3. This would also translate to YoY growth of 29% – which is not to grab the Street’s attention given Broadcom is in the 100% range on ASICs and Nvidia is reporting about 12X higher on a quarterly basis and 50X higher revenue on an annual basis.

The guide also implies a deceleration from $10.3 billion this quarter to $9.8 billion in the upcoming quarter. Although this beats the Street expectations of $9.37 billion, it’s not the beat/raise tempo being set from other AI stocks right now.  

However, there were some bright spots on the call, primarily management stating to expect 60% growth in the data center over the next few years – including 2026. This hint is important as it communicates a strong acceleration into the second half of 2026 given we are starting out at data center growth of 39%, which is what we’ve been expecting.  

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.” 

I could have simply republished last quarter’s write-up except for this one important update that offers 3-5 year visibility; rare for any management team but especially AMD. 

Data Center to see 60%+ Growth for Next 3-5 Years 

Buried in the 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.” 

There have been rumors that AMD may be late in delivering the MI400s, however, that was dismissed today on the call as management reiterated, they are on time for H2:

Later, the CFO also confirmed margins would improve by Q4 based on the launch: “So all those tailwinds we're seeing, we continue to see in the next few quarters. And MI450 ramp, of course, in Q4, our gross margin will be driven largely by mix. And I think we'll give you more color when we get there. But overall, we feel really good about our gross margin progression this year.” 

There was more color provided on timing by CEO Lisa Su: 

“But as we get into the second half of the year, the MI450 is really an inflection point for us. So that revenue will start in the third quarter, but it will ramp significant volume in the fourth quarter as we get into 2027. So that gives you a little bit of sort of what the data center ramp looks like throughout the year.” 

Later she reiterated this again when asked for specific timing: “And our expectation is that we will be on track for our second half launch.” 

However, I do expect some analysts to be turned off by the company choosing to not breakout Instinct GPU revenue from EPYC revenue in the data center segment. That would imply GPU revenue is pretty low still. That is not too relevant to our thesis, yet if you see negative notes coming out, it’s due to lack of visibility in the current size of the AI accelerator business. 

The only visibility provided was a mention of “tens of billions” in revenue in 2027.

Aaron Rakers   Wells Fargo Securities 

Lisa, at your Analyst Day back in November, you seem to kind of endorse the high $20 billion AI revenue expectation that was out there on the Street for 2027. I know today you're reaffirming the path to strong double-digit growth. So I guess my question is, can you talk a little bit about what you've seen as far as customer engagements, how those might have expanded? I think you've alluded to in the past multiple multi-gigawatt opportunities. Just any — just double-click on what you've seen from the MI455 and Helios platform from a demand shaping perspective as we look into the back half of the year? 

Lisa Su   Chair, President & CEO 

Yes. Sure, Aaron. Thanks for the question. So first of all, I think the MI450 Series development is going extremely well. So we're very happy with the progress that we have. We're right on track for a second half launch and beginning of production. And as it relates to sort of the shape of the ramp and the customer engagement, I would say the customer engagements continue to proceed very well. We have obviously a very strong relationship with OpenAI, and we're planning that ramp starting in the second half of the year going into 2027. That is on track. We're also working closely with a number of other customers who are very interested in ramping MI450 quickly, just given the strength of the product, and we see that across both inference and training. And that is the opportunity that we see in front of us. So we feel very good about sort of the data center growth overall for us in 2026. And then certainly going into 2027, we've talked about tens of billions of dollars of data center AI revenue, and we feel very good about that. 

Financials: 

By Royston Roche 

Q4 Revenue Grew by 34% 

AMD’s Q4 revenue grew by 34.1% YoY and 11.1% QoQ to $10.27 billion, beating estimates by 6.2%. Revenue growth was primarily driven by continued growth in the Data Center segment from both server and data center AI business, as well as a return to YoY growth in the Embedded segment. 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. It is primarily driven by growth in the Data Center and Client and Gaming segments and modest growth in the Embedded segment. Although this beats the Street expectations of $9.37 billion, it’s not the beat/raise tempo being set from other AI stocks right now.   

Full year 2025 revenue grew by 34.3% YoY to $34.6 billion. Looking ahead, analysts expect revenue to grow 33.2% YoY to $46.1 billion in 2026 and accelerate to 37.9% YoY to $63.6 billion in 2027. 

Q4 Data Center Revenue Grew by 39% YoY 

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. In server, adoption of fifth gen EPYC CPUs accelerated in the quarter, accounting for more than half of the total server revenue. AMD had record server CPU sales to both cloud and enterprise customers in the quarter and exited the year with record share. 

In cloud, hyperscaler demand was very strong as North American customers expanded deployments. EPYC-powered public cloud offerings grew significantly in the quarter with AWS, Google and others launching more than 230 new AMD instances. In the enterprise, AMD is witnessing a meaningful shift in EPYC adoption, driven by leadership performance, expanded platform availability, broad software enablement, and increased go-to-market programs. Looking ahead, management expects server CPU demand to be very strong as hyperscalers are expanding their infrastructure to meet growing demand for cloud services and AI while enterprises are modernizing their data centers. 

The company delivered record Instinct GPU revenue in the fourth quarter, led by the ramp of MI 350 Series shipments. In addition to the partnership entered in October with OpenAI to deploy 6 gigawatts of Instinct GPUs, AMD is in active discussions with other customers on at-scale multiyear deployments starting with Helios and MI450 later this year. AMD expanded the ROCm ecosystem in the fourth quarter, enabling customers to deploy Instinct faster and with higher performance across a broader range of workloads. The company remains on track to launch its MI500 chips in 2027 and expects to deliver a major increase in AI performance.

Client and Gaming Segment revenue grew by 37% 

Q4 client and gaming segment revenue grew by 37% YoY and down (3%) QoQ to $3.94 billion. Client segment revenue grew by 34% YoY and 13% QoQ to $3.1 billion. Management highlighted the strong demand for Ryzen processors for laptops and PCs, which have been gaining market share against Intel. 

In gaming, revenue grew by 50% YoY and down (35%) QoQ to $843 million. Management said in the Q4 earnings call, “Semi-custom sales increased year-over-year and declined sequentially as expected. For 2026, we expect semi-custom SoC annual revenue to decline by a significant double-digit percentage as we enter the seventh year of what has been a very strong console cycle.” 

Embedded revenue returned to growth in Q4. It grew by 3% YoY and up 11% QoQ to $950 million, up from the YoY decline of (8%) and up 4% QoQ in Q3 2025. 

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. Adjusted gross margin was 57%, and it benefitted from the $360 million previously written down MI308 inventory reserves. Excluding the inventory reserve release and MI308 revenue from China, gross margin would have been 55%, up 100 basis points YoY and QoQ, driven by favorable product mix. Management has guided 55% adjusted gross margin in Q1 
• Q4 operating income grew by 101% YoY and 38% QoQ to $1.75 billion. Operating margin improved by 600 basis points YoY and 300 basis points QoQ to 17%. Adjusted operating income grew by 41% YoY and 28% QoQ to $2.85 billion. 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. 
• Net income was up 213% YoY to $1.5 billion or 15% of revenue, up 900 basis points YoY. Adjusted net income was up 42% YoY to $2.5 billion or 25% of revenue, up 200 basis points YoY. 

Adjusted EPS grew by 40% 

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 27.6% YoY to $1.22 and 185.8% YoY to $1.37 in Q2 2026. 

Cash Flow and Balance Sheet 

The company’s cash flows are growing primarily driven by higher revenue and profits.   

• 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. 
• 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. 
• The company had 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.  
• Inventories rose 8% QoQ to $7.92 billion to support the strong future AI demand.

Conclusion: 

There was no big reveal in AMD’s earnings report, yet interesting enough, that has been the case for years and yet the stock is starting to see movement as AMD outpaced Nvidia’s returns last year by 3X. In my most recent Top 15 AI Stocks report, I highlighted how the element of surprise can work in AMD’s favor. Most importantly, AMD’s management team remains among the most conservative in the space, which makes it notable that we’re already hearing expectations for roughly 60% growth in the data center segment for 3-5 years. That’s a nice clue for what may be on the horizon.

Royston Roche, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in AMD at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• Meta Q4 Earnings: A New Era Driven by AI Agents
• Palantir Q4: Highest Growth as Public Company; US Commercial to Accelerate
• Lumentum Q2: Capacity Constrained (and Loving It)
• The I/O Fund’s Top 15 Stocks for Q1 2026

Lumentum 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.  

To learn more about upcoming networking shifts, read our Top 15 report here.our Top 15 report here. 

Increasing Capacity for H2 2026 

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.” 

200G/1.6T Ramping Faster than Expected 

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: 

Ruben Roy   Stifel, Nicolaus & Company, Incorporated 

Great. Just a very quick follow-up. Has anything changed with the way you guys are thinking about 800 gig versus [ 1.6T ] module mix this year one way or the other? Is it accelerated towards 1.60T for any reason in terms of volumes from a single customer or multiple customers? Or is it relatively unchanged from how you're thinking about it 90 days go? 

Michael E. Hurlston   President, CEO & Director 

Yes. 1.6T is definitely stronger than we felt than we felt 90 days ago. So 1.6T is definitely accelerating. Our 800-gig volume actually is doing better than we would have expected. So an 800-gig that what you're seeing right now from us is an acceleration in revenue on our 800-gig shipments. But in the market, to your question, Ruben, 1.6T is definitely going better. We have exposure to a couple of customers, a couple of large customers on 1.6T, and we've been surprised by how quickly they're trying to push us to deliver and their forecast to us relative to the different SKUs that we're being asked to deploy. 

OCS Starting to Move the Needle, CPOs Incoming for 2027 

Optical circuit switches are starting to move the needle. By the end of fiscal 2027, management now expects roughly $400 million in revenue, up from the $100 million originally guided—representing an incremental $300 million upside to prior expectations. 

Management stated the following on the call: 

“If you remember, you and I discussed last time, we believed our calendar Q4 would be around $100 million. It looks like it will be quite a bit higher than that, although we're not breaking up that $400 million between the two quarters. So we — it's a broad-based — there's multiple customers making up that backlog. We've talked about shipping to three customers, and that continues — but those customers are increasing their demands rather significantly. And thus, the demand on us has gone up quite appreciably. So we feel pretty good about that. I think as we enter calendar year 2027, it should go up from there in terms of what we see in our backlog and in terms of our revenue.” 

Co-packed optics are expected to contribute early 2027 with stronger contribution by year-end 2027: “An industry pivot is underway to bypass the scaling limits of copper. By late calendar 2027, we would expect our first scale-out CPO shipments, replacing longer copper connections.” 

Financials 

Revenue Growth to Accelerate to 89% in Q3 

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.  

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.  

Segment Breakdown 

As we had noted in our prior analysis, Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth, Lumentum changed its reportable segments in fiscal Q1, dropping Cloud & Networking and Industrial Tech and transitioning to Components and Systems.  

For a brief recap, Components include laser chips, laser subassemblies, line subsystems and wavelength management subsystems, while Systems includes full stand-alone products such as optical transceivers, optical circuit switches and industrial lasers.  

Lumentum expects Components to be the cornerstone for revenue growth and profitability while Systems will scale rapidly with transceivers, OCS and other high-performance solutions.  

Components Revenue Accelerates Slightly 

Components revenue was $443.7 million in Q2, up 68.3% YoY and 17% QoQ. YoY growth accelerated 3.4 points, though 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. Ultra-high-power lasers for CPO continued to grow, with Lumentum outlining a broader ramp in the second half of calendar 2026, aligning with Nvidia’s Spectrum-X switch roadmap with its Vera Rubin platform.  

Lumentum also noted that Q2 saw an eighth consecutive quarter of sequential growth for narrow linewidth lasers for data center interconnect (DCI) applications, while pump lasers for scale-across and sub-sea applications reached a record. 

Systems Revenue up 43% QoQ, New Record for Transceivers 

Systems revenue saw much stronger sequential growth than Components as it is coming off a much smaller base, with cloud transceivers likely to be the main driver as optical circuit switching (OCS) is still very early in its ramp.  

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.  

Lumentum did note that OCS shipments exceeded a $10 million quarterly run rate in the quarter, while manufacturing readiness is proceeding ahead of schedule as the company prepares to begin fulfilling its >$400 million OCS backlog later in 2026. 

Margins Expand Substantially, Operating Margin up 22.5 Points YoY 

Complementing the strong revenue growth is substantial margin expansion, with Lumentum showing GAAP operating margin expand 22.5 points YoY to nearly crack into double digit territory. Lumentum’s cost profile also shows that operating margin expansion will continue as costs rise at a much slower pace than revenue growth. 

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%.  

However, operating leverage was quite prevalent and visible in the quarter as opex rose just 0.6% QoQ and 16.3% YoY. This, combined with gross margin expansion, drove significant expansion in operating margins on a YoY and QoQ basis.  

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. 

To note, Lumentum is well ahead of its target financial model, which called for >20% adjusted operating margin and 39-42% adjusted gross margin at a $750 million/quarter. Lumentum is already above both metrics on $665 million/quarter base.  

Net margins followed, with GAAP net margin expanding 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%. An important takeaway here is that this AI-driven growth is driving strong earnings leverage for Lumentum, as adjusted net margin was single-digits just three quarters ago.  

Adjusted EPS Beats by 18%, Q3 Guided 40% Above Estimates 

As just mentioned, this AI growth and margin expansion is driving visible earnings leverage for Lumentum. Not only did adjusted EPS beat estimates by 18% in Q2, but Lumentum’s Q3 guide was more than 40% above estimates, signaling >290% growth will be maintained for another quarter.  

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.  

Combined, Q3 and Q4 adjusted EPS came in $0.91 ahead of estimates, and if Q4 is to see a similar ~$0.60 to $0.70 upward revision to match Q3’s beat, full-year adjusted EPS estimates could move to around the $7.50 range, up from the current $5.92 prior to the report. 

Cash Flows and Balance Sheet 

Cash and equivalents increased slightly to $1.16 billion while debt was $3.29 billion.   

Inventories were $570.4 million, up more than 7% QoQ, while accounts receivable surged nearly 23% QoQ (again) to $376.8 million, supporting the upcoming product ramps over the coming quarters.  

Conclusion:

AI networking is entering a new phase, one where silicon photonics plays a much larger role alongside system-level shifts such as optical circuit switches and co-packaged optics.  

The goal is to continually optimize for AI workloads rather than rely on traditional networking as ever-expanding AI clusters seek speed, lower latency and reduced power consumption. Pluggable optics work well enough today but are quickly becoming a limiting factor as power and port density doesn’t scale well with traditional optics. This dynamic is pushing hyperscalers to rethink network architectures, and positions Lumentum well, as its solutions help future-proof AI networks for the years ahead.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in LITE at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• Meta Q4 Earnings: A New Era Driven by AI Agents
• Palantir Q4: Highest Growth as Public Company; US Commercial to Accelerate
• The I/O Fund’s Top 15 Stocks for Q1 2026
• Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth

Memory is typically a cyclical industry that is lower margin and lumpy, yet it is seeing a newfound resurgence from AI that is strong enough to transform commoditized hardware into a secular trend as the AI economy is built out. AI servers use more DRAM and NAND than traditional servers, relying heavily on high-bandwidth memory (HBM) for training and inference.   

We first touched upon the rising importance of the memory market in AI GPUs in the summer of 2023 within our AMD and Lam Research analyses, and provided a closer look in November 2023 in the analysis, 2024 Trend: Memory and PC Rebound. We also dove further into HBM’s growth opportunities in December 2023 with our Micron deep dive, Micron: AI Offers a Multifaceted Secular Growth Tailwind. 

As is stands, the AI-driven demand for memory (especially HBM and high-performance DRAM) is still in the early stages of a multiyear growth cycle. The company’s CEO and Chairman, Sanjay Mehrotra, also mentioned in the September earnings call, “Memory is very much at the heart of this AI revolution. This means a tremendous opportunity for memory and certainly a tremendous opportunity for HBM."  

The HBM market is projected to reach $35 billion this year, doubling YoY, with Micron’s September results confirming that the market was well on track to be over $30 billion as of Q3. Looking ahead, the shift to HBM4 with Nvidia’s Rubin architecture and AMD’s MI400 series will represent another important growth lever come 2026 as HBM content per GPU and per rack surges, paving the way for HBM to potentially triple again by as early as 2028. 

Not only is HBM a focal point due to its rising importance and thus increasing content per GPU, but other memory products are quickly coming to the forefront, notably low-power DDR5 memory (LPDDR5X) and data center solid state drives (SSDs). 

Below, we look at the memory products front and center of this AI-driven cycle, structural drivers behind rising AI memory demand such as AI inference, supply and inventory constraints driving prices rapidly higher, long-term growth outlooks and more. 

Overview: DRAM and NAND’s Role in AI 

Demand for high-capacity memory is driven by generative AI and LLMs, which both require significant amounts of computing power and substantial amounts of DRAM to meet elevated performance requirements. Within DRAM, demand is focused more specifically around high-bandwidth memory (HBM) and double-data rate 5 DRAM (DDR5) for its increasing content in AI accelerators — SK Hynix’s head of DRAM marketing Park Myung-soo has explained in the past that “an AI server requires 500-gigabyte (GB) or larger high bandwidth memory (HBM) chips and at least 2-terabyte (TB) DDR5 chips.” Now, we’re seeing nearly that amount of HBM being put on a single GPU rather than an 8-GPU system.  

High bandwidth memory (HBM) offers higher bandwidth, capacity, performance, and lower power by vertically stacking up to twelve DRAM memory chips to shorten how far data has to travel, while also allowing for smaller form factors. Stacked memory chips are connected through something called “through silicon vias” or TSVs.  

HBM is now mission-critical, especially for inference, as increasing bandwidth and capacity per HBM generation paves the way for significant leaps in throughput with each GPU generation. Currently, the leading AI accelerators from Nvidia and AMD utilize HBM3e, an enhanced HBM3, while the next-gen Rubin and MI400 architectures are set to bring HBM4 mainstream next year – more on this below. 

DDR5 DRAM, or double data rate 5, is aimed to double bandwidth and data transfer speeds at a lower latency and power consumption than its predecessor, DDR4. DDR5 memory chips can be mounted on circuit boards to create memory modules, for use in servers or PCs. DDR5’s increased bandwidth allows for faster processing for memory-intensive applications, such as generative AI and training LLMs. Memory giant SK Hynix saw high-capacity DDR5 (>128GB) revenue more than double QoQ in Q3. 

Demand for a low-power DDR5 variant, LPDDR5X, is rising sharply due to its role in Nvidia’s Grace and Vera CPUs, as LPDDR5x is delivering up to 5X better throughput, a more than 35% increase in memory bandwidth with up to 77% lower power consumption. This combination can improve system power efficiency (performance per watt) by up to 10%, per Micron.  

Emerging with Nvidia’s GB300 racks (and soon Rubin racks) are SOCAMM modules, which combine LPDDR5X with a Compression Attached Memory Module (CAMM). SOCAMM modules can deliver up to 2.5X higher bandwidth with lower power consumption and a smaller footprint versus traditional RDIMMs (registered dual in-line memory modules).  

Not to be forgotten is memory’s second half, NAND, as it also plays a vital yet less visible role in AI, as its high-capacity, reliable storage is increasingly important in meeting growing inference demands. NAND’s importance is primarily concentrated to NAND-flash based data center solid-state drives (SSD). 

SSDs can boast superior performance in speed, latency, energy efficiency and reliability compared to hard disk drives (HDD). The high-speed read capabilities of SSDs help process vast datasets in training large LLMs and multi-modal models, as well as store model checkpoints. For inference, SSDs are typically used to store ‘hot’ data, or data needing to be accessed frequently, making them crucial for inference workloads.  

HBM3e, HBM4 and the Need for Increased Memory Bandwidth 

HBM3e is the primary version of HBM shipping currently, supporting both Nvidia’s H200 and Blackwell architectures, as well as AMD’s Instinct MI350 series and Google’s TPU v7 Ironwood. HBM4, the next generation, is expected to support Nvidia’s upcoming Rubin platform later in 2026, along with AMD’s Instinct MI400 series.  

The reason that AI accelerators are quickly upgrading to the next generation of HBM is because HBM capacity and bandwidth are consistently increasing, which, when combined with increasing capacity per chip, translates directly to massive leaps in throughput, or tokens processed per second. This means that newer chip generations, such as the shift from Nvidia’s Hopper generation to Blackwell, are exponentially more performant on LLM inference workloads.  

For example, HBM2e, used on Nvidia’s H100, delivered a modest 3.6 Gb/s data rate (speed of data transfer), leading to 461 GB/s of bandwidth per HBM cube. With HBM3, data rates improved to 6.4 Gb/s and stack heights moved from 12 to 16, thus boosting bandwidth by more than 75% to 819 GB/s.  

Source: Rambus 

With HBM3e, data rates increased substantially to 9.6 Gb/s, boosting bandwidth to 1.23TB/s, or nearly 3X that of HBM2e. Translating this to Nvidia’s H200 meant that it could deliver 1.4X to 2X faster LLM inference versus the H100 as bandwidth per chip rose from 3TB/s to 4.8TB/s and HBM capacity rose 1.76X from 80GB to 141GB.  

Source: Nvidia 

With HBM4, the main upgrade is a doubling of interface bits from 1,024 to 2,048, or the number of data bits that can be transferred simultaneously between the memory chip and the GPU. This means that even at the JEDEC standard of 8 Gb/s data rate, a modest decline from HBM3e, bandwidth per HBM4 stack rises to more than 2 TB/s, a 2.5X boost from HBM3 and a ~66% increase from HBM3e. However, Micron claims that its HBM4 boosts data rate to 11 Gb/s, delivering 40% higher bandwidth at 2.8 TB/s per stack, along with 20% better power efficiency and 60% better performance versus HBM3e.  

While these increases may not seem significant when looking simply at upgrades per HBM generation, looking at the exponential increases in bandwidth and inference performance per GPU generation shows a better picture. 

Nvidia’s 8-GPU HGX H100 system delivered a mere 24 TB/s of aggregate memory bandwidth, yet the HGX B200 system boosted that ~2.6X to 62 TB/s with the shift to HBM3e and more HBM3e content (more on this below). 

The scale-up architecture of Nvidia’s GB200 and GB300 NVL72 brought a massive boost to aggregate memory, with both rack-scale solutions offering 576 TB/s, or 24X more than the HGX H100 systems. Nvidia says the GB200 can offer throughput of up to 116 tokens/s on GPT-MoE-1.8T model, a 30X improvement on real-time LLM inference versus the HGX H100, with performance gains also aided by improvements in NVLink and CX8 network interface cards. 

Source: Nvidia 

Nvidia’s upcoming Vera Rubin architecture will boost aggregate memory bandwidth by as much as 8X from here over the next two years. 

Nvidia’s upcoming Vera Rubin NVL144 is expected to take aggregate memory bandwidth to 1.4 PB/s, and 1.7 PB/s with the CPX platform. This is the equivalent of 1,400 TB/s to 1,700 TB/s, or a ~2.4X to ~3X increase versus the GB200/GB300 racks. Jensen Huang claims that the NVL144 system bandwidth is “the entire data usage of the Internet in one second.” 

With the NVL576, aggregate memory bandwidth will continue to surge, with the rack boasting 4.6 PB/s, or 4,600 TB/s of bandwidth. This is another roughly 3X boost to the NVL144, and compared to the GB200/GB300, a massive 8X increase in just two years.  

HBM’s Longer-Term Tailwind: Capacity per Chip Surging 

HBM capacity per chip continues to rise with each new generation of GPU, and this is a primary contributing factor behind the surging aggregate memory bandwidth discussed above, paving the way for accelerated throughput gains and inference performance.  For example, we’ve seen a ~3.5x increase in HBM content in short fashion on Nvidia’s GPUs within about three years’ time frame:  

• The H100 featured 80GB of HBM2e content per chip. This chip began shipping in Q4 2022 and ramped in early 2023.  
• The H200 featured 141GB of HBM3e content per chip, 1.76x higher than its predecessor.  
• The B200 features 180GB of HBM3e content, more than double the H100 and a 28% increase versus the H200. In an 8-GPU server configuration, the B200 boasted 1.44TB of HBM content.   
• The B300 boasts 288GB of HBM3e content, a 60% increase versus the B200 and over 3.5x more than the H100. In an 8-server configuration, the B300 has 2.3TB of HBM content. This chip is beginning to ship now in Q3-Q4 2025.  
• The upcoming Rubin chip will remain at 288GB, but transition to HBM4 for more bandwidth. 

Putting in context Nvidia’s rack-scale solutions, the GB200 and GB300 NVL72, shows just how rapidly HBM content is increasing. The GB200 supports up to 13.4TB of HBM content, while the GB300 supports up to 21.7TB of HBM, nearly 34X higher than the 640GB of HBM content in the 8-GPU DGX H100 servers.  

AMD is also showing surging memory requirements, to the tune of 3.5X across two main generations:   

• The Instinct MI250 featured 128GB of HBM2e memory.  
• The MI350X featured 288GB of HBM3e memory, a 125% increase versus the MI250 and on par with Nvidia’s Blackwell Ultra.   
• The MI400 series is expected to feature 432GB of HBM4 memory, a 50% increase versus the MI350X and the Blackwell Ultra. In the Helios rack configuration slated for 2026, the MI400 will boast 31.1TB of HBM content, 1.5x more than the GB300 NVL72.   

Packing more HBM per chip is also not exclusive to GPUs, with Google’s TPUs notably seeing a 6X jump in HBM capacity over one generation (one year) and a 12X increase in two generations: 

• Google’s TPU v5e, released to general availability in 2023, featured 16GB of HBM capacity. 
• TPU v6e (Trillium), released in 2024, doubled HBM capacity per chip to 32GB. 
• TPU v7 (Ironwood), released this year, boosted HBM capacity by 6X over Trillium to 192GB per chip, or 12X growth from v5e. 

This surge is expected to continue through 2027 as HBM4 and then HBM4e come online – it has been estimated that in a 20-high configuration could pack 80GB of memory per HBM chip, up from 36GB per 12-high HBM3e cube today. Assuming similar usage of eight cubes, this could take memory per GPU up from 288GB in Nvidia’s B300 to 640GB in future chips.   

HBM Market Doubling in 2025, Expected to Triple Again by as Early as 2028 

The HBM market is forecast to double this year to approximately $35 billion, up from less than $18 billion in 2024, with growth driven by increasing content per GPU such as that with Blackwell and Blackwell Ultra as well as from capacity constraints. As of Q3, HBM is now likely above a $30 billion annualized run rate, supported by comments from Micron last quarter that its HBM revenue grew to almost $2 billion; with HBM share expected to nearly match its DRAM share (of 22% in Q2 to 25.7% in calendar Q3), this would imply the HBM market is likely in the mid-$8 billion range, or around $32-33 billion annualized.  HBM’s share of DRAM revenue is also rising sharply, expected to rise ten points this year, from 18% in 2024 to 28% in 2025, with more growth ahead through 2030.  

Through 2026 and 2027, the outlook for HBM remains fairly positive, with SK , SK Hynix, Samsung and Micron already selling out of HBM3e and HBM4 capacity through the end of 2026. This underscores the robust demand environment stemming from AI accelerators, with Micron seeing HBM bit shipments outpacing DRAM bit growth, but also may limit revenue upside as prices have been contracted over the next four quarters.  

On pricing, HBM4 is expected to carry a significant premium to HBM3e, currently used for Nvidia’s Grace Blackwell chips. Analysts from UBS had estimated that HBM4’s price premium could be as much as 30%, though reports of Samsung’s discussions over HBM4 supply with Nvidia dwarfed that – Samsung was said to be targeting price parity with SK Hynix on HBM4 around $500, up ~50% from the mid-$300s for HBM3e. These price increases will support strong growth as HBM4 volumes ramp. 

Looking forward, industry analysts project the HBM market to reach $98 billion to $100 billion by 2030, representing a 31.5% CAGR from 2024’s $18 billion, outpacing DRAM’s growth by 3X, which is expected to rise at an 11.7% CAGR to $194 billion. As a result, HBM’s share of DRAM revenue is expected to surpass 50%.  

However, in its Q1 report, Micron said it now expects the HBM TAM to reach $100 billion as early as 2028, two years sooner than its prior forecast. This would represent a ~42% CAGR from $35 billion, or more than 10 points faster than the base case forecasts.  

HBM’s Challenges: No Tail-End to Shipments, Supplier Shifts  

The reason that HBM can be such a challenging market is two-fold – supplier qualifications can (and do) change rather quickly between generations for SK Hynix, Samsung and Micron, and the winner is oftentimes determined by time-to-market, or whichever company can hit mass production first.  

Micron executives explained that HBM is unlike standard memory products, and that they do not expect a long tail in these products, meaning that once the next generation comes online and ramps (HBM3 to HBM3e, HBM3e to HBM4, etc), demand for the old generation dissipates quickly. This in turn means that whichever suppliers can either qualify first and reach mass production first have an advantage when it comes to revenue and even margins. For example, SK Hynix was the main supplier of HBM for Nvidia’s H100, yet Micron was the main supplier for the H200.  

For HBM4, SK Hynix said in early September that it had finished development of HBM4 and was ready for mass production, while Micron also announced that month that it had begun shipping HBM4 samples to customers. On the other hand, Samsung just finished development of HBM4 in early December and began shipping samples to Nvidia. However, SK Hynix is reportedly delaying the start of mass production from Q2 2026 to Q3 2026 to better align with Rubin’s ramp.  

Rising Demand for LPDDR5X and the DDR5 Profitability Dilemma 

Demand for  LPDDR5X is rising sharply due to its role in Nvidia’s Grace and Vera CPUs, as LPDDR5x is delivering up to 5X better throughput, a more than 35% increase in memory bandwidth with up to 77% lower power consumption versus typical DDR5.  

Low power is critical with Nvidia’s GB racks as well as Rubin, as power consumption has been surging per rack, and already pushing the upper boundaries of what current data center infrastructure can handle. Current builds, such as Vantage’s upcoming 1.4 GW campus in Texas for Oracle, are only designed for ultra-high density racks up to 250kW, meaning these new facilities could quickly be phased out and require new infrastructure to accommodate increasingly power hungry racks. 

Nvidia’s GB300 and Rubin platforms will use a purpose-built SOCAMM module optimized for AI servers, which combines LPDDR5X with a Compression Attached Memory Module (CAMM), aimed at maximizing performance and reducing power consumption. This is currently provided by Micron, which reported 50% QoQ in LPDDR revenue to a new record last quarter.  

When comparing to smartphones, the usual destination for LPDDR memory, the content demands for AI servers are profoundly large:  

“Indeed, each Grace CPU in today's platform is equipped with 480 GB of LPDDR5X memory (a premium smartphone uses 16 GB of LPDDR5X), but this is going to at least double with Vera CPUs, possibly straining LPDDR5X supply.” 

So not only do you have Nvidia’s Blackwell and Blackwell Ultra lines ramping, with those consuming 30X LPDDR5X memory as a typical smartphone, but that gap is poised to widen tremendously later next year as Rubin ramps, with the Vera CPU expected to contain 1.5TB of LPDDR5X, more than 3X the Grace CPU and as much as almost 94 smartphones.  

Keep in mind that the 480GB for the Grace CPU and the 1.5TB for Vera CPU are per chip, per chip, meaning that the GB200 NVL72 rack featuring 36 CPUs will consume 17.28TB per rack. per rack. For the Rubin NVL144, with the same 36 CPU count, LPDDR5X content would surge to 54TB per rack, and in the NVL576, with 144 CPUs, content quadruples to 218TB per rack218TB per rack. That is the equivalent of 13,625 premium smartphones.  

Nvidia’s demand needs are expected to place substantial upward pressure on prices, as Counterpoint Research believes it now is an LPDDR “customer on the scale of a major smartphone maker — a seismic shift for the supply chain which can’t easily absorb this scale of demand.”  

Global DRAM Market Surges 31% QoQ in Q3, Q4 Pricing to Remain Strong 

You would be hard pressed to find another segment of the AI data center industry posting growth to this degree on a sequential basis. Data from TrendForce estimates that the global DRAM market recorded growth of 30.9% QoQ in calendar Q3 to $41.4 billion. In dollar terms, this represented QoQ growth of ~$9.7 billion, or nearly as large of a QoQ jump as Nvidia reported in its most recent quarter.  This growth was driven by “significant increases in conventional DRAM contract prices, higher bit shipments, and growing HBM volumes.”  

For a supplier breakdown, SK Hynix’s revenue grew 12.4% QoQ to $13.75 billion, fueled by seasonal price increases and significant bit shipment growth. Samsung also reported similar significant growth in bit shipments, with revenue up 30.4% QoQ to $13.50 billion. Micron followed with a substantial 53.2% QoQ increase to $10.65 billion, per TrendForce (note that this is for calendar Q3 which does not align with Micron’s fiscal year calendar).  

As of November, TrendForce estimates that DRAM contract prices will accelerate into Q4, predicting conventional DRAM contract prices will surge by another 45% to 50% QoQ, while total contract prices (which includes HBM) will increase by 50% to 55% QoQ – this is a substantial uplift from projections for 18-23% QoQ growth in Q4 at the end of October.  

Contributing to strong pricing is DDR5 DRAM, where prices rapidly skyrocketed – from late September to early November, prices have as much as quadrupled, with impacts felt most on consumer products. Samsung also reportedly just boosted DDR5 prices by 100%, citing no stock left.  

However, revenue growth in Q4 will likely be lower than pricing as bit shipments are projected to decline sequentially due to rapid inventory depletion. DRAM supplier inventory levels are projected to range between two to four weeks, a major crunch from 5.5 weeks on average last quarter and more than 15 weeks at the start of the year.  

Turning to 2025 as a whole, HBM is expected to be a primary growth contributor for the DRAM industry. Current projections have DRAM revenue rising ~35% YoY, or $32 billion, to $127 billion in 2025, meaning HBM is contributing more than half of that dollar growth at ~$17 billion. This is also marking a rapid recovery from 2023’s trough of $52 billion, with the $127 billion projection representing two-year growth of 148%.  

For 2026, it is these tailwinds above, along with tight supply, that can continue to drive strong growth in the DRAM market moving through the year, especially as HBM4 begins to ramp initially with Nvidia’s Rubin platform along with AMD’s MI400 platform.  

Inference is Creating a Secular Tailwind for Data Center NVMe SSDs  

Data center solid state drives (SSDs), such as those based on NAND flash memory, are an often overlooked but equally important memory component when it comes to AI training and inference. This is because data center SSDs offer higher read-write speeds critical for accessing and transferring data rapidly, along with higher performance and energy efficiency, making them vital for larger-scale AI training and inference workloads.  

NVMe (Non-Volatile Memory Express) is a protocol designed specifically for NAND-flash based SSDs that optimizes performance by reducing latency and increasing data transfer speeds by utilizing the PCIe bus. This helps provide the high throughput and fast data transfer speeds necessary for AI workloads – NVMe SSDs can increase performance by more than 2X versus SATA SSDs. 

There are five main types of NAND flash used in SSDs, delineated by the number of bits of data that can be stored per cell: 

• SLC (single-level cell): Stores one bit of data per cell, meaning data can be retrieved faster. SLC offers the best performance and highest endurance, though it is typically the most expensive.  
• MLC (multi-level cell): Stores two bits of data per cell, allowing for a higher data density or higher capacity, though this comes at the expense of performance and endurance. MLC is typically found in consumer NAND products. 
• TLC (triple-level cell): Stores three bits per cell, increasing density and capacity and reducing cost, but also increasing chance for error. 
• QLC (quad-level cell): Stores four bits per cell, providing significant storage capacity (4X that of SLC) and lower costs, making QLC suitable for large-capacity solid-state drives. Meta has made the case for QLC SSDs in data center applications due to the higher density and improved power efficiency versus TLCs at a price that allows for significant scaling, though it says price is not yet competitive enough for broader deployments. 
• PLC (penta-level cell): The next evolution of NAND flash that stores five bits per cell, aiming for significant high density storage but also facing high error rates.    

AI training and inference are two main long-term drivers for SSD demand, which is projected to rise ~6X from 2024 to 2030, from 181 exabytes (EB, or equal to 181,000,000 TB) to 1,078 EB, under McKinsey’s base case scenario. Training demand projected to rise at a 62% CAGR to from 7 EB in 2024 to 127 EB by 2030. On the flipside, demand from AI inference is expected to grow at a 105% CAGR from 6 EB to 447 EB by 2030, giving inference a 41% share of demand versus less than 12% for training. 

This is not only driven by development of more LLMs, but also the increasing size and complexity of frontier models, where training data sets and context windows for inference are getting increasingly large.  

For example, EpochAI estimates that training data set sizes are rising 3.7X per year on average, or nearly doubling every six months. There are some models that are scaling much quicker. For example, Meta’s Llama2-70B from 2023 was trained on 2 trillion tokens, while Llama3-70B, from 2024, was trained on 15 trillion tokens, a 7.5X increase.  Multi-modal models, those integrating audio, video, image or more, are also likely to require significantly more SSD storage, with McKinsey estimating in the hundreds of TBs depending on the mix of data needing to be stored.  

Source: EpochAI 

Looking at tensor parallelism from a memory perspective also shows why the ability to distribute workloads across tens to thousands of GPUs is such a game-changer for AI training and inference. After accounting for memory required to store model parameters and for the activation buffer, a single AMD MI300X GPU can handle a max request of ~6,500 tokens on Llama-70B, per TensorWave. However, when you distribute model parameters across 8 GPUs along with the same buffer, that 8-GPU server could now handle a max request of 523,000 tokens, an ~80X increase, with gains that only compound as server size and memory increase.  

The increasing size and complexity of models also ties directly to a major pain point when it comes to inference: “As models grow in complexity and require longer contexts, their memory footprint expands beyond what a single GPU can handle. This results in inefficiencies where GPUs are memory-starved, causing significant bottlenecks in AI token generation.” 

To put this GPU memory bottleneck in real-life application for inference, AI inference acceleration startup WEKA states that when it tested Llama-3.1 70B with no optimizations, a 100K token prompt took 24 seconds to pre-fill into the model before any output could be generated, but “extending GPU memory to ultra-fast storage [NVMe SSDs] can dramatically improve token processing efficiency.” When configuring an Nvidia DGX H100 server with an 8-node exabyte-scale NVMe SSD pod, WEKA says its “tests demonstrated a staggering 41x reduction in prefill time on LLaMA3.1-70B, dropping from 23.97 seconds to just 0.58 seconds,” significantly improving model efficiency with zero optimizations – simply from adding SSDs to extend GPU memory. This pre-fill time would feel near instantaneous for an end user versus a 24 second delay.  

Other long-term growth vectors include increased adoption of retrieval augmented generation (RAG), which “assembles companies’ own data into vectorized databases, which models then refer to, improving the accuracy and specificity of outputs.” RAG would then require two forms of storage – active storage of useful data, and vector database storage to organize that active data to be accessible by LLMs.  

Source: McKinsey 

A more rapid uptake of RAG or faster multi-modal model adoption could push data center SSD demand up to a 42% CAGR through 2030, reaching 1,490 EB or ~8X 2024’s demand, while slower uptake could see demand rise at a 25% CAGR to 702 EB, or ~4X 2024’s demand.  

Data Center SSD Revenue Up 28% QoQ in Q3 

Similar to DRAM, data center SSD shipments and prices were strong in Q3, driven by hyperscaler demand for AI infrastructure and general-purpose servers. Revenue from the top five companies – Samsung, SK, Micron, Kioxia and SanDisk – rose ~28% QoQ to a new record at $6.54 billion, per TrendForce. Notably, this was broad-based strength, with growth at the five firms all ranging between 26-30% QoQ.   

For Q4, there are a few dynamics in play that are likely to keep prices and thus revenue growth strong. For example, supplier inventories are expected to have fallen sharply, from 10-15 weeks in early Q3 to just 7-10 weeks at the start of Q4, which was said to be ‘below healthy levels’, with enterprise SSD supply growth substantially lagging demand. SanDisk says that its storage-focused SSD is “growing in demand with 2 hyperscaler qualifications underway and a third hyperscaler along with a major storage OEM planned for calendar year '26.” 

In November, TLC and QLC SSDs reportedly experienced strong price increases, with 1 TB TLC SSDs seeing sharp increases and the “most significant shortage due to persistent enterprise SSD demand.” 512 GB TLCs were estimated to see the most significant price hikes at ~65% MoM, while the QLC supply chain tightened and forced prices higher.  

Additionally, TrendForce points out that these inventory and demand dynamics mean “supply shortages in 2026 are becoming increasingly apparent,” providing an additional lever for SSD prices to rise through next year and support more revenue growth as long as inventories and bit shipments do not hinder that.  

Can the Memory Boom Last Through 2028? 

There have been rising discussions regarding the strength of this current memory boom, and whether it can stretch through 2027 or even 2028, as reports from Korea and analysts from Morgan Stanley now estimate. The industry currently has the necessary ingredients for a sustained upcycle: strong demand, supply shortages combined with lean inventories, and strong pricing trends. A multi-year supercycle would likely require persistent supply shortages driving strong pricing power, stemming from elevated demand. such as strong HBM and LPDDR5X content growth with next-gen GPU racks, and strong inference-led tailwinds for SSD growth.  

There are signs emerging that support such a view. Micron said in November that it is seeing “much more supply-demand tightness than we expected” in September and expects this tightness “to continue beyond 2026.” However, perhaps the most important comment from Micron came from Q1’s call this past week, with management saying that “in the medium term, we are only able to meet about 50% to 2/3 of our demand from several key customers.” SK Hynix also believes it will be “difficult to resolve the supply shortage by the first half of 2027.” More specifically on NAND, SanDisk says that demand “continued to outpace our supply, a dynamic we expect to persist through the end of calendar year '26 and beyond.”  

Samsung and SK Hynix have both been rather straightforward about wanting to avoid oversupply, as this could cut the current cycle short and eat into profitability quickly. Samsung executives have said that they will “minimize the risk of oversupply through a capital expenditure strategy that balances customer demand and pricing," and instead of rapidly expanding production, they will focus on profitability.” SK Hynix is on a similar page, though reports have suggested it could boost 1c DRAM production by ~8X by 2026, from 20K units per month to 160K, in order to meet rising SOCAMM and GDDR7 demand.  

The profitability point ties into capacity allocations and exhibits why supply remains tight. For example, HBM3e and DDR5 share production capacity, and through the first part of 2025, HBM3e “commanded a price premium more than four times that of DDR5.” However, with the recent surge in DDR5, profitability is now on track to surpass HBM3e by Q1, meaning suppliers may shift HBM3e capacity to DDR5 to boost profits. Samsung is already planning this shift from HBM3e to DDR5, with the expectation that it will shift ~80K wafers per month, while Micron is shelving its consumer DRAM and SSD unit, Crucial, to focus on HBM, DDR5 and enterprise SSDs.  

Competitive risks aside, evidence of the size of this boom will be visible within revenue growth trajectories and margins. But perhaps the most important question for this cycle is, can the combination of tight supply, low inventories, sharply rising prices and strong (and rising) demand drive margins and earnings power to surpass 2018 levels in a sustainable way? 

Currently, Micron’s revenue estimates and revisions give two primary takeaways into the duration and size of the cycle – analysts are more bullish about the boom lasting into 2028, though they are essentially completely divided on the overall strength of it, with revisions showing a massive range between low and high end forecasts. 

Above shows revenue estimates heading into Micron’s fiscal Q1 report on December 17, with fiscal 2027 and fiscal 2028 both seeing estimates revised 41-43% higher since July. FY27 estimates had moved from $48 billion to $68 billion, while FY28 moved from $51 billion to $72 billion. A majority of the upward revisions have come since September, aligning with surging DRAM prices.  

However, Micron gave a blowout Q2 guide, forecasting revenue of $18.7 billion at midpoint, more than 31% above consensus for $14.23 billion and representing growth of 37.1% QoQ and 132.2% YoY. This has pushed estimates even higher – FY26 and FY27 already see revisions ~$16 billion higher to $74.1 billion (+98% YoY), and $84.3 billion (+15%), while FY28 rose $11 billion and points to flat growth.

Analysts remain essentially completed divided on the potential strength of the cycle, with the gap between the low and high end of revenue estimates doubling from fiscal 2026 to fiscal 2027 and 2028.  

For example, estimates for fiscal 2026 range from $53 billion on the low end to $82 billion on the high end, or a $29 billion range. For fiscal 2027, the low end falls to $46 billion, potentially on pricing peaking much sooner than expected, while the high end rises to $106 billion, a $59 billion range. Fiscal 2028 also sees a $61 billion range between the high and low end of $53 billion to $114.5 billion. 

Source: Seeking Alpha 

More impressively, Micron is showing that gross and operating margins have already surpassed the 2018 peaks, and commentary for expansion through the year suggests some potential upside to already strong earnings estimates.   

For example, Micron’s TTM gross and operating margins, below, have rapidly recovered from 2023’s trough and already pushed past prior cyclical peaks (2010/2015), at 45.6% and 33% respectively as of fiscal Q1 (ending Aug). On a quarterly view, Micron’s FQ1 margins were 56% and 45% respectively, up 11.3 and 12.7 points QoQ. For comparison, SK Hynix reported operating margin at nearly 47% in Q3, up more than five points QoQ. 

Compare this to the 2018 cycle, where DRAM prices tripled over the course of six to eight quarters. Micron’s gross and operating margins peaked at 60% and 50% respectively, and Q2 was guided to far surpass that at 67% and 58.7%, respectively. Again for comparison, SK Hynix’s operating margin peaked at 57% on a quarterly basis in Q3 2018, more than ten points higher than current margins.  

Potential earnings power is where this boom gets interesting, especially for Micron, given the wide range for revenue estimates and the rapid ascent in margins to above >65%/>55%. Considering Micron’s management explained that they “would expect gross margins to expand beyond fiscal Q2” though at a more gradual pace than the last few quarters, it is reasonable to assume upside towards 70-72% and potentially 61-63% on operating margin, assuming similar fall-through. Supporting this would be evidence of strong AI-driven product demand in HBM and LPDDR5X (non-existent factors in the prior 2018 cycle) and strong DRAM pricing. 

Assuming Micron ramps into this margin profile of ~70%/61% by year-end and maintains that through fiscal 2027 (Aug ’27) on tight supply dynamics and demand growth, rough back-of napkin math would place FY27 GAAP EPS at $40.35, or ~7.6% above consensus for $37.50 (although it should be noted that this was $20.77 prior to earnings, or an ~80.5% raise now). 

If FY27 revenue moves to the upper end of the estimated range, or ~$105.8 billion, driven by factors such as strong HBM demand from next-gen platforms and strong LPDDR5X content growth, earnings power could be even stronger. Assuming the same peak margin profile of 70/61/54%, FY27 GAAP EPS could reach $50, or ~33% higher than consensus, though such a scenario could be challenging to execute. 

However, it would be remiss to cover memory without discussing the cyclicality of the industry and risks to the ‘supercycle’ narrative. Some of the main factors that could end this cycle include potential oversupply from capacity additions, or price reverting lower after its current L-shaped trajectory.  

While 16Gb DDR4 and DDR5 prices have seen an “unprecedented spot price rally” to record levels, time and time again DRAM prices have always reverted lower, although cycle timing can differ. The swiftness of the current price rally has already outpaced 2018’s rise, though the duration of the price rally has not nearly been long enough to see when or where it could peak. The next-gen DDR6 is not expected to reach the mass market until 2027, suggesting there is ample runway for DDR5 pricing to remain strong through 2026.   

A more hidden risk to the thesis emerges from consumer electronics. Although Micron has exited its consumer memory business and the focus for the trio of Micron, SK Hynix and Samsung remains squarely on AI, consumer electronics (smartphones and PCs) are still strong drivers of DRAM and NAND demand. For example, some analysts have placed consumer electronics at ~37% of DRAM and ~56% of NAND demand.   

The surging DRAM prices are placing upwards pressure on bill-of-materials content for PCs and smartphones, with Lenovo, Dell, HP, Asus and others already hiking PC prices as a result, estimated at around 15-20%. These dynamics could lead to inventory buildup in consumer electronics markets, or potentially some degree of demand erosion, both major headwinds for pricing strength moving through 2026 and 2027. 

While inventory rebuild and oversupply have previously ended past booms, manufacturers are aiming to preserve strong profitability and avoid flooding the market to keep this cycle intact. However, there can be no assurance that these fears will remain on the back burner come 2027. 

Conclusion 

While this may be a lot to unpack, the primary takeaway here is that the memory market is seeing strong, structural tailwinds from rising HBM and LPDDR5X content in GPUs and SSD use in AI applications. Some of the primary companies located at the heart of this trend include Micron, SK Hynix and Samsung as the primary HBM manufacturers; for enterprise SSDs, the market leaders include Micron, Samsung, SK, Kioxia and SanDisk.  

To help narrow down on this trend, we plan to dive deeper into one of a leading Memory stock to our Discovery tier members the first week of January. 

Subscribe to Discovery and get the Top 10 Emerging Tech Watchlist delivered monthly. Our incoming Top 10 list will be published January 2nd with many new names including a lesser-known memory stock. Current Pro and Advanced Members: To subscribe to Discovery with 30% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY30

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 AI Stocks for Q4 2025
• Meta: Growth is Quietly Benefitting from AI, Though Margin Risks Weigh Heavy
• Broadcom FQ4 Earnings: $73B AI Backlog with Visibility; $162B Consolidated Backlog
• Coherent: Indium Phosphide Capacity to Double, Data Center to Reaccelerate to 10% QoQ