Category: AI Stocks

AI Spending To Exceed A Quarter Trillion Next Year

Posted on by io-fund

This article was originally published on Forbes on ForbesForbes on Nov 14, 2024, 05:29pm EST

Big Tech’s AI spending continues to accelerate at a blistering pace, with the four giants well on track to spend upwards of a quarter trillion dollars predominantly towards AI infrastructure next year.

Though there have recently been concerns about the durability of this AI spending from Big Tech and others downstream, these fears have been assuaged, with management teams stepping out to highlight AI revenue streams approaching and surpassing $10 billion with demand still outpacing capacity.

Below, I take a look at the growth in AI spending from Big Tech this year and yet, as it quickly approaches the quarter-trillion mark, and next week, I’ll discuss exactly what this means for the market’s biggest beneficiary.

Sign up for I/O Fund's free newsletter with gains of up to 2600% because of Nvidia's epic run – Click hereClick hereClick here

AI Capex Accelerating

Big Tech’s AI-fueled capital expenditures serve as a barometer for the broader AI industry, as Microsoft, Meta, Alphabet and Amazon are among the first to recognize multi-billion dollar revenue streams from AI and generative AI offerings. The four are also leading the charge by pouring billions each quarter towards AI infrastructure, signaling that they are still attempting to catch up to AI demand and invest more aggressively in AI come 2025.

To better understand the trajectory of AI spending, let’s take a step back to 2023, where the rapid ascent of ChatGPT at the beginning of the year set the stage for AI to quickly step into the spotlight.

In the first half of 2023, Big Tech spent ~$74 billion on capex. Through Q3, that sum had moved up to ~$109 billion.

In the first half of 2024, Big Tech spent nearly $104 billion, a 47% YoY increase. Through Q3, that sum had surged to $170 billion, up 56% YoY.

So far in 2024, Big Tech has spent nearly $171 billion on capex, predominantly for AI infrastructure, up 56% from 2023.

Big Tech Capex

Source: I/O Fund

To understand why these four are accelerating spending this year and laying the groundwork for even higher spend in 2025, consider this: why is Big Tech spending billions on AI infrastructure globally? Why is Big Tech procuring GPUs en masse or building out custom silicon to deliver AI services in the cloud to millions of enterprise customers?

The answer is three-fold:

1) AI is expected to have a multi-trillion dollar economic impact globally, with a recent estimate from IDC placing AI’s cumulative potential impact through 2030 at $20 trillion. The mobile economy, which sprouted a handful of the trillion-dollar tech behemoths of today, added approximately $5.7 trillion to the economy in 2023. Big Tech’s leaders are well aware of how critical it is to capture and capitalize on an opportunity of this magnitude, and will not miss it.

2) Developing larger models and doubling model sizes requires massive computing power that only Big Tech can afford to develop, meaning a majority of genAI progress is likely to be made primarily in the hyperscalers’ clouds.

3) Big Tech is already realizing AI-related gains, with three of the four saying AI revenue is at least in the mutli-billion dollar range. With millions to billions of users for products to either enhance with AI integrations or target with AI features in subscriptions, the long-term revenue opportunity could dwarf some of their leading revenue streams of today.

I had said in May this year that Big Tech “will likely commit upwards of $200 billion, maybe even $210 billion, combined in capex this year, predominantly for AI infrastructure – from data center construction and expansion, to GPU procurement and custom silicon efforts and more.”

However, that figure is already likely too small, given the pace of acceleration seen in Q3 and commentary for Q4 and full-year spending. Combined, Big Tech spent $64.9 billion in Q3, up 11% QoQ. This increase was driven primarily by Amazon, which boosted capex by ~$5 billion sequentially.

Big Tech Quarterly Capex

Big Tech spent $64.9 billion on capex in Q3, up 11% QoQ and accelerating to 68% YoY. Source: I/O Fund

Microsoft and Amazon combined for $42.6 billion in capex in the third quarter, with Alphabet maintaining its ~$13 billion/quarter rate and Meta beginning to accelerate its spending.

Amazon signaled in Q3 that it was expecting to spend $75 billion on capex this year, with Meta tightening and raising its capex guide to $38 billion to $40 billion – alone, the two are expecting to spend nearly $115 billion in 2024. To meet that target, combined capex from the duo will need to be nearly $35 billion.

Alphabet is expecting Q4’s capex to be relatively in-line with Q3’s as it maintains its pace for ~$50 billion in full year spend, while Microsoft did not lay out a concrete picture for capex this year. Assuming spend is flat sequentially for Microsoft, the two would be spending ~$33 billion in Q4.

Putting this all together, Big Tech could spend another $70 billion in Q4, overwhelmingly for AI infrastructure, putting full year capex at ~$240 billion, or nearly 15% higher than the level they were tracking for at the start of the year.

The I/O Fund will spell out what this means for the biggest beneficiary of this trend in next week’s newsletter – make sure you don’t miss it.

Come 2025, this AI-driven capex surge is set to stay, with executives foreseeing lasting AI demand and a need to still invest to capture growth and meet demand.

Executives Signal AI Demand is Lasting, Requiring More Spend

I want to reiterate this quote from May’s newsletter, Big Tech Q1 Earnings: AI Capex Increases As AI-Related Gains Continue, as it continues to remain relevant for investors: it is no surprise that Big Tech is boosting spending by more than 50% versus 2023 “given positive outlooks on AI’s potential to drive revenue growth in the billions and how demand continues to outstrip GPU supply.”

This theme was evident across Big Tech’s Q3 earnings calls. Listen to what executives had to say:

Microsoft: Microsoft spent close to ~$10 billion this most recent quarter on GPU and CPU servers, primarily to meet cloud demand, with management signaling that “demand continues to be higher than our available capacity.”

CFO Amy Hood explained that Microsoft expects capex “to increase on a sequential basis, given our cloud and AI demand signal,” as they aim to stay aligned with demand signals. Microsoft also “announced new cloud and AI infrastructure investments in Brazil, Italy, Mexico, and Sweden as we expand our capacity in line with our long-term demand signals.”

Hood further clarified that Microsoft has confidence that as they “get a good influx of supply across the second half of the year, particularly on the AI side that we'll be better able to do some supply-demand matching and hence, while we're talking about acceleration [in Azure] in the back half.”

Amazon: Amazon CEO Andy Jassy said that AWS has “more demand that we could fulfill if we had even more capacity today,” and that “pretty much everyone today has less capacity than they have demand for, and it's really primarily chips that are the area where companies could use more supply.” He explained that AWS is growing rapidly in AI, but he believes “the rate of growth there has a chance to improve over time as we have bigger and bigger capacity.”

What Jassy is saying is that AWS and Microsoft are not the only supply-constrained firms, with Alphabet, Oracle, and others all struggling to meet demand because they cannot purchase enough GPUs or deploy enough custom accelerators alongside GPUs to meet demand.

Jassy also dropped a big clue on long-term demand and AWS’ need for rapidly increasing AI investments. He said that he thinks AI is at an “earlier stage [and] more fluid and dynamic than our non-AI part of AWS,” and customers not “showing up for 30,000 chips in a day. They're planning in advance. So we have very significant demand signals giving us an idea about how much we need.”

It’s interesting that this comment comes as Amazon has significantly ramped capex over the past two quarters, from $14.6 billion in Q1 to $22.6 billion in Q3. Jassy’s comment implies that AWS is seeing much larger demand than what they were expecting at the beginning of the year, hence the need to spend much more on AI infrastructure, from data centers to servers to GPUs to custom silicon.

Alphabet: The Search giant was a bit more obscure on AI demand in the cloud, but executives signaled spending to increase in 2025. CFO Anat Ashkenazi said that realizing growth opportunities and innovating in AI “requires global reach, which we have through our products and platforms, as well as continued meaningful capital investment.” Ashkenazi explained that Alphabet thinks that “into 2025, we do see an increase coming in 2025, and we will provide more color on that on the Q4 call, likely not the same percent step-up that we saw between '23 and '24, but additional increase.”

Meta: Though Meta is positioned primarily in advertising as opposed to the cloud, executives still signaled long term opportunities and a need to continually invest in AI. CEO Mark Zuckerberg said that it is “clear that there are a lot of new opportunities to use new AI advances to accelerate our core business that should have strong ROI over the next few years,” while Meta’s “AI investments continue to require serious infrastructure, and I expect to continue investing significantly there too.”

CFO Susan Li added that Meta is “growing our infrastructure investments significantly this year, and we expect significant growth again in 2025.” For Q4, she clarified that Meta foresees the large QoQ jump in part from “increases in server spend and to a lesser extent data center capex” due to delivery and cash recognition dynamics.

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock entries and exits. We offer trade alerts plus an automated hedging signal. The I/O Fund team is one of the only audited portfolios available to individual investors. Learn more here.Learn more hereLearn more here.

AI Revenue Streams Emerging

AI revenue streams are emerging as Big Tech continues to spend prolifically on AI, with Microsoft among the leaders as it sees AI revenue soon to be double digits.

Microsoft: CEO Satya Nadella pointed out that “monetization from these [AI] investments continues to grow, and we're excited that only 2.5 years in, our AI business is on track to surpass $10 billion of annual revenue run rate in Q2. This will be the fastest business in our history to reach this milestone.”

At a closer look, AI contributed 12 points to Azure’s growth in the recent quarter, implying that Azure’s AI run rate has already surpassed $6 billion, with other gains coming from Microsoft’s product suite, with Power Platform seeing 4x YoY growth to 600,000+ users utilizing AI capabilities and 70% of the Fortune 500 using Microsoft 365 Copilot.

Azure AI Quarterly Run Rate

Azure’s AI run rate is estimated to have already surpassed $6 billion. Source: I/O Fund

To read more about how AI could drive the $100 billion in revenue for Microsoft by 2027, read more here: Microsoft – AI Will Help Drive $100 Billion In Revenue By 2027.Microsoft – AI Will Help Drive $100 Billion In Revenue By 2027.

Amazon: Amazon did not provide an exact number for AI revenue, but said that “AWS's AI business is a multibillion-dollar revenue run rate business that continues to grow at a triple-digit year-over-year percentage ,and is growing more than 3 times faster at this stage of its evolution as AWS itself grew.”

For comparison, it took AWS ~2 years to scale from ~$500 million in revenue in 2010 to over $2 billion in revenue in 2012, and then another 3 years to grow to nearly $8 billion. To have AI growing at triple this rate in the multi-billion dollar level already speaks volumes about the magnitude of the AI opportunity ahead and the demand that exists that still can’t be met in the market today.

Alphabet: Alphabet did not provide a new update for AI revenue, with the latest update from Q2 noting that “AI infrastructure and generative AI solutions for Cloud customers have already generated billions in revenues and are being used by more than 2 million developers.”

Management provided a few additional points about the swift uptake of AI across its products, saying that “Gemini API calls have grown nearly 40x in a 6-month period,” while AI Overview in Search “will now reach more than 1 billion users on a monthly basis.”

Meta: Unlike the other three, Meta’s path to monetizing AI in the billion-dollar scale is less clear, as AI’s primary role in operations is driving better ROI and conversions for advertisers, thus driving advertising revenue higher.

Management said that “Meta AI now has more than 500 million monthly active improvements to our AI driven feed and video recommendations have led to an 8% increase in time spent on Facebook and a 6% increase on Instagram this year alone. More than a million advertisers used our Gen AI tools to create more than 15 million ads in the last month and we estimate that businesses using image generation are seeing a 7% increase in conversions and we believe that there's a lot more upside here.” What’s not as clear is the direct impact to revenue growth stemming from these AI-fueled increases, but management has faith in the longer-term of driving strong ROI from AI investments.

Conclusion

Big Tech’s AI spending is only set to surge through the end of 2024 and into 2025, with management teams reiterating the need to invest more to meet demand and build out AI infrastructure. Microsoft leads the pack with AI on the cusp of surpassing a $10 billion run rate, while Amazon and Alphabet see AI revenue in the billions.

In next week’s free newsletter, the I/O Fund will discuss what this surge in spending to more than a quarter trillion will mean for one of AI’s largest beneficiaries. The I/O Fund is also closely tracking the next sectors to benefit from AI, regularly sharing our research on the biggest growth opportunities. We also share our buy and sell plans with real-time alerts for our premium members. Learn more here.

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:

Astera Labs: Hypergrowth AI Networking Stock

Posted on by io-fund

Astera Labs is a stock that simply cannot be ignored despite having a fraction of the market cap compared to much larger AI hardware players.

The company offers a few key products that are enabling larger and faster AI clusters. For data center AI accelerators, the company was first to offer PCIe 5 switches and retimers. Growth will likely continue due to next-generation PCIe 6 products with higher average sales prices that will be released in 2025 and ramp in 2026.

There are additional growth opportunities, including a software architecture that allows hyperscalers to monitor their data center infrastructure and increase utilization rates; the software helps to incentivize hyperscalers to use Astera Labs hardware products. The company also offers ethernet smart cable modules and is introducing a new CXL product for CXL-enabled CPUs next year.

We are interested in Astera Labs for the increased average sales prices that are expected to persist at least through 2025-2026 due to the Aries products and upcoming Scorpio products. Astera Labs is a highly technical company, yet this quote clearly communicates why the growth trajectory can sustain:

“One is generally speaking with each new generation of a protocol like PCIe going from Gen 5 to Gen 6. There is an ASP uplift. That's number one. Number two, of course, we were hinting at the Scorpio product line, which because of the value it delivers to customers is at a higher ASP, as you can imagine. So overall, if you look at the design wins we have today, the dollar content per GPU goes up, that's one way to look at it based on what we've shared before […] So overall, if you look at sort of the increasing speed, additional product line as well as the fact that the internally developed platforms, AI accelerated database platforms, they are starting to gain more and more traction. So when you look at all of them, on an average, our content is on the up.”

Investment Thesis:

  • Current Growth Driver: Increased average sales prices are being driven by CPUs, GPUs and ASICs all moving to the new PCIe 5.0 standards. Arista Labs’ Aries Retimers and PCIe 5.0 components is driving the current growth, and the company is unchallenged in this new generation of PCIe, which came to market for AI accelerators only recently with Nvidia’s H200s.
  • Catalyst: Scorpio is a new product that is expected to expand the TAM to more than $12B by 2028: Astera Labs is releasing a PCIe Gen 6 fabric switch custom designed for AI data flows with high performance per watt compared to incumbents. Rather than building a large switch, the company built a smaller device that is more efficient for high-speed signals. With Scorpio, Astera Labs is defending its dominance in PCIe5 by doubling the bandwidth at lower power requirements than the 5th generation of PCIe.
  • The market demand for Astera Labs is healthy, as a major supplier to Nvidia’s PCIe-enabled GPUs (note: most of Nvidia’s GPUs use their in-house NVLink). Yet, Arista also supplies other AI accelerator platforms for Big Tech, and Arista is the only provider for PCIe5. In addition to hardware, ALAB offers Big Tech data centers software-defined architecture called COSMOS that allows for performance monitoring. There is some favorable vendor lock-in dynamics with COSMOS as the many different product lines can be optimized and monitored with the software.
  • Financials on Astera Labs are impressive as the company is hypergrowth at 206% growth last quarter and 153% guided next quarter. The free cash flow margin is at 41%. The company is not GAAP profitable due to being a recent IPO with outsized stock-based compensation. However, there is a path to GAAP profitability, which is detailed in the financials section below.

Overview of Astera’s Products:

Broadly speaking, Astera’s products are seeing increased relevance as AI clusters grow to support hundred-billion and trillion parameter models. The company’s hardware and software increase AI server performance and productivity for the current generations and future generations of AI accelerators.

Transition to PCIe5 and PCIe5 Retimers are Driving Astera’s Growth:

The Aries products offer PCIe5 interfaces that GPUs and AI accelerators (like custom silicon) use to connect components including ethernet networking. Compared to the previous generation, PCIe5 is twice as fast with data transfer rates that reach up to 128 GBs/s on multi-directional bandwidth in each lane. By increasing the data transfer for each lane, it allows more lanes to become available to help leverage the power of the GPU or AI accelerator.

PCIe5 Retimers are chips that boost signals across high-speed components and are seeing increased demand, starting with Nvidia’s H200s, and also for application-specific chips. Specifically, Aries Retimers and smart cable modules allow hyperscalers to connect multiple racks together with up to 7 meters of copper cables. Aries can also go up to 50 meters with optical fiber. According to a presentation at Nvidia’s GTC event, this is 3X the standard reach defined in PCIe specs.

Despite PCIe5 being out since 2019, it was Nvidia’s H200s released in 2024 that were the first data center GPUs to use PCIe5. What’s interesting is that Astera is said to have captured 95% of the XPU market, which refers to application-specific chips that are specific on a product level. Per an analyst on the earnings call: “Majority of the XPU shipments are still going to be I think Gen 5 based where your market share is still somewhere in the range of, I think, like 95%.”

Management also stated: “The upside to the guidance was driven largely from Aries' revenue, both for the third-party GPUs, but also as well with the strong ramps on new platforms on the internally developed AI accelerators. And we're seeing that across multiple hyperscaler customers, so it's not just one. So the upside was largely driven by that Aries revenue.”

Notably, Aries devices are used to interconnect AI accelerators with CPUs and networking, yet are also used for backend networking between GPUs for larger clusters. Astera supplies the HGX H100 systems with PCIe-based GPUs with up to 25 retimer chips per HGX system.

Scorpio PCIe6 Custom-Built for AI is a Catalyst:

Scorpio is where the excitement is building for continued growth next year with management stating it will “exceed 10% of revenues in 2025” with “good momentum going into 2026.” This is due to the PCI Express switches being custom built for AI purposes, whereas in the past, PCIe was built for storage and then retrofitted for AI purposes.

PCIe 6 doubles the bandwidth from the 5th generation, with up to 256 GB/s of bandwidth per lane, which will require faster supporting components, such as the retimers that Astera Labs offers. The demo from GTC showed the Scorpio fabric switches (name released in October) delivering twice the bandwidth with less power at 11W instead of the 13W from the PCIe 5 interfaces.

There are two Scorpio fabric switches. The Scorpio P-Series is a small chip that connects the CPU, GPU, NIC and NVMe storage. Rather than building a large switch, the company built a smaller device that is more efficient for high-speed signals to help feed GPUs with data. The fewer ports and smaller switch decrease complexity in a bid to compete against Broadcom with twice the lane count.

The X-Series is for back-end networking in GPU-to-GPU configurations, and will offer a higher port count. Astera is essentially building something similar to Nvidia’s NVSwitch with the X-Series, but for PCIe-enabled GPUs: “And this one, like Mike noted, it's a greenfield use case, meaning if you keep Nvidia and NV Switch aside, everyone else is starting to build configurations that are obviously going to need some kind of a switching functionality, which is what we are addressing with our X Series device.” The X-Series improves efficiency for ever-increasing AI cluster sizes. The majority of AI clusters are in the tens of thousands GPUs, but are expected to go to the hundreds of thousands (already has with X and some other Big Tech companies), and will see AI clusters with millions of GPUs over the next couple of years.

Here is another quote from the management team as to why the X-Series Scorpio switch fabric is a big opportunity for their company:

“X series will have a bigger TAM. The TAM today is nearly zero. It's not very commonly used outside of the Nvidia ecosystem. We do expect many hyperscalers to start deploying this, starting with the X family and the designs for which that we have. And we are able to do that because of the architecture that we have. Because of our software-defined architecture, we can customize many parts of the X-Series to cater to the specific requirements of the hyperscalers both on the side of performance, the exact configuration that they require in count and so on and also the diagnostics framework that they require to monitor their infrastructure. So over time, we do expect X-Series to become larger [than the P-Series].”

COSMOS Software:

In order of importance, COSMOS Software ranks higher than some of Astera’s hardware as it offers performance monitoring for data center infrastructure. This is especially important for Big Tech companies concerned with utilization rates for expensive GPU systems.

The adoption of the software stack to monitor for performance is expected to increase with the Scorpio X-Series:

“Where the Scorpio family sits, we have access to a lot more diagnostic information. And we can couple that with the information that we are collecting from our other families deployed such as Aries and even Taurus to provide a holistic view of the AI infrastructure to the data center operators. So both from the hardware side, the kind of the purpose-built nature of these devices as well as the software stack that comes with it is a big differentiator for us.”

Taurus Ethernet Smart Cable Modules:

Astera Lab offers Ethernet smart cable modules which help to alleviate bandwidth issues with 100-gig per lane connectivity over copper cables including AEC. The company recently released 400-gig Ethernet SCMs, which help to stabilize the network. When thinking of ALAB’s investment thesis, Ethernet is not top-of-mind given the sheer size and lead we see from Broadcom, Nvidia’s Spectrum, and Arista Networks.

Right now, the maximum bandwidth supported by PCIe 5.0 is 400Gbps per port. By using 106Gbps PAM4 SerDes, ASICs can be tuned to support 100, 200 and 400 Gbps port speeds. To work around this, and to achieve 800Gbps, larger chip makers are building NICs directly into the accelerator. According to The Register, the 800Gbps ports built into accelerators may reduce bottlenecks before PCIe 6.0 arrives on the market. The larger Ethernet players are moving quickly on this, and we will need to keep an eye on Astera Labs to determine if the company’s Taurus product can remain competitive.

Leo Compute Express Link (CXL):

Leo is slated to impact revenue in 2025 when data center platforms plan to utilize CXL technology for memory bandwidth and capacity bottlenecks. Next year, CXL-capable CPUs will become broadly available. We’ve covered in the past how CXL is a new server architecture that “dynamically assigns memory resources between servers.” The result is boosted memory bandwidth and also at a lower cost than adding more CPUs. Partially-disaggregated racks are expected to deploy in 2024-2025 with separate compute, memory and I/O racks with the interconnect being CXL.

Per Astera’s management team: “In the past, this was done by adding additional CPUs into the server box to provide for more memory channels. But what we have demonstrated is that by using Leo you're not only able to get the higher performance by the added memory. But from an overall TCO standpoint, it's significantly less than adding additional CPUs. “

Financials: Strong Revenue Growth, Expanding Margins

Astera Labs reported solid top-line and bottom-line beats in the recent Q3 results. The company reported record revenue of $113.1 million, up 47% QoQ and 206% YoY, beating estimates by 16.1%. The adjusted operating margin expanded to 32.4% from 2% in the same period last year, which was helped by strong operating leverage. The adjusted EPS of $0.23 beat estimates by 35%.

Revenue:

The company is one of the fastest-growing tech companies and is emerging as a rising star in the AI data center networking space. The company’s Q3 revenue grew by 206.2% YoY and 47% QoQ to $113.1 million.

Jitendra Mohan, CEO and co-founder of the company, said in the Q3 2024 earnings call, “Our business has entered a new growth phase with multiple product families ramping across AI platforms, featuring both third-party GPUs and internally developed AI accelerators, which drove the Q3 sales upside verus our guidance.”

  • Management also provided a strong Q4 revenue guide of $126 million to $130 million, representing YoY growth of 153.4% at the mid-point. This also represents QoQ growth of 13% at the midpoint.

    According to management, the QoQ growth is being driven by “our Aries product family across a diverse set of AI platforms, some of which are just starting to ramp and also from our Taurus SCM for 400-gig applications, and additional preproduction shipments of our Scorpio P-Series switches.”

  • Analysts estimate revenue to grow 103.5% to $132.78 million in Q1 2025 and 85.2% YoY to $142.34 million in Q2 2025. While the growth rates are strong, growth is expected to slow down due to tougher comps.

Jitendra Mohan said, “Looking into Q4, we expect our revenue momentum to continue, largely driven by the Aries PCIe and Taurus Ethernet product lines. The Scorpio Fabric Switches are continuing to ship in preproduction volumes.”

The newly introduced Scorpio Fabric Switches are expected to increase the total market opportunity to more than $12 billion by 2028 for the company. Scorpio Switches are also expected to constitute more than 10% of revenue in 2025.

Source: Company website

Analysts expect 2024 revenue growth of 230.9% YoY to $383.14 million, followed by 55.5% and 40.4% in the subsequent years. Meanwhile, management comments seem to imply the growth recently reported will sustain, implying the growth phase has only begun: “Our business has entered a new growth phase with multiple product families ramping across AI platforms, featuring both third-party GPUs and internally developed AI accelerators, which drove the Q3 sales upside versus our guidance.”

Management also later stated: “Yes, right now, our visibility is very strong, both as always with our backlog position, but also the breadth of designs we have — right now, we're really kind of entering a new phase of growth here where our revenue streams are clearly diversifying […].”

Margins:

The company’s margins are improving, helped by strong operating leverage. However, the product mix might weigh on the margins going forward. Management mentioned in the recent earnings call that they have a long-term gross margin target of 70%.

  • Q3 gross profits grew by 212.7% YoY to $87.88 million or 77.7% of revenue compared to 76.1% in the same period last year.
  • This compares to 73.5% for FY2022 and 68.9% for FY2023.
  • Adjusted gross margin was 77.8% compared to 76.1% in the same period last year.
  • Management gross margin and adjusted gross margin guide for the next quarter are 75%. This is down sequentially due to higher product mix towards hardware solutions during the quarter.
  • Operating margin was (-7.9%) compared to (-5.3%) in the same period last year. The adjusted operating margin expanded to 32.4% from 2% in the same period last year, helped by strong operating leverage.
  • Management operating margin guide for the next quarter is (-4.3%) and adjusted operating margin guide is 32.4%.
  • It’s important to consider that stock-based compensation is quite high due to the recent IPO at 40.3% this quarter, and was at 56% last quarter. Once SBC naturally levels out, this company has strong enough margins to become GAAP profitable.
  • Net loss was (-$7.6 million) or (-6.7%) of revenue compared to (-8.5%) in the same period last year.
  • Adjusted net income improved significantly to $40.28 million or 35.6% of revenue compared to (-$0.41 million) or (-1.1%) of revenue in the same period last year.

The difference between the GAAP and non-GAAP net income is due to high stock-based compensation. Stock-based compensation was $45.5 million or 40.3% of revenue in the recent quarter. Stock-based compensation has been lumpy as the company’s IPO was in March 2024.

EPS:

The company beat EPS estimates helped by solid operating leverage. Q3 GAAP loss per share was ($0.05) compared to ($0.08) in the same period last year, beating estimates by 28.1%. Adjusted EPS was $0.23 and beat estimates by 35%.

  • Management has guided the Q4 GAAP EPS in the range of $0.04 to $0.06.
  • Notably, the company will not be GAAP operating income positive next quarter as the guide suggests an operating loss of (-$5.5 million) and interest income of about $10 million in the next quarter will make it GAAP profitable. However, it’s very close to being GAAP profitable across all margins, and we think it’s only a matter of time before this happens.
  • Adjusted EPS guide is $0.25 to $0.26.
  • Analysts expect adjusted EPS to be $0.25 in Q1 2025 and $0.27 in Q2 2025.
  • Analysts expect strong EPS growth. They expect 2025 adjusted EPS to grow 58.9% YoY to $1.14 and 48.6% YoY to $1.70 in 2026.

Cash Flow and Balance Sheet:

The company’s cash flows margins are high and are also improving due to increased profits.

  • Q3 operating cash flow was $63.5 million or 56.2% of revenue compared to (-0.9%) in the same period last year. It is also a significant improvement from (-17%) for the full year 2022 and (-6.5%) for 2023.
  • Free cash flow was $46.81 million or 41.4% of revenue compared to (-$1.07 million) or (-2.9%) in the same period last year.
  • Inventory was $24.4 million compared to $28.6 million in Q2.
  • The company had cash and marketable securities of $886.8 million with no debt.
  • Net proceeds from the IPO were $672.2 million. The shares began trading on the NASDAQ on March 20, 2024.

Valuation and IPO Risk:

On a sales valuation, Astera Labs is trading higher than Nvidia at 36X Fwd P/S compared to Nvidia’s at 29 Fwd P/S. On the bottom line, whether it’s PE Ratio, EV/EBITDA or Price to FCF, Astera is trading nearly double Nvidia’s valuation.

The company went public on March 20th, 2024 with shares opening for public trading around $62. The company raised $672.2 million with a lockup that expired September 11th, 2024. Insiders saw shares priced at $36 per share at time of IPO and the highest the stock has traded is $95 following the last earnings report.

The valuation is testing the upper limits of AI semi-related stocks. Therefore, we foresee participating now as a momentum play and participating longer-term with a new entry sometime late 2024-early 2025.

This stock requires an active stance, to where if we enter, we will exit for a quick trade, and try again at a lower valuation for a longer-term position. We offer real-time trade alerts on our Advanced tier.

Conclusion:

Management commentary is at odds with analysts’ estimates, as the commentary suggests there is a new growth phase occurring while estimates suggest a drop off in growth over the next 1-2 years. The products are enabling faster data speeds with PCIe5 and also increased back-end networking with PCIe6 for large AI clusters, and thus management’s commentary that this is a new growth phase holds weight. We certainly know the future for AI clusters is going to exponentially increase from 10s of thousands for AI clusters to eventually millions of AI accelerators per cluster. This is not only a GPU opportunity but also a custom silicon opportunity, as Astera Labs exclusively offered PCIe5 switches and retimers, and will now compete against Broadcom on PCIe6.

Astera Labs is technically in the lead and Broadcom is the follower in this case; but where it gets even more interesting is with the new product Scorpio. It is expected to increase the TAM by $5.0 billion with a total TAM of over $12B over the next 3 years. If we assume Astera captures 50% of the total TAM, then what we have is a stock that will remain in hypergrowth territory. If you do the math, that’s a potential 12X increase in revenue by 2028 from the $500M run rate Astera has today.

Certainly, technical analysis is critical given we are dealing with a very stretched valuation, a recent IPO, and ultimately, a stock that will be volatile in the years to come – yet perhaps, with the right timing, the stock will be as equally rewarding. We typically do not participate in IPOs, and Astera Labs illustrates why given its trading 2X higher than the AI juggernaut with impeccable financials (Nvidia). We do not plan to participate long-term in Astera Labs for this reason, yet may participate briefly, and then keep the stock on our watchlist from there.

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:

Astera Labs: Hypergrowth AI Networking Stock

Posted on by io-fund

Astera Labs is a stock that simply cannot be ignored despite having a fraction of the market cap compared to much larger AI hardware players.

The company offers a few key products that are enabling larger and faster AI clusters. For data center AI accelerators, the company was first to offer PCIe 5 switches and retimers. Growth will likely continue due to next-generation PCIe 6 products with higher average sales prices that will be released in 2025 and ramp in 2026.

There are additional growth opportunities, including a software architecture that allows hyperscalers to monitor their data center infrastructure and increase utilization rates; the software helps to incentivize hyperscalers to use Astera Labs hardware products. The company also offers ethernet smart cable modules and is introducing a new CXL product for CXL-enabled CPUs next year.

We are interested in Astera Labs for the increased average sales prices that are expected to persist at least through 2025-2026 due to the Aries products and upcoming Scorpio products. Astera Labs is a highly technical company, yet this quote clearly communicates why the growth trajectory can sustain:

“One is generally speaking with each new generation of a protocol like PCIe going from Gen 5 to Gen 6. There is an ASP uplift. That's number one. Number two, of course, we were hinting at the Scorpio product line, which because of the value it delivers to customers is at a higher ASP, as you can imagine. So overall, if you look at the design wins we have today, the dollar content per GPU goes up, that's one way to look at it based on what we've shared before […] So overall, if you look at sort of the increasing speed, additional product line as well as the fact that the internally developed platforms, AI accelerated database platforms, they are starting to gain more and more traction. So when you look at all of them, on an average, our content is on the up.”

Investment Thesis:

  • Current Growth Driver: Increased average sales prices are being driven by CPUs, GPUs and ASICs all moving to the new PCIe 5.0 standards. Arista Labs’ Aries Retimers and PCIe 5.0 components is driving the current growth, and the company is unchallenged in this new generation of PCIe, which came to market for AI accelerators only recently with Nvidia’s H200s.
  • Catalyst: Scorpio is a new product that is expected to expand the TAM to more than $12B by 2028: Astera Labs is releasing a PCIe Gen 6 fabric switch custom designed for AI data flows with high performance per watt compared to incumbents. Rather than building a large switch, the company built a smaller device that is more efficient for high-speed signals. With Scorpio, Astera Labs is defending its dominance in PCIe5 by doubling the bandwidth at lower power requirements than the 5th generation of PCIe.
  • The market demand for Astera Labs is healthy, as a major supplier to Nvidia’s PCIe-enabled GPUs (note: most of Nvidia’s GPUs use their in-house NVLink). Yet, Arista also supplies other AI accelerator platforms for Big Tech, and Arista is the only provider for PCIe5. In addition to hardware, ALAB offers Big Tech data centers software-defined architecture called COSMOS that allows for performance monitoring. There is some favorable vendor lock-in dynamics with COSMOS as the many different product lines can be optimized and monitored with the software.
  • Financials on Astera Labs are impressive as the company is hypergrowth at 206% growth last quarter and 153% guided next quarter. The free cash flow margin is at 41%. The company is not GAAP profitable due to being a recent IPO with outsized stock-based compensation. However, there is a path to GAAP profitability, which is detailed in the financials section below.

Overview of Astera’s Products:

Broadly speaking, Astera’s products are seeing increased relevance as AI clusters grow to support hundred-billion and trillion parameter models. The company’s hardware and software increase AI server performance and productivity for the current generations and future generations of AI accelerators.

Transition to PCIe5 and PCIe5 Retimers are Driving Astera’s Growth:

The Aries products offer PCIe5 interfaces that GPUs and AI accelerators (like custom silicon) use to connect components including ethernet networking. Compared to the previous generation, PCIe5 is twice as fast with data transfer rates that reach up to 128 GBs/s on multi-directional bandwidth in each lane. By increasing the data transfer for each lane, it allows more lanes to become available to help leverage the power of the GPU or AI accelerator.

PCIe5 Retimers are chips that boost signals across high-speed components and are seeing increased demand, starting with Nvidia’s H200s, and also for application-specific chips. Specifically, Aries Retimers and smart cable modules allow hyperscalers to connect multiple racks together with up to 7 meters of copper cables. Aries can also go up to 50 meters with optical fiber. According to a presentation at Nvidia’s GTC event, this is 3X the standard reach defined in PCIe specs.

Despite PCIe5 being out since 2019, it was Nvidia’s H200s released in 2024 that were the first data center GPUs to use PCIe5. What’s interesting is that Astera is said to have captured 95% of the XPU market, which refers to application-specific chips that are specific on a product level. Per an analyst on the earnings call: “Majority of the XPU shipments are still going to be I think Gen 5 based where your market share is still somewhere in the range of, I think, like 95%.”

Management also stated: “The upside to the guidance was driven largely from Aries' revenue, both for the third-party GPUs, but also as well with the strong ramps on new platforms on the internally developed AI accelerators. And we're seeing that across multiple hyperscaler customers, so it's not just one. So the upside was largely driven by that Aries revenue.”

Notably, Aries devices are used to interconnect AI accelerators with CPUs and networking, yet are also used for backend networking between GPUs for larger clusters. Astera supplies the HGX H100 systems with PCIe-based GPUs with up to 25 retimer chips per HGX system.

Scorpio PCIe6 Custom-Built for AI is a Catalyst:

Scorpio is where the excitement is building for continued growth next year with management stating it will “exceed 10% of revenues in 2025” with “good momentum going into 2026.” This is due to the PCI Express switches being custom built for AI purposes, whereas in the past, PCIe was built for storage and then retrofitted for AI purposes.

PCIe 6 doubles the bandwidth from the 5th generation, with up to 256 GB/s of bandwidth per lane, which will require faster supporting components, such as the retimers that Astera Labs offers. The demo from GTC showed the Scorpio fabric switches (name released in October) delivering twice the bandwidth with less power at 11W instead of the 13W from the PCIe 5 interfaces.

There are two Scorpio fabric switches. The Scorpio P-Series is a small chip that connects the CPU, GPU, NIC and NVMe storage. Rather than building a large switch, the company built a smaller device that is more efficient for high-speed signals to help feed GPUs with data. The fewer ports and smaller switch decrease complexity in a bid to compete against Broadcom with twice the lane count.

The X-Series is for back-end networking in GPU-to-GPU configurations, and will offer a higher port count. Astera is essentially building something similar to Nvidia’s NVSwitch with the X-Series, but for PCIe-enabled GPUs: “And this one, like Mike noted, it's a greenfield use case, meaning if you keep Nvidia and NV Switch aside, everyone else is starting to build configurations that are obviously going to need some kind of a switching functionality, which is what we are addressing with our X Series device.” The X-Series improves efficiency for ever-increasing AI cluster sizes. The majority of AI clusters are in the tens of thousands GPUs, but are expected to go to the hundreds of thousands (already has with X and some other Big Tech companies), and will see AI clusters with millions of GPUs over the next couple of years.

Here is another quote from the management team as to why the X-Series Scorpio switch fabric is a big opportunity for their company:

“X series will have a bigger TAM. The TAM today is nearly zero. It's not very commonly used outside of the Nvidia ecosystem. We do expect many hyperscalers to start deploying this, starting with the X family and the designs for which that we have. And we are able to do that because of the architecture that we have. Because of our software-defined architecture, we can customize many parts of the X-Series to cater to the specific requirements of the hyperscalers both on the side of performance, the exact configuration that they require in count and so on and also the diagnostics framework that they require to monitor their infrastructure. So over time, we do expect X-Series to become larger [than the P-Series].”

COSMOS Software:

In order of importance, COSMOS Software ranks higher than some of Astera’s hardware as it offers performance monitoring for data center infrastructure. This is especially important for Big Tech companies concerned with utilization rates for expensive GPU systems.

The adoption of the software stack to monitor for performance is expected to increase with the Scorpio X-Series:

“Where the Scorpio family sits, we have access to a lot more diagnostic information. And we can couple that with the information that we are collecting from our other families deployed such as Aries and even Taurus to provide a holistic view of the AI infrastructure to the data center operators. So both from the hardware side, the kind of the purpose-built nature of these devices as well as the software stack that comes with it is a big differentiator for us.”

Taurus Ethernet Smart Cable Modules:

Astera Lab offers Ethernet smart cable modules which help to alleviate bandwidth issues with 100-gig per lane connectivity over copper cables including AEC. The company recently released 400-gig Ethernet SCMs, which help to stabilize the network. When thinking of ALAB’s investment thesis, Ethernet is not top-of-mind given the sheer size and lead we see from Broadcom, Nvidia’s Spectrum, and Arista Networks.

Right now, the maximum bandwidth supported by PCIe 5.0 is 400Gbps per port. By using 106Gbps PAM4 SerDes, ASICs can be tuned to support 100, 200 and 400 Gbps port speeds. To work around this, and to achieve 800Gbps, larger chip makers are building NICs directly into the accelerator. According to The Register, the 800Gbps ports built into accelerators may reduce bottlenecks before PCIe 6.0 arrives on the market. The larger Ethernet players are moving quickly on this, and we will need to keep an eye on Astera Labs to determine if the company’s Taurus product can remain competitive.

Leo Compute Express Link (CXL):

Leo is slated to impact revenue in 2025 when data center platforms plan to utilize CXL technology for memory bandwidth and capacity bottlenecks. Next year, CXL-capable CPUs will become broadly available. We’ve covered in the past how CXL is a new server architecture that “dynamically assigns memory resources between servers.” The result is boosted memory bandwidth and also at a lower cost than adding more CPUs. Partially-disaggregated racks are expected to deploy in 2024-2025 with separate compute, memory and I/O racks with the interconnect being CXL.

Per Astera’s management team: “In the past, this was done by adding additional CPUs into the server box to provide for more memory channels. But what we have demonstrated is that by using Leo you're not only able to get the higher performance by the added memory. But from an overall TCO standpoint, it's significantly less than adding additional CPUs. “

Financials: Strong Revenue Growth, Expanding Margins

Astera Labs reported solid top-line and bottom-line beats in the recent Q3 results. The company reported record revenue of $113.1 million, up 47% QoQ and 206% YoY, beating estimates by 16.1%. The adjusted operating margin expanded to 32.4% from 2% in the same period last year, which was helped by strong operating leverage. The adjusted EPS of $0.23 beat estimates by 35%.

Revenue:

The company is one of the fastest-growing tech companies and is emerging as a rising star in the AI data center networking space. The company’s Q3 revenue grew by 206.2% YoY and 47% QoQ to $113.1 million.

Jitendra Mohan, CEO and co-founder of the company, said in the Q3 2024 earnings call, “Our business has entered a new growth phase with multiple product families ramping across AI platforms, featuring both third-party GPUs and internally developed AI accelerators, which drove the Q3 sales upside verus our guidance.”

  • Management also provided a strong Q4 revenue guide of $126 million to $130 million, representing YoY growth of 153.4% at the mid-point. This also represents QoQ growth of 13% at the midpoint.

    According to management, the QoQ growth is being driven by “our Aries product family across a diverse set of AI platforms, some of which are just starting to ramp and also from our Taurus SCM for 400-gig applications, and additional preproduction shipments of our Scorpio P-Series switches.”

  • Analysts estimate revenue to grow 103.5% to $132.78 million in Q1 2025 and 85.2% YoY to $142.34 million in Q2 2025. While the growth rates are strong, growth is expected to slow down due to tougher comps.

Jitendra Mohan said, “Looking into Q4, we expect our revenue momentum to continue, largely driven by the Aries PCIe and Taurus Ethernet product lines. The Scorpio Fabric Switches are continuing to ship in preproduction volumes.”

The newly introduced Scorpio Fabric Switches are expected to increase the total market opportunity to more than $12 billion by 2028 for the company. Scorpio Switches are also expected to constitute more than 10% of revenue in 2025.

Source: Company website

Analysts expect 2024 revenue growth of 230.9% YoY to $383.14 million, followed by 55.5% and 40.4% in the subsequent years. Meanwhile, management comments seem to imply the growth recently reported will sustain, implying the growth phase has only begun: “Our business has entered a new growth phase with multiple product families ramping across AI platforms, featuring both third-party GPUs and internally developed AI accelerators, which drove the Q3 sales upside versus our guidance.”

Management also later stated: “Yes, right now, our visibility is very strong, both as always with our backlog position, but also the breadth of designs we have — right now, we're really kind of entering a new phase of growth here where our revenue streams are clearly diversifying […].”

Margins:

The company’s margins are improving, helped by strong operating leverage. However, the product mix might weigh on the margins going forward. Management mentioned in the recent earnings call that they have a long-term gross margin target of 70%.

  • Q3 gross profits grew by 212.7% YoY to $87.88 million or 77.7% of revenue compared to 76.1% in the same period last year.
  • This compares to 73.5% for FY2022 and 68.9% for FY2023.
  • Adjusted gross margin was 77.8% compared to 76.1% in the same period last year.
  • Management gross margin and adjusted gross margin guide for the next quarter are 75%. This is down sequentially due to higher product mix towards hardware solutions during the quarter.
  • Operating margin was (-7.9%) compared to (-5.3%) in the same period last year. The adjusted operating margin expanded to 32.4% from 2% in the same period last year, helped by strong operating leverage.
  • Management operating margin guide for the next quarter is (-4.3%) and adjusted operating margin guide is 32.4%.
  • It’s important to consider that stock-based compensation is quite high due to the recent IPO at 40.3% this quarter, and was at 56% last quarter. Once SBC naturally levels out, this company has strong enough margins to become GAAP profitable.
  • Net loss was (-$7.6 million) or (-6.7%) of revenue compared to (-8.5%) in the same period last year.
  • Adjusted net income improved significantly to $40.28 million or 35.6% of revenue compared to (-$0.41 million) or (-1.1%) of revenue in the same period last year.

The difference between the GAAP and non-GAAP net income is due to high stock-based compensation. Stock-based compensation was $45.5 million or 40.3% of revenue in the recent quarter. Stock-based compensation has been lumpy as the company’s IPO was in March 2024.

EPS:

The company beat EPS estimates helped by solid operating leverage. Q3 GAAP loss per share was ($0.05) compared to ($0.08) in the same period last year, beating estimates by 28.1%. Adjusted EPS was $0.23 and beat estimates by 35%.

  • Management has guided the Q4 GAAP EPS in the range of $0.04 to $0.06.
  • Notably, the company will not be GAAP operating income positive next quarter as the guide suggests an operating loss of (-$5.5 million) and interest income of about $10 million in the next quarter will make it GAAP profitable. However, it’s very close to being GAAP profitable across all margins, and we think it’s only a matter of time before this happens.
  • Adjusted EPS guide is $0.25 to $0.26.
  • Analysts expect adjusted EPS to be $0.25 in Q1 2025 and $0.27 in Q2 2025.
  • Analysts expect strong EPS growth. They expect 2025 adjusted EPS to grow 58.9% YoY to $1.14 and 48.6% YoY to $1.70 in 2026.

Cash Flow and Balance Sheet:

The company’s cash flows margins are high and are also improving due to increased profits.

  • Q3 operating cash flow was $63.5 million or 56.2% of revenue compared to (-0.9%) in the same period last year. It is also a significant improvement from (-17%) for the full year 2022 and (-6.5%) for 2023.
  • Free cash flow was $46.81 million or 41.4% of revenue compared to (-$1.07 million) or (-2.9%) in the same period last year.
  • Inventory was $24.4 million compared to $28.6 million in Q2.
  • The company had cash and marketable securities of $886.8 million with no debt.
  • Net proceeds from the IPO were $672.2 million. The shares began trading on the NASDAQ on March 20, 2024.

Valuation and IPO Risk:

On a sales valuation, Astera Labs is trading higher than Nvidia at 36X Fwd P/S compared to Nvidia’s at 29 Fwd P/S. On the bottom line, whether it’s PE Ratio, EV/EBITDA or Price to FCF, Astera is trading nearly double Nvidia’s valuation.

The company went public on March 20th, 2024 with shares opening for public trading around $62. The company raised $672.2 million with a lockup that expired September 11th, 2024. Insiders saw shares priced at $36 per share at time of IPO and the highest the stock has traded is $95 following the last earnings report.

The valuation is testing the upper limits of AI semi-related stocks. Therefore, we foresee participating now as a momentum play and participating longer-term with a new entry sometime late 2024-early 2025.

This stock requires an active stance, to where if we enter, we will exit for a quick trade, and try again at a lower valuation for a longer-term position. We offer real-time trade alerts on our Advanced tier.

Conclusion:

Management commentary is at odds with analysts’ estimates, as the commentary suggests there is a new growth phase occurring while estimates suggest a drop off in growth over the next 1-2 years. The products are enabling faster data speeds with PCIe5 and also increased back-end networking with PCIe6 for large AI clusters, and thus management’s commentary that this is a new growth phase holds weight. We certainly know the future for AI clusters is going to exponentially increase from 10s of thousands for AI clusters to eventually millions of AI accelerators per cluster. This is not only a GPU opportunity but also a custom silicon opportunity, as Astera Labs exclusively offered PCIe5 switches and retimers, and will now compete against Broadcom on PCIe6.

Astera Labs is technically in the lead and Broadcom is the follower in this case; but where it gets even more interesting is with the new product Scorpio. It is expected to increase the TAM by $5.0 billion with a total TAM of over $12B over the next 3 years. If we assume Astera captures 50% of the total TAM, then what we have is a stock that will remain in hypergrowth territory. If you do the math, that’s a potential 12X increase in revenue by 2028 from the $500M run rate Astera has today.

Certainly, technical analysis is critical given we are dealing with a very stretched valuation, a recent IPO, and ultimately, a stock that will be volatile in the years to come – yet perhaps, with the right timing, the stock will be as equally rewarding. We typically do not participate in IPOs, and Astera Labs illustrates why given its trading 2X higher than the AI juggernaut with impeccable financials (Nvidia). We do not plan to participate long-term in Astera Labs for this reason, yet may participate briefly, and then keep the stock on our watchlist from there.

Recommended Reading:

AMD Q3 2024: GPU Revenue at 22%, and AI PCs have a Leader

Posted on by io-fund

AMD beat estimates by $110 million on the top line and was in line on adjusted EPS expectations of $0.92. Despite the beat in Q3, AMD guided for Q4 revenue slightly below consensus estimates at midpoint.

AMD is a fundamentally strong company with data center revenue growth accelerating for many quarters now, client revenue rebounded sequentially, and margins are improving. As Jean Hsu stated toward the end of the Q&A: “But when you look at our Data Center segment performance, we more than doubled the revenue year-over-year, but we tripled the operating income year-over-year.”

Regarding the Client segment, AMD and analysts acknowledged there may be consumer weakness in PCs from other companies, yet AMD stated they have the strongest PC line up in company history on the market today. This is evident considering Intel guided Q3 to be flat to down in their Client segment yet AMD’s growth was up 29% YoY and 26% QoQ to $1.88 billion.

The slight miss at the midpoint is likely due to Embedded, as Jean Hu, the CFO, stated next quarter would be: “driven by strong growth in our Data Center and Client segment, more than offset decline in the Gaming and Embedded segments.” It was further stated that embedded “demand continues recovering gradually, led by strength in test and emulation offset by ongoing softness in the industrial market.”

Revenue

AMD reported revenue of $6.82 billion in the third quarter, for YoY growth of 17.6%, a ~770 bp sequential acceleration from 8.9% YoY growth in the prior quarter. Data center drove Q3’s growth, while Client revenue rebounded significantly, offsetting continued weakness in gaming and embedded.

CEO Lisa Su said that AMD delivered “record revenue led by higher sales of EPYC and Instinct data center products and robust demand for our Ryzen PC processors,” with “significant growth opportunities across our data center, client and embedded” moving forward.

Looking ahead, AMD guided for revenue of $7.5 billion, +/- $300 million, for Q4, correlating to YoY growth of ~21.6% at midpoint, or a 400 bp sequential acceleration. While this fell just shy of the consensus estimate for 22.3% YoY growth to $7.55 billion, revenue growth is still expected to accelerate to the low-30% range in the first half of fiscal 2025.

Margins

As noted in our pre-earnings report, AMD’s margins continue to benefit from the increasing mix of data center products, primarily from EPYC CPUs as GPUs are currently dilutive to margins but will eventually be accretive to margins over time.

  • Q3 gross margin was 50%, up from 49% last quarter and 47% in the year ago quarter. Adjusted gross margin was 54%, slightly ahead of the 53.5% guide, and increasing from 53% last quarter and 51% in the year ago quarter.
  • For Q4, management guided for adjusted gross margin of 54%, flat sequentially.
  • Q3 operating margin expanded to the double-digit range, at 11%, driven by increased data center mix – this compares to a 5% operating margin last quarter and a 4% margin in the year ago quarter.
  • Operating income surged sequentially, with Q3’s operating income of $724 million up ~169% QoQ from $269 million. This was driven primarily by data center growth and margin expansion (covered below in Segments).
  • Adjusted operating margin came in at 25%, in line with expectations, while management’s guidance implies Q4’s adjusted operating margin rises to nearly 27%.
  • Q3 net income was $771 million, an increase of 191% QoQ and 158% YoY. Net margin was 11% in the quarter, up from 5% last quarter and last year. Adjusted net income was $1.50 billion, up 34% QoQ and 33% YoY, for a 22% margin.

EPS

GAAP EPS significantly improved in the quarter due to margin substantially improving, with operating and net margin both in the double digit range in the quarter.

  • GAAP EPS of $0.47 beat estimates for $0.41, and represented YoY growth of 161% and QoQ growth of 194%.
  • Adjusted EPS of $0.92 met estimates, and represented YoY growth of 31% and QoQ growth of 33%.

Given the revenue acceleration through Q4 and the first half of 2025 along with improved operating leverage from increased data center mix driving higher operating margins, adjusted EPS growth is expected to accelerate more than 40 percentage points to the mid- to high-70% range by Q2 2025.

Cash and Balance Sheet

Cash flows still have room to improve, as margins contracted in Q3, with AMD reporting operating and free cash flow margins falling by 100 bp QoQ. However, other line items, particularly accounts receivable and inventories, surged sequentially, hinting at potential strong growth ahead.

  • Operating cash flow was $628 million in Q3, or a 9% margin. This compares to a 10% margin in both Q1 and Q2.
  • Free cash flow was $496 million, or a 7% margin, versus an 8% margin in Q2.
  • Cash and equivalents totaled $4.54 billion, while debt totaled $1.72 billion.
  • Inventories totaled $5.37 billion, rising nearly 8% QoQ as AMD continues to ramp data center GPUs and move towards an annual release cadence.
  • Accounts receivable surged 26% QoQ to $7.24 billion, after hovering in the $5 billion range for the last four quarters.

Segments

Data Center

AMD’s data center segment once again drove growth in the quarter, with management boosting FY24’s AI revenue target once more, now seeing AI revenue exceeding $5 billion, versus a prior view for $4.5 billion-plus in AI revenue. Management stated AI revenue is at $1.5 billion per quarter, or 22% of revenue.

Data center revenue accelerated 7 percentage points to 122% YoY, with AMD reporting $3.55 billion in revenue in the segment. QoQ growth was 25%, accelerating from 21% QoQ in Q2. AMD once again witnessed strong demand for AMD Instinct GPUs and EPYC server CPUs.

Data center’s operating margin continues to expand, with segment operating income rising 240% YoY and 40% QoQ to $1.04 billion. This was an operating margin of 29%, expanding from 26% last quarter and 19% in the year ago quarter.

Zen 5 Turin launched this month and will help support data center sales next year. Regarding the AMD versus Intel battle, Lisa Su made it clear that AMD continues to take market share: “We believe we gained server CPU share in the quarter as enterprise wins accelerated. Cloud providers expanded their use of EPYC CPUs across their infrastructure, and we began the initial ramp of fifth-gen EPYC processors” and that “Meta alone has deployed more than 1.5 million EPYC CPUs across their global data center fleet to power their social media platforms.”

The MI325X launched earlier this month with increased memory capacity and bandwidth, with AMD stating it offers 20% higher inferencing than the H200. The MI325X is in production shipment this quarter and “interest for MI325X is high.” The MI350 will launch in H2 2025 and the MI400s with CDNA Next architecture will launch in 2026. We had stated in our pre-earnings writeup that CDNA 4 and also CDNA Next architecture should be a defining moment for AMD in terms of narrowing the product road map with Nvidia.

Regarding the MI300 AI accelerators, Meta and Microsoft are large customers due to TCO advantages (total cost of ownership). Management also offered statements around RocM’s progress, stating that foundational support is growing and performance gains are improving by 2.4X.

Client Segment:

Though there were some weaker data points around PCs, as stated in the pre-earnings writeup, AMD is less of a concern in that regard as the company’s lineup is loaded with stellar releases. The company recently released the Zen 5 architecture including the Ryzen AI 300 laptops with a neural processing unit (NPU) with 50 TOPS of AI performance, and the Ryzen 9000 series for desktops – making them the most powerful units on the market today. Lisa Su stated it “this is the strongest PC portfolio we’ve had in our history.”

AMD’s ‘Zen 5’ Ryzen processors were met with “strong demand,” driving Client revenue up 29% YoY and 26% QoQ to $1.88 billion.

Although it looks like quite a sharp deceleration over the past two quarters, Client revenue has reached the highest level since Q2 2022. Client operating income also increased, up 97% YoY and 210% QoQ to $276 million; this represents an operating margin of 15%, up from 6% last quarter and 10% in the year ago quarter.

Management stated that AMD has “very high” share in the desktop channel, and that AMD “saw some of our highest sell-through.”

Gaming

AMD has still not escaped the trough in gaming, with revenue declining (69%) YoY and (29%) QoQ to $427 million. AMD said that the weakness was due to a decline in semi-custom revenue. Operating income for the segment also dropped substantially, falling to just $12 million, or a 3% margin, compared to a 12% margin last quarter and a 14% margin last year.

Embedded

Embedded revenue has begun to recover, following management’s comments last quarter about order patterns improving. Revenue in the segment rebounded 8% QoQ but declined (25%) YoY to $927 million. Operating margin for the segment was 40%, flat QoQ and down from 49% last year.

Earnings Call:

AI Revenue:

There was a question on the call about how large AI revenue is on a quarterly basis, to which Lisa Su provided more information, stating it exceeds $1.5 billion. The comment that the GPU business is approaching the scale of the CPU business will be key for investors to put into perspective, as it’s a big statement as we move into 2025.

Timothy Arcuri   

I had a quick 1 and then a more intensive question. So the first one is I wanted to ask about the September actuals for Data Center GPU. It seems like it was in the $1.5 billion range. And that would put December in kind of the $2 billion range. Is that about right?

Lisa Su   

So it's a pretty granular question, Timothy. But maybe let me help you with this. We actually did better in the Data Center GPU business relative to our initial expectations. So you would imagine that the business was actually greater than $1.5 billion. I mean we're actually seeing now our GPU business really approaching the scale of our CPU business.

-End Quote

There was an analyst on the call attempting to clarify if AI revenue would be flat QoQ.

Stacy Rasgon:

[…] You said it was approaching the size of Your compute business which you put around what under $1.7 billion, maybe a little more. Is that right? And like if that is right, it implies that at $5 billion for the year, you'd actually be down in Q4. So I'd probably got to be $5.2 billion or $5.3 billion for the full year, just to be flat sequentially and more than that to get growth […]

Lisa Su:

Right, Stacy. So first, a couple of things. You have to remember that in our Data Center segment, we have some other revenue that is not CPUs and GPUs, right? We have some FPGAs and other things. But the question earlier was the revenue of $1.5 billion, and I said that it was greater than $1.5 billion. So take that as a fundamental. And then as — we talked about — we didn't guide an exact number for the data center GPU. We said exceed $5 billion.

-End Quote

Lisa Su also stated: “What I would say about 2025 is we feel very good about the growth opportunities I would say that it might be lumpy. In general, these are large customer acquisitions and it's not always predictable exactly which quarters you would expect the significant build out.”

My comment: Lumpy to the upside … sounds bullish for next year.

Here was another time that Lisa Su clarified that the lumpiness would be to the upside: “So these are large customers that drive deployments. Like for example, the third quarter was a bit higher than we expected. That was driven by some additional customer demand, and we may see that type of lumpiness.”

There was a pointed question about how AMD plans to catch up to Nvidia’s product road map, to which it was stated: “I think MI300, when we launched it was behind H100, H100 was in the market for a much longer time. And we have with our accelerated road map actually closed a good part of that gap. I think MI325 is a great product. It's going to compete very well with H200 and the MI350 series will compete very well with Blackwell.”

It was also insinuated that Blackwell’s AI systems, which are more complex, could be a tailwind for AMD. “In the overarching view of the world, frankly, the market continues to be constrained, particularly in the newer product generations. It takes a long time to go from, let's call it, shipping your first samples to actually ramping in volume production workloads. And I think one of the advantages that we have with the — with our portfolio is that from a data center retrofit standpoint, it's actually a much easier ramp, just the infrastructure is the same.”

If we read between the lines, Lisa Su is stating that AMD is positioned to answer the overflowing demand from Blackwell, and in a way the supply chain can handle, as it’s well-known that Blackwell is running into wafer capacity limits compounded by a larger die size.

Lisa Su also echoed my understanding of roughly when AMD should narrow the product road map with Nvidia: “We feel very good about the progress I think next year is going to be about expanding both customer set as well as workload. And as we get into the MI400 series, we think it's an exceptional product. So — all in all, the ramp is going well, and we will continue to earn and — earn the trust and the partnership of these large customers.” The MI400 is on the new CDNA architecture and is ramping in 2026.

Margins:

As stated in the pre-earnings writeup, margins are an area where AMD and Nvidia offer quite a contrast. AMD’s data center margin is 29% with a company operating margin of 11% compared to Nvidia’s 60%.  The guide is for flat margins next quarter. The CFO was encouraging in terms of what to expect for 2025: “When we scale the company next year, you can see we're going to benefit from economies of scale to continue to drive our operational efficiency to improve gross margin.”

Conclusion:

As we get more earnings reports this quarter, it should become evident that AMD’s Client growth is unusually strong and is truthfully a defining moment. Remember, we had pulled PC data that showed flat to negative PC growth YoY for Q3 industry wide. In the pre-earnings report, I had stated AMD is incrementally stronger and could go unscathed, but this is quite the growth in a quarter where unit sales were flat to declining industry wide. We will not lose sight of this incremental strength as we plan for 2025.

Regarding AMD’s GPU story … slowly but surely, AMD will show the market it should take the company seriously. These things take time. It was our understanding going into this report that the 2025 MI350s and 2026 MI400s is when the product release cycle will start to narrow with Nvidia, and this was echoed on the call.

On the AH price action, we had pointed out on the Q4 webinar that SMH was looking unusually weak. Whether it’s due to potential tariffs or potential consumer-facing weakness in the semiconductor industry, or a combination of both, I’m not sure. But the point is that I do not believe AMD is selling off for reasons that are specific to the stock. The earnings report was strong, and we will look to keep this as a leading position for next year.

Recommended Reading:

Fabrinet: Datacom, Nvidia Driving Optical Revenue Growth

Posted on by io-fund

Last month, we covered the importance of optical interconnects in linking GPUs together in clusters in our thematic deep dive “Optical Interconnects Overview: Strong Growth Expected Ahead.” Optical transceivers address bandwidth, or the speed of data transfer in the data center. As hyperscalers work to expand capacity, it’s expected there will be a surge in optic connections.

Fabrinet provides advanced optical communications components for datacom and telecom end markets, customized optics and glass fabrication, and advanced laser and other electro-mechanical parts. Fabrinet has reported consecutive quarters of strong datacom growth, stemming from one of its flagship customers, Nvidia. Fundamentally, Fabrinet possesses stable margins and strong cash flow generation, as top and bottom line growth is expected to accelerate from the low-teens to the 20% range over the next few years.

Fabrinet is clearly a beneficiary of Nvidia’s surging growth in the AI GPU market, with Nvidia’s contribution nearly tripling from 12.5% to 35.1% over the past twelve months. We’ve covered how Blackwell brings an enormous revenue opportunity for Nvidia and its suppliers, and Fabrinet is expected to be a beneficiary of Nvidia’s upcoming superchip, with long-term growth opportunities for optics in the AI data center and in 800G+ data rates.

Fabrinet reports earnings on Nov 4th.

Revenue Growth Reaccelerated in 2024; 149% Q3 Datacom Growth

After a weak start to fiscal 2024 with low single digit revenue growth, Fabrinet finished its fiscal year with revenue growth reaccelerating in each quarter. Over the long-term, analysts currently estimate revenue growth to accelerate each year through fiscal 2027, as Fabrinet captures tailwinds from growth in optics.

Fabrinet reported 9% YoY growth to $2.88 billion in revenue in fiscal 2024, decelerating from 16.9% YoY growth in fiscal 2023. Much of this decline was weighted in the first half of fiscal 2024, as the broader optics industry suffered from a sharp inventory correction in the telecom industry.

Moving forward, analysts expect Fabrinet’s revenue growth to accelerate again on the back of strong datacom revenue growth, with estimates pointing to an acceleration through fiscal 2027. Fabrinet is currently estimated to report 13.4% growth in FY25 to $3.27 billion in revenue, a 5.4 percentage point acceleration, before rising to 14.1% growth in FY26 and 20.8% growth in FY27.

Here's what CEO Seamus Grady said about the long-term opportunity: “If your time horizon is much longer, I think we're in the very early stages of this. So, you can add or subtract as many various as you like, but I still think we're in the very, very early stages of this. And we're just beginning to see what this explosive growth in AI and the infrastructure that's required to power this network will do and what it will need in terms of optical interconnect.

I think it's because optical is the only way that you can get the speed and the bandwidth that you need to get the signals to move around. You just can't do it with traditional interconnect. So, I think there's a kind of a paradigm shift to optical interconnect becoming kind of almost mainstream. And you have to have optical for this. There's no other way to do it.”

On a quarterly basis, fiscal Q1 2024 was Fabrinet’s weakest quarter, with revenue growth of 4.6% YoY, decelerating 7 percentage points sequentially and 16 percentage points from the year ago quarter. This was both in part due to a quarter that was a week shorter (growth was 8% YoY when normalizing for weeks), as well as telecom revenue declining more than (28%) YoY, offsetting 161% YoY datacom growth.

Telecom headwinds continued to persist through Q2 and Q3, with declines of nearly (29%) YoY and (25%) YoY respectively. Datacom growth of 154% and 149% YoY in both quarters offset the lingering weakness in telecom, aiding revenue growth and pushing growth rates up to 10% YoY by Q3. Revenue growth accelerated nearly 5 percentage points QoQ to 14.9% in Q4, as telecom declines moderated as data center interconnect growth ticked up. Datacom remained strong in Q4, with 800G products leading growth, offset by the wind down of 100G products.

For Q1, Fabrinet expects revenue between $760 million and $780 million, for YoY growth of 12.3% at midpoint, with sequential growth in all product categories. This would represent a 2.6 percentage point deceleration at the midpoint, and a 1.1 percentage point decel at the high range to 13.8% growth. With 800G demand remaining strong, datacom is set to remain a primary driver for quarterly performance moving through 2025.

Surging Datacom Revenue Drives Optical Revenue Growth

Similar to what we discussed last month in our optics overview, Fabrinet’s management sees 800G data center transceivers as its largest growth tailwind, followed by 400 ZR and 400G transceivers as the next largest tailwinds. In just eight quarters, Fabrinet’s quarterly datacom revenue has grown nearly 3.5x, from $92.7 million in Q1 of fiscal 2023 to nearly $315 million in Q4 of fiscal 2024.

Taking a step back, growth visibly accelerated sharply in Q4 of fiscal 2023, or the June 2023 quarter, where datacom revenue surged to a record high at $192.5 million, more than doubling YoY and rising more than 50% QoQ. Management said that the “datacom growth was primarily driven by an 800-gig AI data center transceiver program for one of our customers.”

While not named, it’s likely that the customer being referenced is Nvidia, as this growth coincided with Nvidia’s breakout quarter (the July 2023 quarter) with Hopper driving more than 100% YoY and 88% QoQ revenue growth. Management explained that that AI data transceiver program “is ramping very fast, and has obviously become a meaningful contributor to our revenue and our growth rates and has really helped us to absorb the decline in the telecom business.”

They further clarified that they believe it is “very much in the early days of this [particular] program and this [broader] opportunity, very, very much in the early days. We're really just a couple of quarters into this of what we believe, as we understand, it will be a very long cycle and a very long trend.”

Four quarters later, in fiscal 2024 (June 2024 quarter, most recent reported), Fabrinet reported $314.7 million in datacom revenue, an increase of more than 63% YoY. Datacom now accounts for nearly 42% of total revenue, up from 29% a year ago.

However, Q4 saw the start of revenue deceleration in the segment as Fabrinet laps stronger comps with its ramp cycle. Datacom revenue accelerated extremely rapidly, rising from 4.4% YoY to 161.1% YoY in the span of four quarters, before hovering at ~150% YoY for three quarters in a row. While Fabrinet did not provide an exact guide for datacom revenue in Q1, it’s likely that there will be a slight deceleration sequentially as the company laps its peak growth quarter in Q1 2024.

Nvidia Jumps to 35% of Revenue

Despite passing peak growth, management remains optimistic about datacom’s opportunities, especially with core 800G customer Nvidia, which significantly boosted its purchases and relationship with Fabrinet to meet red-hot GPU demand.

Analysts pressed about the potential impacts of Nvidia’s once-rumored Blackwell delay, and if that would affect Q1’s guide, with Fabrinet CEO Seamus Grady saying that Nvidia “continue[s] to see strong demand for their products. And our understanding is that they will extend and expand production based on current GPUs to meet the demand that's there, and we're happy to continue to support them.”

Nvidia has rapidly become a core customer for Fabrinet through fiscal 2024, with its contribution to revenue nearly tripling from 2023. Nvidia was a non-significant customer through fiscal 2022, contributing anywhere from 0% to 9.99% of revenue, before contributing 12.5% of revenue in fiscal 2023 and now 35.1% in fiscal 2024 (June 2023 to June 2024).

Source: Fabrinet 10-K

In dollar terms, Nvidia’s revenue surged 206% YoY, from approximately $331 million in fiscal 2023 to $1.01 billion in fiscal 2024. Cisco remained Fabrinet’s second-largest customer, contributing 13.4% of revenue (~$386 million) in fiscal 2024, down from 15.6% in fiscal 2023 (~$413 million). Lumentum and Infinera had previously been major customers, accounting for more than 10% of revenue each in fiscal 2022 and 2023, but both fell below the 10% reporting threshold in fiscal 2024.

While Nvidia presents a strong growth opportunity, it’s also a risk, as the significant concentration in Nvidia opens up the door to a large chunk of lost revenue if Fabrinet lost Nvidia as a customer. However, management is working on additional opportunities outside of Nvidia in merchant transceivers and with hyperscalers, noting that they “really don't mind whether it's Ethernet or InfiniBand or anything else” supporting AI infrastructure buildouts, having the flexibility to work across different networking infrastructure.

Additionally, Fabrinet is one of a handful of suppliers to Nvidia, who is now expected to be expanding its supplier base in order to expand its GPU supply. Analysts from B. Riley stated earlier in October that its “latest checks indicate that Nvidia may have added another supplier for 1.6T, which it believes is Eoptolink. As such, Nvidia’s 1.6T allocations will be in question between incumbers Coherent, Innolight, Fabrinet and a newcomer in Eoptolink.”

Margins

Compared to other companies in the optics space that we covered in our prior update, Fabrinet has a much thinner gross margin profile in the 12% range, but stable and strong operating margins and a strong bottom-line.

Fabrinet has maintained its GAAP gross margin above 12% since Q1 of fiscal 2022, with some minor FX-impacted fluctuations. GAAP operating margin has steadily risen since 2020, rising from ~7% to the high 9% range in fiscal 2023 and 2024. Given the thin gross margins, this is a great example of Fabrinet’s operational leverage, to drive a 200 bp+ increase in operating margin on a <100 bp expansion in gross margin.

GAAP net margins reflect the strength of Fabrinet’s operating margin profile, with the company reporting a 10.3% GAAP net margin for fiscal 2024, expanding 70 bp from 9.6% in fiscal 2023. This is what gives Fabrinet a very strong bottom line: GAAP EPS was $8.10 in fiscal 2024, a 20.4% increase from $6.73 in fiscal 2023.

Looking ahead, GAAP earnings are expected to continue growing, with analyst estimates calling for 15.1% YoY growth to $9.32 in EPS in fiscal 2025 and 19.0% growth to $11.09 in fiscal 2026.

Balance Sheet and Cash Flows

Fabrinet also has a robust balance sheet alongside rapidly increasing cash flows.

  • Cash, equivalents and investments totaled $858.6 million at the end of fiscal 2024, up from $550.5 million at the end of fiscal 2023 due to strong operating cash flow growth.
  • Fabrinet reported zero debt at the end of fiscal 2024.
  • Operating cash flow was $83.1 million, or 11% of revenue, in Q4. For fiscal 2024, operating cash flow was $413.1 million, or 14.3% of revenue; this was an increase of nearly 94% YoY from $213.3 million, or 8.1% of revenue, in fiscal 2023. 
  • Free cash flow was $70.4 million, or 9.3% of revenue, in Q4. For fiscal 2024, free cash flow was $365.6 million, or 12.7% of revenue, increasing more than 140% YoY from $152.0 million, or 5.7% of revenue, in fiscal 2023.

Valuation

Despite its strong bottom line, Fabrinet is trading at elevated multiples relative to historic trends. This is important to track as well given the longer duration of both its top line and bottom line acceleration, with growth expected to accelerate to above 20% by FY27.

Fabrinet is trading slightly above 3x sales and 2.7x forward sales, both elevated relative to its prior highs in 2021 at ~2.2x sales and 2x forward sales. Fabrinet has also struggled to maintain multiples above 3x sales – in each of the last four times Fabrinet traded above 3x sales in 2024, it pulled back to below 2.75x to 2.25x within the next six weeks.

On the bottom line, Fabrinet is trading at a ~30x PE ratio and a 24x forward PE, both elevated compared to historical highs. Through much of 2021, Fabrinet failed to sustain a 27x PE ratio, with shares currently more than 10% above that level on a TTM basis. Fabrinet’s 5-year average PE is ~22.7x, with shares briefly trading below that level just once since AI tailwinds from Nvidia became visible in August 2023.

Conclusion

Fabrinet has caught our attention due to Nvidia’s rising revenue contribution, and ahead of Blackwell’s imminent launch this quarter. Fabrinet’s datacom revenue has been strong, and a primary driver of this recent quarterly revenue growth acceleration.

Despite management guiding for a slight sequential deceleration in fiscal Q1, Fabrinet’s revenue growth is expected to accelerate in fiscal 2025 and beyond, as the company captures tailwinds from high-data rate optics and tailwinds from Nvidia. Unlike Coherent, Lumentum, and Marvell that we previously covered in our prior optics overview, Fabrinet has a solid margin profile, a strong bottom line, and exceptional cash flow growth.

This analysis is a preview of what you can expect in our upcoming Discovery tier, which will provide additional analysis on new idea generation stocks that are not currently in the I/O Fund portfolio. We look forward to launching this tier November/December. There will be no changes to our current service tiers, rather I/O Fund Discovery is a service for those who want more new stock ideas beyond what our service currently provides. Stay tuned for more information!

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 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:

Fabrinet: Datacom, Nvidia Driving Optical Revenue Growth

Posted on by io-fund

Last month, we covered the importance of optical interconnects in linking GPUs together in clusters in our thematic deep dive “Optical Interconnects Overview: Strong Growth Expected Ahead.” Optical transceivers address bandwidth, or the speed of data transfer in the data center. As hyperscalers work to expand capacity, it’s expected there will be a surge in optic connections.

Fabrinet provides advanced optical communications components for datacom and telecom end markets, customized optics and glass fabrication, and advanced laser and other electro-mechanical parts. Fabrinet has reported consecutive quarters of strong datacom growth, stemming from one of its flagship customers, Nvidia. Fundamentally, Fabrinet possesses stable margins and strong cash flow generation, as top and bottom line growth is expected to accelerate from the low-teens to the 20% range over the next few years.

Fabrinet is clearly a beneficiary of Nvidia’s surging growth in the AI GPU market, with Nvidia’s contribution nearly tripling from 12.5% to 35.1% over the past twelve months. We’ve covered how Blackwell brings an enormous revenue opportunity for Nvidia and its suppliers, and Fabrinet is expected to be a beneficiary of Nvidia’s upcoming superchip, with long-term growth opportunities for optics in the AI data center and in 800G+ data rates.

Fabrinet reports earnings on Nov 4th.

Revenue Growth Reaccelerated in 2024; 149% Q3 Datacom Growth

After a weak start to fiscal 2024 with low single digit revenue growth, Fabrinet finished its fiscal year with revenue growth reaccelerating in each quarter. Over the long-term, analysts currently estimate revenue growth to accelerate each year through fiscal 2027, as Fabrinet captures tailwinds from growth in optics.

Fabrinet reported 9% YoY growth to $2.88 billion in revenue in fiscal 2024, decelerating from 16.9% YoY growth in fiscal 2023. Much of this decline was weighted in the first half of fiscal 2024, as the broader optics industry suffered from a sharp inventory correction in the telecom industry.

Moving forward, analysts expect Fabrinet’s revenue growth to accelerate again on the back of strong datacom revenue growth, with estimates pointing to an acceleration through fiscal 2027. Fabrinet is currently estimated to report 13.4% growth in FY25 to $3.27 billion in revenue, a 5.4 percentage point acceleration, before rising to 14.1% growth in FY26 and 20.8% growth in FY27.

Here's what CEO Seamus Grady said about the long-term opportunity: “If your time horizon is much longer, I think we're in the very early stages of this. So, you can add or subtract as many various as you like, but I still think we're in the very, very early stages of this. And we're just beginning to see what this explosive growth in AI and the infrastructure that's required to power this network will do and what it will need in terms of optical interconnect.

I think it's because optical is the only way that you can get the speed and the bandwidth that you need to get the signals to move around. You just can't do it with traditional interconnect. So, I think there's a kind of a paradigm shift to optical interconnect becoming kind of almost mainstream. And you have to have optical for this. There's no other way to do it.”

On a quarterly basis, fiscal Q1 2024 was Fabrinet’s weakest quarter, with revenue growth of 4.6% YoY, decelerating 7 percentage points sequentially and 16 percentage points from the year ago quarter. This was both in part due to a quarter that was a week shorter (growth was 8% YoY when normalizing for weeks), as well as telecom revenue declining more than (28%) YoY, offsetting 161% YoY datacom growth.

Telecom headwinds continued to persist through Q2 and Q3, with declines of nearly (29%) YoY and (25%) YoY respectively. Datacom growth of 154% and 149% YoY in both quarters offset the lingering weakness in telecom, aiding revenue growth and pushing growth rates up to 10% YoY by Q3. Revenue growth accelerated nearly 5 percentage points QoQ to 14.9% in Q4, as telecom declines moderated as data center interconnect growth ticked up. Datacom remained strong in Q4, with 800G products leading growth, offset by the wind down of 100G products.

For Q1, Fabrinet expects revenue between $760 million and $780 million, for YoY growth of 12.3% at midpoint, with sequential growth in all product categories. This would represent a 2.6 percentage point deceleration at the midpoint, and a 1.1 percentage point decel at the high range to 13.8% growth. With 800G demand remaining strong, datacom is set to remain a primary driver for quarterly performance moving through 2025.

Surging Datacom Revenue Drives Optical Revenue Growth

Similar to what we discussed last month in our optics overview, Fabrinet’s management sees 800G data center transceivers as its largest growth tailwind, followed by 400 ZR and 400G transceivers as the next largest tailwinds. In just eight quarters, Fabrinet’s quarterly datacom revenue has grown nearly 3.5x, from $92.7 million in Q1 of fiscal 2023 to nearly $315 million in Q4 of fiscal 2024.

Taking a step back, growth visibly accelerated sharply in Q4 of fiscal 2023, or the June 2023 quarter, where datacom revenue surged to a record high at $192.5 million, more than doubling YoY and rising more than 50% QoQ. Management said that the “datacom growth was primarily driven by an 800-gig AI data center transceiver program for one of our customers.”

While not named, it’s likely that the customer being referenced is Nvidia, as this growth coincided with Nvidia’s breakout quarter (the July 2023 quarter) with Hopper driving more than 100% YoY and 88% QoQ revenue growth. Management explained that that AI data transceiver program “is ramping very fast, and has obviously become a meaningful contributor to our revenue and our growth rates and has really helped us to absorb the decline in the telecom business.”

They further clarified that they believe it is “very much in the early days of this [particular] program and this [broader] opportunity, very, very much in the early days. We're really just a couple of quarters into this of what we believe, as we understand, it will be a very long cycle and a very long trend.”

Four quarters later, in fiscal 2024 (June 2024 quarter, most recent reported), Fabrinet reported $314.7 million in datacom revenue, an increase of more than 63% YoY. Datacom now accounts for nearly 42% of total revenue, up from 29% a year ago.

However, Q4 saw the start of revenue deceleration in the segment as Fabrinet laps stronger comps with its ramp cycle. Datacom revenue accelerated extremely rapidly, rising from 4.4% YoY to 161.1% YoY in the span of four quarters, before hovering at ~150% YoY for three quarters in a row. While Fabrinet did not provide an exact guide for datacom revenue in Q1, it’s likely that there will be a slight deceleration sequentially as the company laps its peak growth quarter in Q1 2024.

Nvidia Jumps to 35% of Revenue

Despite passing peak growth, management remains optimistic about datacom’s opportunities, especially with core 800G customer Nvidia, which significantly boosted its purchases and relationship with Fabrinet to meet red-hot GPU demand.

Analysts pressed about the potential impacts of Nvidia’s once-rumored Blackwell delay, and if that would affect Q1’s guide, with Fabrinet CEO Seamus Grady saying that Nvidia “continue[s] to see strong demand for their products. And our understanding is that they will extend and expand production based on current GPUs to meet the demand that's there, and we're happy to continue to support them.”

Nvidia has rapidly become a core customer for Fabrinet through fiscal 2024, with its contribution to revenue nearly tripling from 2023. Nvidia was a non-significant customer through fiscal 2022, contributing anywhere from 0% to 9.99% of revenue, before contributing 12.5% of revenue in fiscal 2023 and now 35.1% in fiscal 2024 (June 2023 to June 2024).

Source: Fabrinet 10-K

In dollar terms, Nvidia’s revenue surged 206% YoY, from approximately $331 million in fiscal 2023 to $1.01 billion in fiscal 2024. Cisco remained Fabrinet’s second-largest customer, contributing 13.4% of revenue (~$386 million) in fiscal 2024, down from 15.6% in fiscal 2023 (~$413 million). Lumentum and Infinera had previously been major customers, accounting for more than 10% of revenue each in fiscal 2022 and 2023, but both fell below the 10% reporting threshold in fiscal 2024.

While Nvidia presents a strong growth opportunity, it’s also a risk, as the significant concentration in Nvidia opens up the door to a large chunk of lost revenue if Fabrinet lost Nvidia as a customer. However, management is working on additional opportunities outside of Nvidia in merchant transceivers and with hyperscalers, noting that they “really don't mind whether it's Ethernet or InfiniBand or anything else” supporting AI infrastructure buildouts, having the flexibility to work across different networking infrastructure.

Additionally, Fabrinet is one of a handful of suppliers to Nvidia, who is now expected to be expanding its supplier base in order to expand its GPU supply. Analysts from B. Riley stated earlier in October that its “latest checks indicate that Nvidia may have added another supplier for 1.6T, which it believes is Eoptolink. As such, Nvidia’s 1.6T allocations will be in question between incumbers Coherent, Innolight, Fabrinet and a newcomer in Eoptolink.”

Margins

Compared to other companies in the optics space that we covered in our prior update, Fabrinet has a much thinner gross margin profile in the 12% range, but stable and strong operating margins and a strong bottom-line.

Fabrinet has maintained its GAAP gross margin above 12% since Q1 of fiscal 2022, with some minor FX-impacted fluctuations. GAAP operating margin has steadily risen since 2020, rising from ~7% to the high 9% range in fiscal 2023 and 2024. Given the thin gross margins, this is a great example of Fabrinet’s operational leverage, to drive a 200 bp+ increase in operating margin on a <100 bp expansion in gross margin.

GAAP net margins reflect the strength of Fabrinet’s operating margin profile, with the company reporting a 10.3% GAAP net margin for fiscal 2024, expanding 70 bp from 9.6% in fiscal 2023. This is what gives Fabrinet a very strong bottom line: GAAP EPS was $8.10 in fiscal 2024, a 20.4% increase from $6.73 in fiscal 2023.

Looking ahead, GAAP earnings are expected to continue growing, with analyst estimates calling for 15.1% YoY growth to $9.32 in EPS in fiscal 2025 and 19.0% growth to $11.09 in fiscal 2026.

Balance Sheet and Cash Flows

Fabrinet also has a robust balance sheet alongside rapidly increasing cash flows.

  • Cash, equivalents and investments totaled $858.6 million at the end of fiscal 2024, up from $550.5 million at the end of fiscal 2023 due to strong operating cash flow growth.
  • Fabrinet reported zero debt at the end of fiscal 2024.
  • Operating cash flow was $83.1 million, or 11% of revenue, in Q4. For fiscal 2024, operating cash flow was $413.1 million, or 14.3% of revenue; this was an increase of nearly 94% YoY from $213.3 million, or 8.1% of revenue, in fiscal 2023. 
  • Free cash flow was $70.4 million, or 9.3% of revenue, in Q4. For fiscal 2024, free cash flow was $365.6 million, or 12.7% of revenue, increasing more than 140% YoY from $152.0 million, or 5.7% of revenue, in fiscal 2023.

Valuation

Despite its strong bottom line, Fabrinet is trading at elevated multiples relative to historic trends. This is important to track as well given the longer duration of both its top line and bottom line acceleration, with growth expected to accelerate to above 20% by FY27.

Fabrinet is trading slightly above 3x sales and 2.7x forward sales, both elevated relative to its prior highs in 2021 at ~2.2x sales and 2x forward sales. Fabrinet has also struggled to maintain multiples above 3x sales – in each of the last four times Fabrinet traded above 3x sales in 2024, it pulled back to below 2.75x to 2.25x within the next six weeks.

On the bottom line, Fabrinet is trading at a ~30x PE ratio and a 24x forward PE, both elevated compared to historical highs. Through much of 2021, Fabrinet failed to sustain a 27x PE ratio, with shares currently more than 10% above that level on a TTM basis. Fabrinet’s 5-year average PE is ~22.7x, with shares briefly trading below that level just once since AI tailwinds from Nvidia became visible in August 2023.

Conclusion

Fabrinet has caught our attention due to Nvidia’s rising revenue contribution, and ahead of Blackwell’s imminent launch this quarter. Fabrinet’s datacom revenue has been strong, and a primary driver of this recent quarterly revenue growth acceleration.

Despite management guiding for a slight sequential deceleration in fiscal Q1, Fabrinet’s revenue growth is expected to accelerate in fiscal 2025 and beyond, as the company captures tailwinds from high-data rate optics and tailwinds from Nvidia. Unlike Coherent, Lumentum, and Marvell that we previously covered in our prior optics overview, Fabrinet has a solid margin profile, a strong bottom line, and exceptional cash flow growth.

This analysis is a preview of what you can expect in our upcoming Discovery tier, which will provide additional analysis on new idea generation stocks that are not currently in the I/O Fund portfolio. We look forward to launching this tier November/December. There will be no changes to our current service tiers, rather I/O Fund Discovery is a service for those who want more new stock ideas beyond what our service currently provides. Stay tuned for more information!

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

Recommended Reading:

AMD Q3 Earnings Preview: AI Revenue Estimated to be $5 Billion

Posted on by io-fund

AMD will release its Q3 2024 results tomorrow. Analysts expect Q3 revenue to grow 15.8% YoY to $6.71 billion and adjusted EPS to grow 30.8% YoY to $0.92.

There are three primary growth segments for AMD. In data center CPUs, AMD has been a force to reckon with and continues to stomp on Intel. The most recent numbers for data center market share is at 34%, up from 4% in March of 2020 when our firm first predicted AMD was setting up for an epic comeback. The Turin EPYC processors were previewed in June and are expected to ramp with broad availability into the second half of the year.

Secondly, given Nvidia sparked an AI boom, the market is hyper focused on AI accelerators, and thus, there is outsized pressure on AMD to raise its revenue on MI300 GPUs. The Street whisper is that AMD will raise this number to $5 billion for this year (up from $4.5 billion), with some analysts predicting AMD will report $10 billion in GPU revenue next year. Regarding analyst expectations of $5 billion this year, this aligns with AMD’s commentary: “When you look at the second half we will continue to see Data Center to be the major driver of our top-line revenue growth” and expectations data center will grow sequentially. Last quarter, data center grew 115% year-over-year in Q2.

If we assume the $5 billion will be guided in tomorrow’s call, this equates to 19.5% of revenue this year, and if we assume the $10 billion materializes, it means AMD will end 2025 with 30% of the company’s revenue from AI. Management has already pointed out the MI300 series marks the fastest ramp in company history, and given they are sitting on tech’s second-best comeback of all time with EPYC processors, this is not a casual statement to make. For perspective, Broadcom will end this year with 23% in AI revenue, if we take their current fiscal year guide. Recently, AMD moved to an annual cadence for its GPUs with the MI325X expected to ship in Q4 2024 and MI350 expected in H2 2025.

Investors who own AMD are contending with market psychology around what it means to be second place in the highly competitive industry of tech. Additionally, AMD is known for undercutting on price, and thus, the MI300s are expected to be dilutive to margins, whereas Nvidia’s H100s and H200s have provide historic margins. Long-term, management has emphasized that GPUs will be accretive and above the corporate average (see paragraph below). With that said, for this report, AMD is expected to report an expansion in adjusted operating margins from 22% last quarter to 25% this quarter due to EPYC CPUs helping product mix. If reported, it will mark the highest adjusted operating margin in two years.

The third point to consider going into tomorrow’s report, is that it’s widely expected that client-facing semiconductor segments will be softer than expected this quarter. We’ve seen evidence of this in ASML, Texas Instruments and ON Semi. For AMD, this refers to the company’s PC exposure. We covered in the webinar the PC data the I/O Fund is closely tracking, which shows Q3 being meaningfully softer than expected, and that ultimately PCs will miss the 2024 estimates that industry analysts had going into this year. Last quarter, Client revenue was $1.492 billion, up 49% YoY and up 9% QoQ. Management stated at the time: “In summary, we delivered strong second quarter results and are well positioned to grow revenue significantly in the second half of the year, driven by our data center and client segments.”

There is QoQ growth in PCs from Q2, yet within the industry estimates for Q3 across the PC industry, YoY is negative. AMD has a strong CFO (which helps with guiding correctly) and the company incrementally stronger than its peers on PCs. There is a scenario where AMD squeaks by unscathed by PC softness and its peers do not, as the company recently released the Zen 5 architecture including the Ryzen AI 300 laptops with a neural processing unit (NPU) with 50 TOPS of AI performance, and the Ryzen 9000 series for desktops. Management was positive about this upcoming quarter, stating to expect “above-typical seasonality given the strength of our product launches.”

Canalys has the most optimistic estimates for PCs, whereas IDC and Gartner are negative on a YoY basis at (-2.4%) and (-1.3%) respectively.

Revenue

Revenue is expected to accelerate in the coming quarters, driven by strong demand for data center CPUs and GPUs, coupled with the expected rebound in the Client segment.

  • Analysts expect Q3 revenue to grow 15.8% YoY to $6.71 billion and accelerate to 22.3% in Q4. Revenue is expected to further accelerate to 33.5% in Q1 2025.
  • Management revenue guide is $6.7 billion, representing YoY growth of 15.5% at the midpoint. This represents a 6.6-point acceleration from 8.9% growth to $5.84 billion in Q2.

Lisa Su said in the Q2 earnings call, “We delivered strong second quarter results and are well positioned to grow revenue significantly in the second half of the year, driven by our data center and client segments. Our data center GPU business is on a steep growth trajectory as shipments ramp across an expanding set of customers. We're also seeing strong demand for our next generation Zen 5 EPYC and Ryzen processors that deliver leadership performance and efficiency in both data center and client workloads.”

  • Analysts expect 2024 revenue to grow 12.9% YoY to $25.61 billion and accelerate to 28.3% growth in 2025.
  • 2026 revenue is expected to grow 19.9% YoY to $39.38 billion.

Margins:

AMD’s margins are benefiting from a higher mix of data center revenue. This quarter is expected to report the highest adjusted operating margin in two years.

  • Q2 gross margin was 49%, compared to 46% last year. Adjusted gross margin improved to 53% from 50% in the same period last year, helped by a higher portion of data center revenue. Management guide for Q3 is 53.5%, up from 51% in Q3 2023.
  • Q2 adjusted operating margin was 22%, compared to 20% in the same period last year. Operating expenses increased 15% YoY to $1.8 billion due to higher R&D expenses required to address AI growth. Management guide for Q3 is 25%, up from 22% in Q3 2023.
  • Q2 net income was $265 million or 5% of revenue compared to $27 million or 1% of revenue in Q2 2023. Adjusted net income was $1.13 billion or 19% of revenue compared to $948 million or 18% of revenue in the same period last year.

EPS

EPS is expected to grow significantly in the coming quarters.

  • Analysts expect Q3 adjusted EPS to grow 30.8% YoY to $0.92, by 50.2% to $1.16 in Q4, and by 76.6% to $1.09 in Q1 2025.
  • Analysts expect 2024 adjusted EPS to grow 27.9% YoY to $3.39 and accelerate to 59.7% growth to $5.41 in 2025.
  • Analysts expect 2026 adjusted EPS to grow 36% YoY to $7.36.

Cash Flow and Balance Sheet

The company has increasing cash flows with room for improvement. At its peak in 2021, AMD reported over 25% cash flow margins.

  • Q2 operating cash flow was $593 million or 10% of revenue compared to 7% in Q2 2023.
  • Free cash flow was $439 million or 8% of revenue compared to 5% in Q2 2023.
  • Inventories were $4.99 billion, compared to $4.65 billion in Q1. They were up primarily due to the continued ramp-up of a data center GPU product.
  • Cash and short-term investments were $5.34 billion, and debt of $1.72 billion, compared to $6.04 billion and $2.47 billion in Q1. The company repaid $750 million in debt that matured in June with existing cash. It repurchased shares worth $352 million, with $5.2 billion of share authorization remaining.
  • The company completed the Silo AI acquisition in August for about $665 million in cash. AMD also announced the acquisition of ZT Systems for $4.9 billion, which will close in the second half.

Segments

Data Center revenue grew 115% YoY and 21% QoQ to $2.83 billion. The company reported record data center revenue in Q2, accounting for 49% of revenue. The company witnessed strong demand for AMD Instinct GPUs and double-digit EPYC server revenue growth. Data Center is expected to continue to be a major driver of top-line growth in the second half of the year. Data Center’s operating margin was 26% compared to 11% in the same period last year.

Client segment revenue grew by 49% YoY and 9% QoQ to $1.49 billion. It was primarily helped by strong demand for Ryzen processors and initial shipments of the next-generation Zen 5 processors. The client segment is expected to grow sequentially in Q3, and management said in the Q2 earnings call that the customer response for the new Ryzen processors was strong, and it is expected to capture additional revenue market share (see below). Operating margin was 6% compared to (-7%) in the same period last year.

The gaming segment continues to struggle due to soft demand. Revenue declined by (-59%) YoY and (-30%) sequentially to $648 million. Management expects gaming segment revenue to decline double digits sequentially in Q3. “Semi-custom demand remains soft, as we are now in the fifth-year of the console cycle and we expect sales to be lower in the second half of the year compared to the first half.” The operating margin was 12% compared to 14% in Q2 2023.

Embedded segment revenue declined by (-41%) YoY and was up 2% sequentially to $861 million. Management observed initial signs of improving order trends and expect embedded revenue to gradually recover in the second half of the year, with revenue expected to be up sequentially in Q3. Operating margin was 40% compared to 52% in the same period last year.

Other Key Points

AI Revenue

AMD’s AI accelerator, the MI300, is the fastest-ramping product in AMD’s history. The company reported over $1 billion in MI300 revenue in Q2. Management expects MI300 revenue to ramp in Q3 and Q4. During Q2 results, the company raised the data center GPU guide from $4 billion to $4.5 billion for the year 2024. The company also cited in the earnings call that Microsoft was the first hyperscaler to announce the general availability of MI300X instances.

“Turning to our data center AI business, we delivered our third straight quarter of record data center GPU revenue with MI300 quarterly revenue exceeding $1 billion for the first time. Microsoft expanded their use of MI300X Accelerators to power GPT-4 Turbo and multiple co-pilot services including Microsoft 365 Chat, Word, and Teams. Microsoft also became the first large hyperscaler to announce general availability of public MI300X instances in the quarter.”

Strong Product Roadmap

The company announced earlier this year its expanded AMD Instinct accelerator roadmap and annual cadence for chip release. During the recent Advancing AI Event, the company also confirmed that MI325X chips are expected to be shipped in Q4 2024 and the launch of MI350 chips in the second half of next year.

Lisa Su said in the Q2 earnings call, “Looking ahead from a roadmap perspective, we are accelerating and expanding our Instinct roadmap to deliver an annual cadence of AI accelerators, starting with the launch of MI325X later this year. MI325X leverages the same infrastructure as MI300 and extends our generative AI performance leadership by offering twice the memory capacity and 1.3 times more peak compute performance than competitive offerings. We plan to follow MI325X with the MI350 series in 2025 based on the new CDNA 4 architecture, which is on track to deliver a 35x increase in performance compared to CDNA 3. And our MI400 series powered by the CDNA “Next” architecture is making great progress in development and is scheduled to launch in 2026.”

Lisa Su predicts that AI Data Center Accelerators TAM to reach $500 billion by 2028 growing at a CAGR of 60% from $45 billion in 2023, compared to the earlier prediction last December of reaching $400 billion in 2027. She also highlighted the strong MI300X performance during the Advancing AI Event. “If you look today at MI300x performance, we have more than doubled our inferencing performance and significantly improved our training performance on the most popular models. Today, over 1 million models run seamlessly out of the box on Instinct, and that's more than 3x the number when we launched in December.”

She further pointed out, “MI300X consistently outperforms the competition, which is H100 in inferencing. So, for example, using Llama 3.1 405B, which is one of the most newest and demanding models out there, MI300 outperforms H100 with the latest optimizations by up to 30% across a wide variety of use cases.”

I/O Fund note: It would be stronger to benchmark the MI300X against the H200s but the competition in releases is likely to become tighter with each generation. The MI325 is due out this quarter, and thus AMD is about two quarters behind Nvidia’s H200.

AI Software

AMD completed the acquisition of Silo AI in August. Silo AI specializes in large language model development, which will further enhance AMD's AI inference and training tools. The acquisition will also help to tap the talent pool of engineers and scientists of Silo AI who have used AMD hardware and provide customized AI solutions to its clients.

AMD announced in August that it would acquire ZT Systems for $4.9 billion. The deal is expected to boost data center AI solutions. Once the deal closes, AMD plans to sell the ZT Systems manufacturing business. The acquisition is expected to close in the first half of 2025 and be accretive on a non-GAAP basis by the end of 2025.

AI PCs and Zen 5 EPYC Processors

At the Advancing AI Event, Lisa Su discussed the success of EYPC CPUs since their launch in 2017. She pointed out that “EYPC has become the CPU of choice for the modern data center.” The cloud providers offer more than 950 EPYC instances, and on the enterprise side, larger server OEMS offer over 350 EPYC platforms, increasing the company’s server CPU market share to 34%.

Recently, AMD launched the 5th Gen AMD EYPC CPUs, formerly codenamed Turin. They are suited for cloud, enterprise, and AI use cases. They use the advanced 3nm/4nm process technology. The new Zen 5 core architecture, provides up to 17% better instructions per clock (IPC) for enterprise and cloud workloads and up to 37% higher IPC in AI and high-performance computing (HPC) compared to Zen 4 architecture.

During the Q2 earnings call, Lisa Su also said that the Client segment is also expected to do well in the second half of the year along with Data Center segment. It is expected to be above seasonal due to the launch of new products. The new Ryzen AI 300 laptops and the Ryzen 9000 series for desktops are powered by the 5th generation of the Zen architecture. The Ryzen AI 300 laptop has a XDNA 2 neural processing unit (NPU) that is designed for Microsoft Copilot+ AI software. This will deliver 50 TOPS of AI performance, exceeding Apple’s M4.

“We are launching Zen 5 desktops and notebooks with volume ramping in the third quarter. And that’s the primary reason that we see above-seasonal. The AI PC element is certainly 1 element of that, but there is just the overall refresh. Usually, desktop launches going into a third quarter are good for us, and we feel that the products are very well positioned. So those are the primary reasons.”

Note on GPU Margins:

It’s prudent to make a note that another area where AMD is not keeping pace with Nvidia is pricing power, leading to GPU margins that are currently below the corporate average. Here was a statement from management when questioned on the margins last quarter:

“Yes. On your second question about the profitability, first our team has done a tremendous job to ramp the product MI300. It is a very complex product. So we ramped it successfully. At the same time, the team also started to implement operational optimization to continue to improve gross margin. So we continue to see the gross margin improvement. Over time, in the longer term, we do believe gross margin will be accretive to corporate average.”

Valuation

The company trades at a P/E ratio of 185.5 and a forward P/E ratio of 44.9.

P/S ratio is 11.5 compared to the five-year average of 8.6. The forward P/S is 10 and the 1-year forward is 7.8.

Conclusion

Fundamentally, AMD is quite strong due to the continued strength in CPUs, the ongoing growth in GPUs, and the expected rebound in the Client Segment. We feel AMD is a win-win for our portfolio. Should the company beat, we will participate. If the company misses or something more broad weighs on the company (such as tariffs on semiconductors) then we will gladly buy shares lower.

We like Lisa Su reiterating the 2027 time frame for a $400B TAM, and increasing the estimate to $500B the following year for 2028. We will match that timeline and say we hope to see AMD be a leader in the market and in our portfolio by 2027-2028. CDNA 4 architecture is due out in 2025-2026, and is the most likely catalyst that I see today to narrow the product road map with Nvidia.

Royston Roche, Equity Analyst at the I/O Fund, contributed to this article. Beth Kindig, Lead Tech Analyst, contributed to this article.

Recommended Reading:

This Stock Is Crushing Salesforce, MongoDB And Snowflake In AI Revenue

Posted on by io-fund

This article was originally published on Forbes on Updated Oct 17, 2024, 09:02pm EDTForbesForbes on Updated Oct 17, 2024, 09:02pm EDT

Palantir has been one of the top-performing AI software stocks this year with a 156% YTD return, thanks to accelerating revenue growth and strong business momentum from its Artificial Intelligence Platform (AIP) released last year.

AIP sets Palantir apart from the rest of the SaaS universe, driving visible AI-related growth and acceleration in multiple different metrics – at this time, other leading AI favorites such as Snowflake or MongoDB can’t say the same. Outside of the cloud hyperscalers, Palantir is one of the rare few that sees AI drive both real returns for its business and real value for its customers due to AIP.

Below, I break down how Palantir’s AIP is putting it a step above peer Salesforce, MongoDB and Snowflake with visible AI growth, and its undeniable ‘secret sauce’.

Palantir’s AI Growth is Visible

AIP has driven tremendous growth for Palantir’s business since its release, with primary impacts arising in the commercial segment. A clear inflection point in Palantir’s growth is visible following AIP’s release, while other ‘AI’ cloud peers can’t say the same about AI-driven growth.

Palantir said that “US commercial continues to accelerate in Q2 2024 alongside [the] AIP revolution” with “unprecedented demand”, and the numbers to back this up:

  • 55% YoY revenue growth in US commercial to $159 million, accelerating from 40% YoY in Q1.
Palantir US Commercial Revenue

US Commercial revenue growth accelerated to 55% YoY in Q2 as revenue rose to $159 million.

Source: I/O Fund

  • 83% YoY growth in US commercial customers to 295 and 98% YoY growth in US commercial deals closed to 123.
  • 103% YoY growth in US commercial remaining deal value and 152% YoY growth in US commercial total contract value to $262 million. Chief Revenue Officer Ryan Taylor explained that “one of the most notable indicators of our delivery is the volume of existing customers who are signing expansion deals, many of which are a direct result of AIP.”

Here’s what the growth in US commercial customers looks like:

US Commercial Customer Count

Palantir's US Commercial customer growth has reaccelerated over the past few quarters thanks to AIP.

Source: I/O Fund

US commercial customer growth began to stagnate through late 2022 and early 2023, but following AIP’s release in Q2 2023, customer count re-accelerated. There is a clear inflection point from where QoQ customer additions were decelerating – from 12 net adds in Q1 2023 to six net adds in Q2 2023. Following the AIP-driven acceleration, net adds rose to 20 QoQ in Q3 2023, then 40 QoQ in Q4 2023.

This matches a similar acceleration in commercial customer growth as Palantir quickly became a market darling following its IPO, which was seen as a way to drive growth in the commercial sector. From Q4 2020 to Q4 2021, commercial customers grew nearly 5X. Now, as a stock market darling once more with a unique and unbeatable AI offering, Palantir is seeing commercial growth resume.

Sign up for I/O Fund's free newsletter with gains of up to 2600% because of Nvidia's epic run – Click hereClick hereClick here

Palantir is King of AI Among Cloud SaaS Stocks

Other leading cloud ‘AI’ stocks are struggling to put up AI-driven growth numbers like Palantir.

Salesforce reported 8% YoY revenue growth in Q2, decelerating from 11% YoY in Q1, as subscription revenue growth decelerated to 9% YoY, down from 12% YoY in Q1. Salesforce sees Q3 revenue growth of 7%, another deceleration. The full-year revenue growth of just 8% to 9% translates to the SaaS giant struggling to realize AI gains. Furthermore, Salesforce’s more AI-aligned offerings, MuleSoft and Tableau, decelerated sharply in Q2, from 27% YoY to 13% YoY for MuleSoft and 21% YoY to 11% YoY for Tableau.

MongoDB witnessed a much steeper deceleration in Q2, as Atlas and new workload wins struggled at the start of the year. In Q1, MongoDB reported 22% YoY growth with Atlas growth of 32% YoY, and this decelerated to 13% YoY revenue growth in Q2 as Atlas declined 5 percentage points QoQ to 27% YoY. For the full year, MongoDB guided to about 14.6% YoY growth in Q2 as it slightly boosted its outlook, a steep deceleration from 31% YoY growth in fiscal 2024.

Snowflake’s product revenue growth decelerated from 34% YoY in Q1 to 30% YoY in Q2, and while this was ahead of its guidance by 3 percentage points, growth is set to decelerate further in Q3. Management guided for 22% YoY growth in product revenue for the third quarter, a steeper QoQ deceleration rate, with the full year product revenue guide of 26% YoY. Despite management saying that they see “great traction” in early stages of AI products, there’s no visible inflection or acceleration in growth.

In sharp contrast, AIP has helped Palantir drive a significant topline acceleration over the past four quarters.

Palantir Quarterly Revenue Growth, YoY

AIP's strong momentum has helped drive quarterly revenue growth to 27.2% YoY in Q2 2024, up from 12.7% in Q2 2023.

Source: I/O Fund

Palantir reported 27.2% YoY revenue growth in Q2, aided by strength in US commercial stemming from AIP as well as government revenue accelerating significantly. Palantir’s YoY revenue growth bottomed in Q2 2023 at 12.7%, the same quarter as AIP’s release, with revenue growth now 15 points higher. Despite guiding for a slight 2 percentage point deceleration in Q3 to 25.2% YoY growth, Palantir would only need to beat its guide by 1.5% to keep this revenue acceleration intact.

Fundamentally, what’s most critical for shares is maintaining a revenue growth rate above 20% for the foreseeable future – analysts currently estimate fiscal Q2 2025 to be the one quarter of the next eight with revenue growth just below that threshold. Given AIP’s strength just one year following its launch, with clear inflections in customer and revenue growth, it will be the telling sign of Palantir’s AI status if it can maintain these revenue growth rates as it scales.

Palantir’s AIP Separates it From the SaaS Universe

Palantir’s standout performance so far in 2024 against SaaS peers can be attributed to the success of AIP, which, at its core, is a comprehensive AI platform that lets enterprises lever Palantir’s AI and machine learning tools and harness the power of the latest large language models (LLMs) within Foundry and Gotham.

Gotham was the company’s first product and is built for government operatives in defense and intelligence sectors. The platform enables users to identify patterns hidden deep within datasets using semantic, temporal, geospatial and full-text analysis, with mixed reality capabilities to allow operations to be run in virtual environment as well. Graph allows data objects to be seen as nodes and edges, while Map track geo-located objects, run searches and display key data.

Foundry was built for the commercial sector, and is centered around the three-layer Ontology Core, integrating semantic, kinetic, and dynamic layers for real-time data analytics and AI-powered decision making capabilities:

  • The Semantic layer brings volumes of data into one place, and lets users generate detailed object properties
  • The Kinetic layer brings operations and business behaviors into a real-time graph linked back to the Semantic layer, creating the basis for AI-driven analytics, real-time monitoring, identification of inefficiencies, and ability to optimize workflows
  • The Dynamic layer connects models to objects and actions, reasoning across both the Semantic and Kinetic layers for AI-powered automation and AI-driven decision making, alongside multi-step simulations with AI predictive analytics to explore possibilities of changing actions or events

AIP combines with Foundry’s data operations suite and Apollo’s autonomous software deployment capabilities as part of Palantir’s ‘AI Mesh’, providing enterprise and government customers with a full suite of AI products from the web to mobile to the edge. With the Ontology, linking data and logic into an AI-accessible environment, Palantir brings generative AI directly to an enterprise’s operations, delivering real-time AI-driven operational decision-making abilities.

Palantir describes Gotham and Foundry as the “ability to construct a model of the real world from countless data points.” AIP links this all together, and this is what separates Palantir as a standout in the SaaS space — outside of the cloud hyperscalers, Palantir is one of the rare few that sees AI drive both real returns for its business and real value for its customers due to AIP.

What further sets AIP apart is its scalability, interoperability and versatility. With AI Mesh, organizations can integrate AI across different operations and applications, while its design facilitates interoperability with existing enterprise software and systems. AIP is also extremely versatile, having been successfully and seamlessly integrated into enterprises spanning a wide range of industries from tech to healthcare to aerospace, while still driving value to customers.

The uniqueness of Palantir’s AIP and value that it can quickly provide has driven growth for the company. CEO Alex Karp said in Q2 that “growth across the commercial and government markets has been driven by an unrelenting wave of demand from customers for artificial intelligence systems that go beyond the merely performative and academic.”

Essentially, there is constant strong demand for an applicable, scalable, versatile AI platform that can drive real-time results with an instant value-add for an organization. Chief Revenue Officer Ryan Taylor added that Q2’s “exceptional results are a reflection of a market that is quickly awakening to a reality that our customers have already known, we stand alone in our ability to deliver enterprise AI production impact at scale.”

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock entries and exits. We offer trade alerts plus an automated hedging signal. The I/O Fund team is one of the only audited portfolios available to individual investors. Learn more here.Learn more hereLearn more here.

Government is Palantir’s Secret Sauce

While Palantir is undoubtedly seeing strong business momentum in the commercial sector, the government sector remains Palantir’s bread and butter, being that the government sector has funded the company and allowed it to aggressively invest in AIP while expanding margins, with a recent growth acceleration

In Q2, US government revenue accelerated to 24% YoY growth, up 12 percentage points from 12% YoY growth in Q1. Overall, government revenue growth was 23% YoY, up 7 percentage points from 16% YoY in Q1. Management noted that Palantir was “selected for several notable awards in Q2, which led to the strongest US government bookings quarter since 2022, reflecting the growing demand for our government software offerings.”

This included a production contract from the DoD, Chief Digital and Artificial Intelligence Office (CDAO) for an AI-enabled operating system for the DoD, with an initial $153 million order and additional awards for up to $480 million over a 5-year period.

The acceleration in the government segment aided overall revenue growth in the quarter, as the government continues to remain Palantir’s primary revenue source, accounting for nearly 55% of total revenue. This is why the government segment is vital for Palantir, and is its ‘secret sauce’ – these long-term, high-value government contracts provide consistent and recurring revenue and financial stability, allowing the company to venture and invest to scale AIP while expanding margins and increasing its profitability.

Conclusion

Palantir has been on a tear this year, and is outperforming major cloud competitors, thanks to the strength and uniqueness of its AIP offering. Palantir has the best of both worlds in government contracts and AI exposure, as well as accelerating enterprise AI adoption and strong customer and revenue growth.

The one caveat is Palantir’s valuation, at 34x FY24 revenue and 29x FY25 revenue, is increasingly challenging to sustain. In the past, the low 20x revenue multiple range has tended to be the level that even the industry’s leading SaaS names have struggled to break past over the last few years.

Given the outsized valuation, the I/O Fund is looking for a lower entry in Palantir before adding the stock to our portfolio. Join the I/O Fund’s next webinar on Thursday, October 24th where Knox Ridley, Technical Analyst, will discuss the firm’s buy zones and targets for AI leaders. Learn more here.

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:Recommended Reading:

Nova and Onto Innovation: Growth in Metrology and Semiconductor Process Control

Posted on by io-fund

While Nvidia’s dominance in AI GPUs is more clearly visible downstream in the HBM market, it’s also driving growth in a lesser-known but equally as important market, in metrology/process control equipment.

Metrology, by definition, is the study of measurement. In the semiconductor industry, metrology plays a critical role, as the ability to measure circuit dimensions, microstructures inside chips, film thickness, accuracy of layers and overlays are all necessary to ensure chipmakers can consistently produce high-quality, flaw-free chips at high yields.

While KLA dominates the metrology market, reportedly holding a more than 50% market share for over 20 years, Onto Innovation and Nova are two smaller players in the market putting up accelerating revenue growth with strong margins, with exposure to AI megatrends and long-term growth drivers.

Metrology Has Multiple Drivers

The metrology market has multiple drivers of future growth, and while it’s not limited to the AI data center mega-trends spearheaded by Big Tech, there are many levers within this AI buildout to drive demand for Onto Innovation and Nova.

This is how Nova’s management team recently described the opportunity: “Artificial intelligence is reshaping our digital world and expediting the implementation of scaling agents in integrated circuits to boost performance and power efficiency which are at the core of enabling large language models through training and infra stages. Scaling is manifested in more complex designs through three avenues, dimensions, materials and advanced packaging.

It is also manifested in the size of manufactured dies and in the number of dies required in solutions such as high bandwidth memory. These developments profoundly impact our industry in the number of the wafers needed to supply demand and in the yield ratio, you must obtain to secure profitability with a lower number of dies per wafer. Couple that with the growing complexity of architecture and materials and you have the growth drivers of process control in the coming years.”

Here’s a brief look at how AI and the data center are generating demand for metrology equipment across these different drivers:

Shift to Advanced Nodes

Advanced nodes require more process steps as node sizes shrink, so for a chipmaker or foundry like TSMC or Intel to move from primarily producing on 5nm nodes to 2nm or below over the next couple of years, there will be a greater need for metrology equipment. Currently, the 5nm and 4nm nodes are primarily being used for AI chip production, such as that for Nvidia’s Hopper and Blackwell chips, while 3nm production is ramping at TSMC with volume production at the 2nm node expected in 2025, primarily for smartphone applications.

This is because the manufacturing tolerances shrinks as nodes shrink in size – the chipmaking process now becomes increasingly more sensitive to minute deviations in the process. Moving to more advanced nodes warrants much greater precision throughout the entire manufacturing process, as the smallest of deviations could greatly affect the process’ yield.

For example, Samsung has reportedly been struggling with severe yield issues for its most advanced nodes, with unstable yields reportedly around 50% or below for its gate all-around GAA 3nm second-gen process, and yields as low as 20% for its 2nm process.

TSMC sees higher yields for the 3nm process, reportedly between 60% and 70%, allowing it to command a majority of production on advanced nodes and thus a majority of market share.

Source: Nova

Rising Chip Complexity

Metrology needs also rise as chip designs get increasingly complex, from the entire GPU to memory cubes.

AI accelerators are getting larger and more complex with each generation packing more transistors than the last. Nvidia’s upcoming B200 GPU packs 208 billion transistors and 192GB of HBM3e memory, versus 80 billion transistors and 80GB HBM3 memory for the H100. Though Blackwell is in a dual-die configuration, it’s still a more complicated chip to produce using TSMC’s customized N4P process (the same as for the H100).

Having a more complex chip design increases the chances of errors, as there are more steps in the manufacturing process to produce said chips. This also goes for DRAM and NAND cubes – as HBM cubes stack higher, from 8-high cubes to 12-high cubes for the current HBM3e generations from SK Hynix, Samsung and Micron, it also increases process steps. Other advanced memory chips have seen a 6x increase in layers from 64 to 400 over the past few years. 3D stacking and building more layers necessitates more sophisticated metrology equipment to ensure critical measurements are met within each stack and layer.

Gate All Around Advanced Packaging

TSMC is phasing out FinFet after the 3nm node, and its upcoming 2nm node will be the first to use gate-all-around field-effect transistors (GAA FETs). GAA will increase chip density, and increase performance-per-watt to enable higher levels of output and efficiency.

These GAA nanosheet transistors have channels surrounded by gates on all sides to reduce leakage yet it will also uniquely widen the channels to provide a performance boost (with another option to narrow the channels to optimize power cost). The 2nm node will also allow chipmakers to customize the width and height of cells. Because of the increased complexity of GAA FETs compared to FinFETs, Nova estimates that GAA requires 20% to 30% more metrology steps than FinFET.

We’ve covered gate-all-around in a TSMC analysis that stated the following in terms of timing for N2 starting production in the last quarter of 2025 and reaching “meaningful revenue” by the Q1 to Q2 of 2026.

Accelerated Chip Development Timelines

Nvidia and AMD are both shifting to annual release cadences, aiming to bring next-generation GPUs to market once per year, compared to prior cadences of every two years. This is a major technological feat – as we had said previously for Nvidia, it’s a ‘move-fast-break-things’ problem, where Nvidia is pushing the boundaries of what had previously been seen as impossible in the chip industry.

By moving to these quicker release cycles, there’s a much greater emphasis on metrology and process control to ensure that the manufacturing process remains sharp, while also ensuring a faster ramp, and high yields to meet mass production thresholds and demand.

Fab Buildouts

Another driver for metrology equipment is chip fab buildouts, such as HBM fab construction and investments from SK Hynix, Micron and Samsung, as well as other expansion plans, such as TSMC’s planned expansion in Arizona.

Onto’s management put this in perspective from the HBM side, saying that “even if AI just stayed flat, which nobody is projecting, AI is supposed to go up, the HBM side is already doubling. … customers [are] talking about supply constraints and that they're looking at expanding factories into next year, early half of next year in order to alleviate these supply constraints.”

Currently, there are 19 global high-volume chip fabs under construction this year, with another 10 expected to be added next year to meet rising demand for semiconductors, driven by AI/HPC. Total equipment spending for these 29 fabs is expected to be approx. $140 billion. Through 2030, there are more than 100 fab projects planned, with more than $200 billion in global funding and incentives for these projects.

Building out new fabs from the ground-up will require metrology and process control equipment, providing a long-term growth runway for the industry.

Product Snapshot: Addressing Critical Challenges in Advanced Packaging, Memory, More

Onto and Nova’s management teams highlighted a few core products that are driving growth or seeing strong orders and adoption from major customers. These products address some of the more pressing challenges in advanced packaging, HBM and memory, and leading-edge nodes.

Israeli-based Nova has a diverse metrology equipment and software portfolio, from dimensional to materials to chemical metrology. One of Nova’s leading products is PRISM 2, the next-generation PRISM which offers improved sensitivity and accuracy in critical dimensions metrology.

Nova noted that its PRISM 2 product saw strong adoption last quarter, especially in GAA, with around half of its record bookings stemming from “advanced packaging processes, such as Through-Silicon Via, where PRISM has a unique advantage in filtering information from specific underlayers.”

PRISM 2 provides improved accuracy and reliability for critical component measurements in processes such as advanced logic chip fabrication and stacking nanosheets for GAA fabrication, as well as 3D NAND and DRAM.

Nova also added that “demand is also consistently high for our advanced integrated metrology and XPS material metrology platforms,” as it saw “record booking of VeraFlex XPS platforms this quarter, of which over 40% resulted from capacity growth” in FinFET advanced nodes.

Nova’s VeraFlex XPS (X-ray photoelectron spectroscopy/materials metrology) helps analyze surface structures, identifying bonding, contaminants or defects in advanced nodes for process control in high-volume manufacturing.

US-based Onto’s core products span defect inspection equipment, dimensional (optical CD) and material metrology, and lithography tools. Onto’s Dragonfly systems, which drove Q2’s performance, witnessed “better-than-expected demand” for advanced packaging for AI chips as it reached another revenue record.

Dragonfly G3 provides combo 2D and 3D inspection and process control for advanced packaging, HBM and chip-on-wafer GPUs. For example, Onto noted last August that it finalized more than $100 million in Dragonfly G3 orders for chip packages “that combine a graphics processor (GPU) and numerous high bandwidth memory (HBM) devices to create an AI GPU in a single package.”

Onto also said that Atlas and Iris demand for GAA investments were a driver of Q2’s performance, while new product launches in panel lithography, 3D bump metrology sensors for smaller interconnects in HBM, and subsurface defect inspection sensors are expected to see strong adoption through the end of the year.

By comparison, leader KLA offers a comprehensive portfolio of metrology equipment for measuring “pattern dimensions, film thicknesses, layer-to-layer alignment, pattern placement, surface topography and electro-optical properties” for chip and substrate manufacturing, from new process development, product verification, and high volume manufacturing. KLA provides advanced wafer packaging equipment, overlay metrology equipment, reticle metrology and inspection tools, OCD, shape and film metrology equipment, sheet resistance metrology equipment, and more.

Customer Concentration Extremely High

It’s common for small cap semiconductors to have high customer concentration, especially at the equipment level, as there are a limited number of fabs globally. Nova and Onto are no exception.

For Nova, its top five customers accounted for 52% of revenue in 2023, down from 57% of revenue in 2022 and 70% in 2021. Nova’s largest customer accounted for 19% of revenue, while its fifth largest customer contributed 5% of revenue. The customers are not named.

Revenue from Nova’s largest customer(s) declined from ~$130 million in 2021 and 2022 to $108 million in 2023.

Source: Nova

Onto Innovation has provided a breakout for its top three customers, along with another figure that highlight its customer concentration.

For Onto, its three largest customers accounted for 56% of revenue in the first half of fiscal 2024. TSMC accounted for 23% of revenue, up from 13% in the same period last year. Samsung accounted for 20% of revenue, down slightly from 21% last year.

SK Hynix has risen to be Onto’s third largest customer at 13% of revenue, up from <10% last year. Onto also has shared that TSMC accounted for 21% of its net accounts receivable in the first half of 2024, with Samsung accounting for 8%.

Source: Onto Innovation

Revenue Accelerating

Both Nova and Onto are reporting accelerating revenue growth in Q2 after returning to growth in Q1 this year after four consecutive quarter of declines. However, Onto guided for revenue growth to decelerate in Q3, while Nova forecast this acceleration to continue but be short-lived.

Onto reported 27.1% YoY growth to $242.3 million in revenue in Q2, accelerating from 14.9% growth in Q1. Management said that Q2’s growth and revenue above its guided $230-240 million range was “driven by additional pilot line expansions for high-performance computing using gate-all-around transistor architecture and high bandwidth memory (HBM) supporting AI market growth.” This was likely driven primarily by increased spend by TSMC and SK Hynix, seen within the customer concentration figures.

Advanced packaging and specialty device revenue reached a fourth-consecutive record at $164 million, up 49% YoY, while Onto added that it “closed over $300 million of volume purchase agreements issued by two customers for their AI advanced packaging and gate-all-around investments through 2025,” again likely to be coming from TSMC. Management expects that GAA and advanced packaging at multiple customers will be primary growth drivers through 2025.

Nova reported revenue growth of 27.8% YoY to $156.9 million in revenue in Q2, accelerating from 7.3% in Q1. Metrology product revenue was $124.6 million, accelerating to 30% YoY growth from just 6% YoY in Q1 and (16%) in Q4.

Management said that it saw “record revenue from chemical metrology, driven by demand in high-bandwidth-memory and front-end logic processes,” while demand for GAA and advanced packaging was increasing with “faster-than-expected adoption” for its PRISM 2 platform. Management added that they also saw record bookings for advanced packaging and materials metrology.

Q2 also marked both companies’ fastest YoY growth rates in two years — here’s a look at how Onto and Nova have recovered from 2023’s slowdown:

Looking ahead for Q3, Onto guided for revenue growth of 20.7% YoY to $250 million in revenue at midpoint, a more than 6 percentage point deceleration. Nova forecast a nearly 6 percentage point acceleration to 33.5% growth to $172 million in revenue at midpoint.

Through 2025, analysts expect Onto to maintain growth in the high-teens each quarter, while Nova’s higher growth rate of 30%+ is expected to decelerate into the low-teens and high-single digit range five quarters out.

Onto is expected to see quarterly revenues rise to above $300 million, while Nova is currently forecast to see revenues begin to flatline in the mid-$190 million range. Nova may be more exposed to cyclicality in the metrology industry with its focus on film metrology equipment, which may see faster growth in the ramp phase that stabilizes later on.

Onto provided a quick look into what’s driving 2025 growth: “We expect to benefit from continued investments in gate-all-around capacity, and the announced capacity expansions from several high-bandwidth memory and logic packaging manufacturers, specifically for HBM. We expect new capacity coming online in the first half of the year to support the increase of HBM content for NVIDIA's AI processors from 80 to 192 gigabytes and for AMD's AI processors from 192 gigabytes to 288 gigabytes.”

However, it’s not all AI revenue for the two. Onto clarified that AI packaging drove over half of specialty and advanced packaging revenue in Q2, implying AI packaging revenue of at least $82 million, or just over one-third of quarterly revenue. Nova is seeing demand stem from AI-driven end markets such as HBM, GAA and TSVs, though they have not broken out AI’s specific impact on revenue.

Long-Term Financial Targets:

Revenue

Both companies have set out long-term financial targets, including revenue, margins and EPS. The revenue targets will likely be achieved in 2027 to 2028 based on current expectations and historical trends. Both companies expect to have enough capacity by year end to reach these long-term revenue targets.

Nova set its long-term target of $1 billion, more than 50% higher than FY24’s projected $648 million. There are currently six analysts covering Nova with only two analysts providing estimates through Dec of 2027 with consensus at $995.7 million. 

Here's Nova’s target model:

Source: Nova

Onto has its long-term revenue target at $1.8 billion, nearly double FY24’s projected $980 million. There are eight analysts covering Onto with only three analysts covering the stock through Dec of 2027 with estimates at $1.56 billion. The Dec 2028 estimate is more in line with Onto’s projections with only one analyst providing an estimate of $1.84 billion.

Here’s Onto’s target model, which provides a more detailed view down the line as revenue increases incrementally by $200 million:

Source: Onto

Margins

Nova is currently ahead of its long-term margin targets, currently benefiting from higher ASPs due to product mix, while Onto is behind on gross margin targets for its $1 billion model.

Here’s how Nova’s margins look relative to its long-term target (referencing non-GAAP margins):

  • Nova reported GAAP gross margin of 59% in Q2, flat QoQ and up 200 bp YoY. Non-GAAP gross margin was 61% in Q2, again flat QoQ and up 200 bp YoY.
  • For Q3, Nova guided gross margins to dip 300 bp QoQ sequentially, to a GAAP gross margin of 56% and non-GAAP gross margin of 58%. Nova said that faster adoption of higher ASP/margin products has previously boosted margins, but they remain at the high-end of targets despite this fluctuation.
  • GAAP operating margin was 29% in Q2, expanding 300 bp QoQ and 600 bp YoY. Non-GAAP operating margin was 34%, up 200 bp QoQ and 600 bp YoY. Management pointed out that non-GAAP operating margin was ahead of targets of 27% to 31%, and while Q3’s non-GAAP margin was guided to decline sequentially to 32%, this is in line with previous quarters with Q2 being a strong outlier.

Onto sits behind gross margin targets, though it sees margins expanding in the coming quarter.

  • Onto reported GAAP and non-GAAP gross margin of 53% in Q2, up 100 bp QoQ and flat YoY. Though current FY revenue estimates are just 2% below the $1 billion target model, gross margins are shy of its range, with Onto targeting non-GAAP gross margins of 56% to 57%.
  • For Q3, management guided to gross margins between 53% to 55%, flat to up 200 bp QoQ, driven by growth in advanced nodes and optimization in manufacturing. Though it’s a step in the right direction, it’s still short of targets by ~200 bp at midpoint.
  • GAAP operating margin was 20% in Q2, up 100 bp QoQ and 700 bp YoY. Non-GAAP operating margin was 27%, up 200 bp QoQ and 600 by YoY.
  • Excluding amortization (to align with long-term targets), non-GAAP operating margin was 32%, in line with the high end of the $1 billion model despite the gross margin shortfall.
  • For Q3, management’s guidance implies non-GAAP gross margin of 28%, up 100 bp QoQ.

EPS

Both Nova and Onto are reporting strong EPS growth, with Nova reporting 48% YoY growth to $1.41 in GAAP EPS of $1.41 and Onto reporting 102% YoY growth to $1.07 in GAAP EPS. It’s rare to see such high EPS numbers from small cap stocks, and long-term targets are both strong and rather achievable.

For Nova, management is targeting non-GAAP EPS of at least $7 per share by the time it reaches $1 billion in revenue. For the first half of 2024, Nova’s non-GAAP EPS is currently $3 per share, suggesting that if margins can be maintained at the high end of its model, $7 in EPS is easily achievable. Analyst estimates currently project $9.58 in EPS on $996 million in revenue in 2027, foreseeing Nova surpassing the $7 EPS target next year on $757 million in revenue.

Onto is targeting non-GAAP EPS of $5.50 to $6.00 at $1 billion in revenue, and long-term up to $8.50 in EPS at $1.4 billion in revenue. Non-GAAP EPS for the first half of 2024 was $2.50, and with Q3 guided to $1.30 at midpoint, Onto would need to report $1.70 in non-GAAP EPS in Q4 to reach the low end of its target at $5.50.

Onto’s longer-term view emphasizes increased operating leverage to drive earnings growth as gross margin expansion is minimal. Management is eyeing just a 2% expansion in gross margin but a 4% expansion in operating margin. This is expected to drive EPS more than 40% higher, from $6 to $8.5 per share at the high end of the range. Analysts are much more optimistic on the degree of operating leverage that Onto can drive at $1.4 billion in revenue, with estimates calling for about $9.35 in EPS on $1.44 billion in revenue in 2026. Again, it’s rare to see EPS numbers this high with small cap semiconductor stocks.

Balance Sheets, Cash Flows Healthy

With operating and net margins in the high-20% to low-30% range, Onto and Nova enjoy healthy balance sheets and cash flows.

Onto has $786 million in cash, equivalents and marketable securities on hand, zero debt, and total liabilities of $174 million. Operating cash flow for 1H 2024 increased 50% YoY to $122 million, or 26% of revenue (versus 21% of revenue in same period last year). Free cash flow rose 49% YoY to $103 million, or 22% of revenue (versus 18% of revenue in the same period last year).

Nova has $759 million in cash, equivalents, and marketable securities, and $198 million in convertible senior notes, which, if converted, would equal about 3.4% dilution to existing shareholders. Nova has no outstanding debt outside of the convertible notes. Operating cash flow for 1H increased nearly 154% YoY to $120 million, or 40% of revenue (versus 19% of revenue in the year ago period). Free cash flow rose more than 178% YoY to $115 million, or 38% of revenue (versus 17% of revenue in the year ago period).

Inventories are where the two differ, with Nova increasing inventory as Onto is working to reduce inventory. Nova reported inventory of $156.6 million at the end of Q2, up more than 13% from the end of 2023. Onto is working on inventory management, reporting $319.7 million in inventory in Q2, down from $327.8 million at the end of 2023. Management expects to further reduce inventory by $10 million to $15 million in Q3 and exit 2024 with inventory below $300 million, or a $50 million YoY reduction.

Valuations Have Enjoyed a Premium

With clear exposure to AI and HPC trends in HBM and advanced packaging, as well as strong operating, net and cash flow margins — and triple digit cash flow growth for Nova – the two have recently enjoyed premium valuations to other chip equipment stocks and metrology leader KLA.

Onto currently trades at nearly 62x TTM earnings and Nova at almost 42x TTM earnings, compared to 38x TTM earnings for KLA and 22x TTM earnings for Applied Materials, which also has exposure to metrology and advanced packaging. Multiples have compressed slightly, with Onto trading above 80x earnings and Nova above 55x earnings earlier this year; however, it’s quite a large premium to their three-year average TTM earnings multiples of 37x for Onto and 31x for Nova.

On a forward basis, strong EPS growth of 40% YoY for Onto and 31% YoY for Nova has brought forward multiples lower, though still elevated relative to peers. Onto trades at nearly 40x forward earnings of $5.22 for FY24, while Nova trades at 31x forward earnings of $6.37 for FY24. KLA trades at 26x forward earnings with 25% YoY growth expected, and Applied Materials at 23x forward earnings with just 6% YoY growth expected.

On a top-line basis, Onto, Nova and KLAC are currently trading in the high 10x, to mid 11x PS range. Multiples for the trio have been in lockstep for more than a year, following each other quite closely since the 5-6x range around Nvidia’s blowout earnings report in May 2023, the first signs of the present AI and HBM boom.

Looking longer-term, it’s quite clear that these are some of the highest top-line multiples Onto and Nova have traded at in the past decade. Both companies’ 10-year average PS multiple hovers around 4.5x to 5x, so the two are currently trading at more than double their 10-year average. If earnings or revenue growth falters, multiple compression is a possibility given the premium valuations of the two.

Technical Analysis

By Knox Ridley

What caught our eye about the technical patterns in both NVMI and ONTO was how similar their long-term trends appear to be unfolding. Like many AI names that we track, they appear to be setting up for a correction within a larger uptrend that should go on for several years.

Most stocks and markets appear to have either topped in 2021 or are in their final swing higher in 2024/2025. So, while the broad market makes a series of lower highs into the coming years, these names would trend higher.

Onto Innovation (ONTO)

The below chart is a weekly chart. In other words, every candle is one week’s worth of price data. This better helps us decipher the long-term trend. Note the vertical move off the 2009 low, which was followed by a multi-year, messy and overlapping correction into 2015. This best resembles waves 1 and 2 in a very large 5 wave uptrend.

This is further supported by the price action into today’s developing top. The below Elliott Wave count has us in the final swing of wave 3 within the larger 3rd wave. We ccould see a final push into the $257 – $339 region. However, this is not a move that we would chase, considering the warning signs.

For one, we have a completed 5 wave pattern off thew COVID low. What follows 5 waves higher is usually a 3 wave retrace. Secondly, since mid-2023, we have seen price make higher highs while momentum is making lower highs. This is characteristic of 5 wave pattern, where peak momentum typically is seen in wave 3, as we wave 5 is met with less volume and less momentum.

Nova (NVMI)

NVMI has a similar long-term chart as ONTO. We can see a near vertical 5 wave move off of the 2009 low, followed by a pullback into 2011. This lines up well as waves 1 and 2 of a very large 5 wave pattern.

Like ONTO, Nova is in the final swing of wave 3 of the larger 3rd wave. If this count is accurate, the 4th wave pullback should take us back to around projected the halfway point of this larger 5 wave pattern, which is where we would look to potentially enter both stocks.

If we zoom in on the current correction, it appears that NOVA is setting up for the final drop in this correction. A move below $186 will confirm this as we target $152 – $138 for our first entry. If NOVA can instead hold $186 and then breakout above $230, we will be in the final swing of this 3rd wave. This stock has a lot to prove before we’d consider it a longer-term buy rather than a momentum stock.

Conclusion

Nvidia, TSMC, and Micron are bellwethers for AI accelerator, advanced packaging and HBM demand, with the three showing no signs of slowing any time soon. This will continue to create opportunities for growth in the metrology equipment market, as AI accelerator-related HBM and advanced packaging demand continues to grow and outstrip supply.

Metrology equipment is positioned well to capture the increase in complexity from advanced nodes, memory stacking and other chipmaking processes where demand is outstripping supply into the foreseeable future.

Onto and Nova share in these tailwinds, evidenced by accelerating quarterly revenue in Q2 and strong margins, cash flows and healthy balance sheets. We will be watching for further evidence that these companies are participating in the Ai boom in the upcoming quarters.

This analysis is a preview of what you can expect in our upcoming Discovery tier, which will provide additional analysis on new idea generation stocks that are not currently in the I/O Fund portfolio. We look forward to launching this tier November/December. There will be no changes to our current service tiers, rather I/O Fund Discovery is a service for those who want more new stock ideas beyond what our service currently provides. Stay tuned for more information!

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 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:

Nova and Onto Innovation: Growth in Metrology and Semiconductor Process Control

Posted on by io-fund

While Nvidia’s dominance in AI GPUs is more clearly visible downstream in the HBM market, it’s also driving growth in a lesser-known but equally as important market, in metrology/process control equipment.

Metrology, by definition, is the study of measurement. In the semiconductor industry, metrology plays a critical role, as the ability to measure circuit dimensions, microstructures inside chips, film thickness, accuracy of layers and overlays are all necessary to ensure chipmakers can consistently produce high-quality, flaw-free chips at high yields.

While KLA dominates the metrology market, reportedly holding a more than 50% market share for over 20 years, Onto Innovation and Nova are two smaller players in the market putting up accelerating revenue growth with strong margins, with exposure to AI megatrends and long-term growth drivers.

Metrology Has Multiple Drivers

The metrology market has multiple drivers of future growth, and while it’s not limited to the AI data center mega-trends spearheaded by Big Tech, there are many levers within this AI buildout to drive demand for Onto Innovation and Nova.

This is how Nova’s management team recently described the opportunity: “Artificial intelligence is reshaping our digital world and expediting the implementation of scaling agents in integrated circuits to boost performance and power efficiency which are at the core of enabling large language models through training and infra stages. Scaling is manifested in more complex designs through three avenues, dimensions, materials and advanced packaging.

It is also manifested in the size of manufactured dies and in the number of dies required in solutions such as high bandwidth memory. These developments profoundly impact our industry in the number of the wafers needed to supply demand and in the yield ratio, you must obtain to secure profitability with a lower number of dies per wafer. Couple that with the growing complexity of architecture and materials and you have the growth drivers of process control in the coming years.”

Here’s a brief look at how AI and the data center are generating demand for metrology equipment across these different drivers:

Shift to Advanced Nodes

Advanced nodes require more process steps as node sizes shrink, so for a chipmaker or foundry like TSMC or Intel to move from primarily producing on 5nm nodes to 2nm or below over the next couple of years, there will be a greater need for metrology equipment. Currently, the 5nm and 4nm nodes are primarily being used for AI chip production, such as that for Nvidia’s Hopper and Blackwell chips, while 3nm production is ramping at TSMC with volume production at the 2nm node expected in 2025, primarily for smartphone applications.

This is because the manufacturing tolerances shrinks as nodes shrink in size – the chipmaking process now becomes increasingly more sensitive to minute deviations in the process. Moving to more advanced nodes warrants much greater precision throughout the entire manufacturing process, as the smallest of deviations could greatly affect the process’ yield.

For example, Samsung has reportedly been struggling with severe yield issues for its most advanced nodes, with unstable yields reportedly around 50% or below for its gate all-around GAA 3nm second-gen process, and yields as low as 20% for its 2nm process.

TSMC sees higher yields for the 3nm process, reportedly between 60% and 70%, allowing it to command a majority of production on advanced nodes and thus a majority of market share.

Source: Nova

Rising Chip Complexity

Metrology needs also rise as chip designs get increasingly complex, from the entire GPU to memory cubes.

AI accelerators are getting larger and more complex with each generation packing more transistors than the last. Nvidia’s upcoming B200 GPU packs 208 billion transistors and 192GB of HBM3e memory, versus 80 billion transistors and 80GB HBM3 memory for the H100. Though Blackwell is in a dual-die configuration, it’s still a more complicated chip to produce using TSMC’s customized N4P process (the same as for the H100).

Having a more complex chip design increases the chances of errors, as there are more steps in the manufacturing process to produce said chips. This also goes for DRAM and NAND cubes – as HBM cubes stack higher, from 8-high cubes to 12-high cubes for the current HBM3e generations from SK Hynix, Samsung and Micron, it also increases process steps. Other advanced memory chips have seen a 6x increase in layers from 64 to 400 over the past few years. 3D stacking and building more layers necessitates more sophisticated metrology equipment to ensure critical measurements are met within each stack and layer.

Gate All Around Advanced Packaging

TSMC is phasing out FinFet after the 3nm node, and its upcoming 2nm node will be the first to use gate-all-around field-effect transistors (GAA FETs). GAA will increase chip density, and increase performance-per-watt to enable higher levels of output and efficiency.

These GAA nanosheet transistors have channels surrounded by gates on all sides to reduce leakage yet it will also uniquely widen the channels to provide a performance boost (with another option to narrow the channels to optimize power cost). The 2nm node will also allow chipmakers to customize the width and height of cells. Because of the increased complexity of GAA FETs compared to FinFETs, Nova estimates that GAA requires 20% to 30% more metrology steps than FinFET.

We’ve covered gate-all-around in a TSMC analysis that stated the following in terms of timing for N2 starting production in the last quarter of 2025 and reaching “meaningful revenue” by the Q1 to Q2 of 2026.

Accelerated Chip Development Timelines

Nvidia and AMD are both shifting to annual release cadences, aiming to bring next-generation GPUs to market once per year, compared to prior cadences of every two years. This is a major technological feat – as we had said previously for Nvidia, it’s a ‘move-fast-break-things’ problem, where Nvidia is pushing the boundaries of what had previously been seen as impossible in the chip industry.

By moving to these quicker release cycles, there’s a much greater emphasis on metrology and process control to ensure that the manufacturing process remains sharp, while also ensuring a faster ramp, and high yields to meet mass production thresholds and demand.

Fab Buildouts

Another driver for metrology equipment is chip fab buildouts, such as HBM fab construction and investments from SK Hynix, Micron and Samsung, as well as other expansion plans, such as TSMC’s planned expansion in Arizona.

Onto’s management put this in perspective from the HBM side, saying that “even if AI just stayed flat, which nobody is projecting, AI is supposed to go up, the HBM side is already doubling. … customers [are] talking about supply constraints and that they're looking at expanding factories into next year, early half of next year in order to alleviate these supply constraints.”

Currently, there are 19 global high-volume chip fabs under construction this year, with another 10 expected to be added next year to meet rising demand for semiconductors, driven by AI/HPC. Total equipment spending for these 29 fabs is expected to be approx. $140 billion. Through 2030, there are more than 100 fab projects planned, with more than $200 billion in global funding and incentives for these projects.

Building out new fabs from the ground-up will require metrology and process control equipment, providing a long-term growth runway for the industry.

Product Snapshot: Addressing Critical Challenges in Advanced Packaging, Memory, More

Onto and Nova’s management teams highlighted a few core products that are driving growth or seeing strong orders and adoption from major customers. These products address some of the more pressing challenges in advanced packaging, HBM and memory, and leading-edge nodes.

Israeli-based Nova has a diverse metrology equipment and software portfolio, from dimensional to materials to chemical metrology. One of Nova’s leading products is PRISM 2, the next-generation PRISM which offers improved sensitivity and accuracy in critical dimensions metrology.

Nova noted that its PRISM 2 product saw strong adoption last quarter, especially in GAA, with around half of its record bookings stemming from “advanced packaging processes, such as Through-Silicon Via, where PRISM has a unique advantage in filtering information from specific underlayers.”

PRISM 2 provides improved accuracy and reliability for critical component measurements in processes such as advanced logic chip fabrication and stacking nanosheets for GAA fabrication, as well as 3D NAND and DRAM.

Nova also added that “demand is also consistently high for our advanced integrated metrology and XPS material metrology platforms,” as it saw “record booking of VeraFlex XPS platforms this quarter, of which over 40% resulted from capacity growth” in FinFET advanced nodes.

Nova’s VeraFlex XPS (X-ray photoelectron spectroscopy/materials metrology) helps analyze surface structures, identifying bonding, contaminants or defects in advanced nodes for process control in high-volume manufacturing.

US-based Onto’s core products span defect inspection equipment, dimensional (optical CD) and material metrology, and lithography tools. Onto’s Dragonfly systems, which drove Q2’s performance, witnessed “better-than-expected demand” for advanced packaging for AI chips as it reached another revenue record.

Dragonfly G3 provides combo 2D and 3D inspection and process control for advanced packaging, HBM and chip-on-wafer GPUs. For example, Onto noted last August that it finalized more than $100 million in Dragonfly G3 orders for chip packages “that combine a graphics processor (GPU) and numerous high bandwidth memory (HBM) devices to create an AI GPU in a single package.”

Onto also said that Atlas and Iris demand for GAA investments were a driver of Q2’s performance, while new product launches in panel lithography, 3D bump metrology sensors for smaller interconnects in HBM, and subsurface defect inspection sensors are expected to see strong adoption through the end of the year.

By comparison, leader KLA offers a comprehensive portfolio of metrology equipment for measuring “pattern dimensions, film thicknesses, layer-to-layer alignment, pattern placement, surface topography and electro-optical properties” for chip and substrate manufacturing, from new process development, product verification, and high volume manufacturing. KLA provides advanced wafer packaging equipment, overlay metrology equipment, reticle metrology and inspection tools, OCD, shape and film metrology equipment, sheet resistance metrology equipment, and more.

Customer Concentration Extremely High

It’s common for small cap semiconductors to have high customer concentration, especially at the equipment level, as there are a limited number of fabs globally. Nova and Onto are no exception.

For Nova, its top five customers accounted for 52% of revenue in 2023, down from 57% of revenue in 2022 and 70% in 2021. Nova’s largest customer accounted for 19% of revenue, while its fifth largest customer contributed 5% of revenue. The customers are not named.

Revenue from Nova’s largest customer(s) declined from ~$130 million in 2021 and 2022 to $108 million in 2023.

Source: Nova

Onto Innovation has provided a breakout for its top three customers, along with another figure that highlight its customer concentration.

For Onto, its three largest customers accounted for 56% of revenue in the first half of fiscal 2024. TSMC accounted for 23% of revenue, up from 13% in the same period last year. Samsung accounted for 20% of revenue, down slightly from 21% last year.

SK Hynix has risen to be Onto’s third largest customer at 13% of revenue, up from <10% last year. Onto also has shared that TSMC accounted for 21% of its net accounts receivable in the first half of 2024, with Samsung accounting for 8%.

Source: Onto Innovation

Revenue Accelerating

Both Nova and Onto are reporting accelerating revenue growth in Q2 after returning to growth in Q1 this year after four consecutive quarter of declines. However, Onto guided for revenue growth to decelerate in Q3, while Nova forecast this acceleration to continue but be short-lived.

Onto reported 27.1% YoY growth to $242.3 million in revenue in Q2, accelerating from 14.9% growth in Q1. Management said that Q2’s growth and revenue above its guided $230-240 million range was “driven by additional pilot line expansions for high-performance computing using gate-all-around transistor architecture and high bandwidth memory (HBM) supporting AI market growth.” This was likely driven primarily by increased spend by TSMC and SK Hynix, seen within the customer concentration figures.

Advanced packaging and specialty device revenue reached a fourth-consecutive record at $164 million, up 49% YoY, while Onto added that it “closed over $300 million of volume purchase agreements issued by two customers for their AI advanced packaging and gate-all-around investments through 2025,” again likely to be coming from TSMC. Management expects that GAA and advanced packaging at multiple customers will be primary growth drivers through 2025.

Nova reported revenue growth of 27.8% YoY to $156.9 million in revenue in Q2, accelerating from 7.3% in Q1. Metrology product revenue was $124.6 million, accelerating to 30% YoY growth from just 6% YoY in Q1 and (16%) in Q4.

Management said that it saw “record revenue from chemical metrology, driven by demand in high-bandwidth-memory and front-end logic processes,” while demand for GAA and advanced packaging was increasing with “faster-than-expected adoption” for its PRISM 2 platform. Management added that they also saw record bookings for advanced packaging and materials metrology.

Q2 also marked both companies’ fastest YoY growth rates in two years — here’s a look at how Onto and Nova have recovered from 2023’s slowdown:

Looking ahead for Q3, Onto guided for revenue growth of 20.7% YoY to $250 million in revenue at midpoint, a more than 6 percentage point deceleration. Nova forecast a nearly 6 percentage point acceleration to 33.5% growth to $172 million in revenue at midpoint.

Through 2025, analysts expect Onto to maintain growth in the high-teens each quarter, while Nova’s higher growth rate of 30%+ is expected to decelerate into the low-teens and high-single digit range five quarters out.

Onto is expected to see quarterly revenues rise to above $300 million, while Nova is currently forecast to see revenues begin to flatline in the mid-$190 million range. Nova may be more exposed to cyclicality in the metrology industry with its focus on film metrology equipment, which may see faster growth in the ramp phase that stabilizes later on.

Onto provided a quick look into what’s driving 2025 growth: “We expect to benefit from continued investments in gate-all-around capacity, and the announced capacity expansions from several high-bandwidth memory and logic packaging manufacturers, specifically for HBM. We expect new capacity coming online in the first half of the year to support the increase of HBM content for NVIDIA's AI processors from 80 to 192 gigabytes and for AMD's AI processors from 192 gigabytes to 288 gigabytes.”

However, it’s not all AI revenue for the two. Onto clarified that AI packaging drove over half of specialty and advanced packaging revenue in Q2, implying AI packaging revenue of at least $82 million, or just over one-third of quarterly revenue. Nova is seeing demand stem from AI-driven end markets such as HBM, GAA and TSVs, though they have not broken out AI’s specific impact on revenue.

Long-Term Financial Targets:

Revenue

Both companies have set out long-term financial targets, including revenue, margins and EPS. The revenue targets will likely be achieved in 2027 to 2028 based on current expectations and historical trends. Both companies expect to have enough capacity by year end to reach these long-term revenue targets.

Nova set its long-term target of $1 billion, more than 50% higher than FY24’s projected $648 million. There are currently six analysts covering Nova with only two analysts providing estimates through Dec of 2027 with consensus at $995.7 million. 

Here's Nova’s target model:

Source: Nova

Onto has its long-term revenue target at $1.8 billion, nearly double FY24’s projected $980 million. There are eight analysts covering Onto with only three analysts covering the stock through Dec of 2027 with estimates at $1.56 billion. The Dec 2028 estimate is more in line with Onto’s projections with only one analyst providing an estimate of $1.84 billion.

Here’s Onto’s target model, which provides a more detailed view down the line as revenue increases incrementally by $200 million:

Source: Onto

Margins

Nova is currently ahead of its long-term margin targets, currently benefiting from higher ASPs due to product mix, while Onto is behind on gross margin targets for its $1 billion model.

Here’s how Nova’s margins look relative to its long-term target (referencing non-GAAP margins):

  • Nova reported GAAP gross margin of 59% in Q2, flat QoQ and up 200 bp YoY. Non-GAAP gross margin was 61% in Q2, again flat QoQ and up 200 bp YoY.
  • For Q3, Nova guided gross margins to dip 300 bp QoQ sequentially, to a GAAP gross margin of 56% and non-GAAP gross margin of 58%. Nova said that faster adoption of higher ASP/margin products has previously boosted margins, but they remain at the high-end of targets despite this fluctuation.
  • GAAP operating margin was 29% in Q2, expanding 300 bp QoQ and 600 bp YoY. Non-GAAP operating margin was 34%, up 200 bp QoQ and 600 bp YoY. Management pointed out that non-GAAP operating margin was ahead of targets of 27% to 31%, and while Q3’s non-GAAP margin was guided to decline sequentially to 32%, this is in line with previous quarters with Q2 being a strong outlier.

Onto sits behind gross margin targets, though it sees margins expanding in the coming quarter.

  • Onto reported GAAP and non-GAAP gross margin of 53% in Q2, up 100 bp QoQ and flat YoY. Though current FY revenue estimates are just 2% below the $1 billion target model, gross margins are shy of its range, with Onto targeting non-GAAP gross margins of 56% to 57%.
  • For Q3, management guided to gross margins between 53% to 55%, flat to up 200 bp QoQ, driven by growth in advanced nodes and optimization in manufacturing. Though it’s a step in the right direction, it’s still short of targets by ~200 bp at midpoint.
  • GAAP operating margin was 20% in Q2, up 100 bp QoQ and 700 bp YoY. Non-GAAP operating margin was 27%, up 200 bp QoQ and 600 by YoY.
  • Excluding amortization (to align with long-term targets), non-GAAP operating margin was 32%, in line with the high end of the $1 billion model despite the gross margin shortfall.
  • For Q3, management’s guidance implies non-GAAP gross margin of 28%, up 100 bp QoQ.

EPS

Both Nova and Onto are reporting strong EPS growth, with Nova reporting 48% YoY growth to $1.41 in GAAP EPS of $1.41 and Onto reporting 102% YoY growth to $1.07 in GAAP EPS. It’s rare to see such high EPS numbers from small cap stocks, and long-term targets are both strong and rather achievable.

For Nova, management is targeting non-GAAP EPS of at least $7 per share by the time it reaches $1 billion in revenue. For the first half of 2024, Nova’s non-GAAP EPS is currently $3 per share, suggesting that if margins can be maintained at the high end of its model, $7 in EPS is easily achievable. Analyst estimates currently project $9.58 in EPS on $996 million in revenue in 2027, foreseeing Nova surpassing the $7 EPS target next year on $757 million in revenue.

Onto is targeting non-GAAP EPS of $5.50 to $6.00 at $1 billion in revenue, and long-term up to $8.50 in EPS at $1.4 billion in revenue. Non-GAAP EPS for the first half of 2024 was $2.50, and with Q3 guided to $1.30 at midpoint, Onto would need to report $1.70 in non-GAAP EPS in Q4 to reach the low end of its target at $5.50.

Onto’s longer-term view emphasizes increased operating leverage to drive earnings growth as gross margin expansion is minimal. Management is eyeing just a 2% expansion in gross margin but a 4% expansion in operating margin. This is expected to drive EPS more than 40% higher, from $6 to $8.5 per share at the high end of the range. Analysts are much more optimistic on the degree of operating leverage that Onto can drive at $1.4 billion in revenue, with estimates calling for about $9.35 in EPS on $1.44 billion in revenue in 2026. Again, it’s rare to see EPS numbers this high with small cap semiconductor stocks.

Balance Sheets, Cash Flows Healthy

With operating and net margins in the high-20% to low-30% range, Onto and Nova enjoy healthy balance sheets and cash flows.

Onto has $786 million in cash, equivalents and marketable securities on hand, zero debt, and total liabilities of $174 million. Operating cash flow for 1H 2024 increased 50% YoY to $122 million, or 26% of revenue (versus 21% of revenue in same period last year). Free cash flow rose 49% YoY to $103 million, or 22% of revenue (versus 18% of revenue in the same period last year).

Nova has $759 million in cash, equivalents, and marketable securities, and $198 million in convertible senior notes, which, if converted, would equal about 3.4% dilution to existing shareholders. Nova has no outstanding debt outside of the convertible notes. Operating cash flow for 1H increased nearly 154% YoY to $120 million, or 40% of revenue (versus 19% of revenue in the year ago period). Free cash flow rose more than 178% YoY to $115 million, or 38% of revenue (versus 17% of revenue in the year ago period).

Inventories are where the two differ, with Nova increasing inventory as Onto is working to reduce inventory. Nova reported inventory of $156.6 million at the end of Q2, up more than 13% from the end of 2023. Onto is working on inventory management, reporting $319.7 million in inventory in Q2, down from $327.8 million at the end of 2023. Management expects to further reduce inventory by $10 million to $15 million in Q3 and exit 2024 with inventory below $300 million, or a $50 million YoY reduction.

Valuations Have Enjoyed a Premium

With clear exposure to AI and HPC trends in HBM and advanced packaging, as well as strong operating, net and cash flow margins — and triple digit cash flow growth for Nova – the two have recently enjoyed premium valuations to other chip equipment stocks and metrology leader KLA.

Onto currently trades at nearly 62x TTM earnings and Nova at almost 42x TTM earnings, compared to 38x TTM earnings for KLA and 22x TTM earnings for Applied Materials, which also has exposure to metrology and advanced packaging. Multiples have compressed slightly, with Onto trading above 80x earnings and Nova above 55x earnings earlier this year; however, it’s quite a large premium to their three-year average TTM earnings multiples of 37x for Onto and 31x for Nova.

On a forward basis, strong EPS growth of 40% YoY for Onto and 31% YoY for Nova has brought forward multiples lower, though still elevated relative to peers. Onto trades at nearly 40x forward earnings of $5.22 for FY24, while Nova trades at 31x forward earnings of $6.37 for FY24. KLA trades at 26x forward earnings with 25% YoY growth expected, and Applied Materials at 23x forward earnings with just 6% YoY growth expected.

On a top-line basis, Onto, Nova and KLAC are currently trading in the high 10x, to mid 11x PS range. Multiples for the trio have been in lockstep for more than a year, following each other quite closely since the 5-6x range around Nvidia’s blowout earnings report in May 2023, the first signs of the present AI and HBM boom.

Looking longer-term, it’s quite clear that these are some of the highest top-line multiples Onto and Nova have traded at in the past decade. Both companies’ 10-year average PS multiple hovers around 4.5x to 5x, so the two are currently trading at more than double their 10-year average. If earnings or revenue growth falters, multiple compression is a possibility given the premium valuations of the two.

Technical Analysis

By Knox Ridley

What caught our eye about the technical patterns in both NVMI and ONTO was how similar their long-term trends appear to be unfolding. Like many AI names that we track, they appear to be setting up for a correction within a larger uptrend that should go on for several years.

Most stocks and markets appear to have either topped in 2021 or are in their final swing higher in 2024/2025. So, while the broad market makes a series of lower highs into the coming years, these names would trend higher.

Onto Innovation (ONTO)

The below chart is a weekly chart. In other words, every candle is one week’s worth of price data. This better helps us decipher the long-term trend. Note the vertical move off the 2009 low, which was followed by a multi-year, messy and overlapping correction into 2015. This best resembles waves 1 and 2 in a very large 5 wave uptrend.

This is further supported by the price action into today’s developing top. The below Elliott Wave count has us in the final swing of wave 3 within the larger 3rd wave. We ccould see a final push into the $257 – $339 region. However, this is not a move that we would chase, considering the warning signs.

For one, we have a completed 5 wave pattern off thew COVID low. What follows 5 waves higher is usually a 3 wave retrace. Secondly, since mid-2023, we have seen price make higher highs while momentum is making lower highs. This is characteristic of 5 wave pattern, where peak momentum typically is seen in wave 3, as we wave 5 is met with less volume and less momentum.

Nova (NVMI)

NVMI has a similar long-term chart as ONTO. We can see a near vertical 5 wave move off of the 2009 low, followed by a pullback into 2011. This lines up well as waves 1 and 2 of a very large 5 wave pattern.

Like ONTO, Nova is in the final swing of wave 3 of the larger 3rd wave. If this count is accurate, the 4th wave pullback should take us back to around projected the halfway point of this larger 5 wave pattern, which is where we would look to potentially enter both stocks.

If we zoom in on the current correction, it appears that NOVA is setting up for the final drop in this correction. A move below $186 will confirm this as we target $152 – $138 for our first entry. If NOVA can instead hold $186 and then breakout above $230, we will be in the final swing of this 3rd wave. This stock has a lot to prove before we’d consider it a longer-term buy rather than a momentum stock.

Conclusion

Nvidia, TSMC, and Micron are bellwethers for AI accelerator, advanced packaging and HBM demand, with the three showing no signs of slowing any time soon. This will continue to create opportunities for growth in the metrology equipment market, as AI accelerator-related HBM and advanced packaging demand continues to grow and outstrip supply.

Metrology equipment is positioned well to capture the increase in complexity from advanced nodes, memory stacking and other chipmaking processes where demand is outstripping supply into the foreseeable future.

Onto and Nova share in these tailwinds, evidenced by accelerating quarterly revenue in Q2 and strong margins, cash flows and healthy balance sheets. We will be watching for further evidence that these companies are participating in the Ai boom in the upcoming quarters.

This analysis is a preview of what you can expect in our upcoming Discovery tier, which will provide additional analysis on new idea generation stocks that are not currently in the I/O Fund portfolio. We look forward to launching this tier November/December. There will be no changes to our current service tiers, rather I/O Fund Discovery is a service for those who want more new stock ideas beyond what our service currently provides. Stay tuned for more information!

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 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: