Applied Digital easily beat estimates in Q1 with revenue up 69% QoQ with fit-out revenue contributing significantly in the quarter as the company prepares to roll out its first phase for CoreWeave. Barely two weeks after the earnings report, Applied signed a $5 billion, 15-year lease at Polaris Forge 2 with an unnamed, investment-grade hyperscaler, which helps de-risk its story from being tied solely to CoreWeave. Combined with the neocloud’s lease at PF1, the two deals for will generate $1.07 billion in average annual revenue over the lifetime of the contracts, or ~5X fiscal 2025 revenue, highlighting the attractiveness of pivoting to AI data centers.
More importantly, Applied also quietly disclosed that they have 4GW in the pipeline with additional capacity under review, doubling from our prior update Applied Digital: Bitcoin Miner Hinting at Rare, Hyperscaler Deal, and nearly 3X more than the 1.4GW disclosed two quarters ago. With only 600MW currently contracted at PF1 and PF2, this hints at substantial upside to contracted revenue and net operating income at full scale.
Applied Extends Active Pipeline to 4.3GW, the Highest Among Miners
Perhaps the most important update coming from Q1’s earnings call was that Applied has quickly and quietly expanded its active development pipeline – just two quarters ago, the company disclosed 1.4GW in the pipeline, yet now it has tripled this to 4.3GW this quarter across nine sites. This would rank Applied atop the leading miners, outpacing Galaxy’s 3.5GW and IREN’s 2.9 GW of grid-connected power.
For the 4.3 GW, CEO Wesley Cummins explained that these are projects “we feel could move into that construction box in the next 6 to 12 months, and some of those could be even sooner. So those are things we're actively working on with permitting, with power, with all of those pieces that we think in the next 6 to 12 months can move into the construction pipeline.” He added that there is demand for sites ranging from hundreds of MW to a multi-GW scale, with emphasis on sites “built in a single location so that you get the cost advantages of building a scale in a single location,” which Applied’s pipeline covers with sizes from 250MW to 1GW+.
This is especially important as Applied continues to reiterate its ability to shorten its construction timelines, from 24 months down to 12 to 14 months. Management is working to match the pace of building with power delivery, starting construction early to ensure buildings are prepped and ready once power is available. Essentially, Applied is hinting that with limited holdups from permitting, with smooth power delivery and necessary financing, it could bring its pipeline to power in as quickly as two and a half years.
This could make the company increasingly more attractive from hyperscalers as other miners are not targeting having even 1GW online by the end of 2027 – management also disclosed that they have “entered negotiations with 2 additional hyperscalers for 2 new locations,” with 100MW under negotiation.
Should this pipeline materialize to operational capacity, Applied’s revenue and NOI opportunities could be 6X its current contracted capacity of 600MW. Assuming deal terms similar to PF1 and PF2, the remaining 3.7 GW pipeline could be worth $6.1 billion to $6.7 billion in average annual revenue, compared to the $1.07 billion in average annual revenue it has currently contracted out.
$5 Billion Lease Secured at Polaris Forge 2
Applied broke ground on Polaris Forge 2 in September, with the facility having an initial 300MW capacity. Applied said in Q1’s call that it has secured financing for the project via Macquarie with an expected cost of $3 billion, or $10 million per MW.
On October 22, Applied announced that it had signed a $5 billion, 15-year deal with an unnamed hyperscaler for 200MW capacity at PF2. On the headline, this is a slight discount to CoreWeave’s lease at $1.67 million per MW per year on average versus $1.83 million per MW per year, with a slightly lower NOI margin of ~86% +/- 3% versus 88% for CoreWeave’s deal. Management explained that having the hyperscaler provides a lower cost of capital, thus the spread between capital cost and revenue is approximately equal.
Securing this second deal with a major hyperscaler is important as it helps de-risk the story from being linked to CoreWeave, whose financials are upside down and require creative ways to raise cash to finance lofty growth ambitions.
“Firmly” On Track to Reach $1B NOI Target in 5 Years
While the hyperscaler engagement at PF2 is certainly good news to hear, management provided a snapshot into long-term net operating income (NOI) targets, providing a clearer view of how the deals will translate into earnings.
Management stated that they believe they can reach an “annualized NOI run rate of approximately $500 million once Polaris Forge 1 is fully operational,” while the “tenant signing at our second campus should put us firmly on the path toward our $1 billion NOI target within the next five years.”
At full scale, the 600MW of contracted capacity would translate into approximately $932 million in NOI based on expected margins of 88% and 86% across its two deals. Looking further out to Applied’s current active pipeline, the remaining 3.7GW could generate around $5.5 billion in annual NOI on average at full scale at similar margins.
While not a true comparison to NOI, analysts currently project Applied’s EBITDA to rise more than 10X by 2028, from $60.7 million expected this fiscal year to $640.2 million as these two deals begin to ramp towards full capacity. This would represent an expansion of EBITDA margin from 20.4% to 66%, still below targeted NOI margins. Additionally, there is the potential for EBITDA to rise by another factor of 7-8X in the long run if Applied can successfully commercialize its entire active development pipeline.
Project Financing Deal with Macquarie Unlocks 5X More Capital
As we discussed in our prior analysis, financing partnerships and capital raises are central to funding Applied and its HPC buildout. In Q1, the company drew $112.5 million from its $5 billion preferred equity financing with Macquarie, which management says helped fund the completion of PF1.
Applied also secured $50 million from Macquarie Equipment Capital, to help fund the groundbreaking for PF2, while also adding that it intends to tap the $5 billion vehicle to help fund the subsequent buildout. Additionally, Applied noted that subsequent to the quarter, it raised an $200 million from an expanded offering of its Series G Preferred Stock, providing more capital to fund these buildouts.
In the earnings calls, management noted that they may have the ability to finance both PF1 and PF2 by themselves, but they would prefer to tap the project financing from Macquarie as it lets them unlock significant capacity growth:
“When you look from a capital perspective, what we're seeking to do there is we could finance the Ellendale campus Polaris Forge 1 by ourselves. We probably even finance Polaris Forge 2 by ourselves.
But what we're trying to put in place and what we have put in place now is the ability for us to scale much larger. We're looking more into the future and putting a mechanism in place that eliminates or minimizes the dilution at the public company for a set amount at the subsidiary for Macquarie. And this allows us to go forward. The Macquarie Capital, $5 billion of capital really unlocks $20 billion to $25 billion of total capital for us when you include project finance and that allows us to build a significant amount of capacity.”
Instead of being capital constrained with two builds, Applied believes the $5 billion line from Macquarie could allow them to build >2GW with the amount of capital it can unlock.
Brief Update on Polaris Forge 1, South Dakota Development
Applied Digital this week announced that the first 50MW phase for CoreWeave is now ready for service, with the remaining 350MW to be rolled out in phases through 2027.
The fit-out of PF1 contributed $26.3 million in revenue in the quarter, with this expected to ramp significantly in the first part of fiscal Q2 leading up to the start of service in late October. Now that the first 50MW phase is online, lease revenues will begin ramping in the latter half of fiscal Q2 ending November and ramp further in Q3 as the next 50MW comes online by year-end.
Applied also provided a brief update on progress in South Dakota, where it was reported back in May 2025 that the company was planning to construct a $16 billion, 430 MW data center. Management said that power would be available in South Dakota in 2026, though the one piece they say is the gating factor for development is a sales tax exemption for IT data center equipment.
Financials
Revenue Surges 69% QoQ, Driven by CoreWeave Fit-out
Applied’s revenue rose 69% QoQ and 84% YoY to $64.2 million, driven primarily by the fit-out of Polaris Forge 1, which contributed $26.3 million in tenant-fit out revenue. This was more than 41% ahead of estimates for $45.5 million in revenue.
For fiscal Q2, revenue is expected to be $82.2 million, up 28.7% YoY and 28% QoQ. Fiscal Q3 (ending Feb 2026) is currently projected to see $71.4 million in revenue, up 35% YoY but down (13.1%) QoQ as fit-out revenue shifts to lease revenue.
Fiscal 2026 revenue is expected to be $297.3 million for YoY growth of 106.2%, with fiscal 2027 (ending May 2027) currently projected at $553.0 million for 86% YoY growth.
Gross margins felt a pinch in Q1 due to the ramp in fit-out activity, though operating margins improved from Q4 yet remain a decent distance from GAAP profitability.
GAAP gross margin was 13.4% in Q1, down from 20.5% in Q4 and 27.6% a year ago due to increase in low margin fit-out revenue. Applied said the $26.3 million in fit-out revenue carried a cost of $25 million, implying barely a 5% gross margin. The ramp of fit-out in Q2 may further pressure gross margin though this should ease by Q3 as lease revenue arises.
GAAP operating margin was (34.7%) in Q1, up from (54.5%) in Q4; the 72.6% year-ago comp is not necessarily comparable due to a $24.8M gain on assets held for sale. Adjusted operating margin was (5.6%), improving from (8.1%) in Q4 but down from 6.2% in the year ago quarter.
GAAP net margin was (28.8%) in Q1, improving from (70%) in Q4 and not comparable to the 45.5% from the year ago quarter. Adjusted net margin was (11.8%), improving from (19.9%) in Q4 but down from (2.3%) a year ago.
EPS Beats, but Not Yet Profitable
Applied beat on EPS in the quarter, with adjusted EPS of ($0.03) coming in well ahead of the ($0.16) estimate. GAAP EPS also beat at ($0.07) versus the ($0.13) estimate.
Looking ahead to Q2, GAAP EPS is expected to dip slightly to ($0.11), likely driven by margin pressure related to the ramp in fit-out revenue, before rebounding slightly to ($0.09) in Q3. For fiscal 2026, GAAP EPS is projected at ($0.45), before improving to ($0.27) in fiscal 2027 and shifting to a profit of $0.86 in 2028.
Cash Flows Heavily Negative on High Capex
Cash flows were heavily negative, with FCF margin widening to (516%) in Q1 driven by a sharp increase in capex.
Operating cash flow was ($82.0 million) for (127.7%) margin, down from 18.0% in Q4 but improving from (217.8%) in the year ago quarter.
Free cash flow was ($331.4 million) for a (516.1%) margin, driven by $249 million PP&E purchases. This compared to a (503.5%) margin in Q4 and a (375%) margin in the year ago quarter.
Cash and equivalents totaled $114.1 million, not including Applied’s $362.5 million raise subsequent to quarter-end. Debt totaled $687.3 million.
Conclusion
Applied’s second deal with a hyperscaler customer at PF2 boosts confidence in its AI data center hosting story and de-risks it from CoreWeave, putting it firmly on track to reach its $1 billion net operating income target by 2030, up from $60.7 million expected this fiscal year. Additionally, Applied disclosed that they have an active pipeline of 4.3 GW but with only 700 MW of capacity under construction, highlighting that revenue and NOI opportunities at full scale could be up to 6X larger at similar terms.
Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.
Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in APLD at the time of writing and may own stocks pictured in the charts.
The energy crisis brought about by AI is a strong thematic hat will surface many opportunities for stock investors, yet it’s those companies that supply power quickly that we are building our portfolio with.
As a Bitcoin miner, Applied Digital has repositioned itself as an AI data center energy stock, with campuses primarily in the Dakotas. With long-term lease agreements — including multi-hundred-megawatt deals with AI cloud providers like CoreWeave — Applied Digital is a front runner that offers lower power usage effectiveness (PUE), an important metric that measures how much electricity is consumed by a data center. This PUE is achieved through free cooling as the Dakota region is colder than other regions and has surplus power the State exports. The company also hinted they are in advanced, direct negotiations with a hyperscaler. If this is confirmed, it will be the first among the miners to sign a direct deal.
With that said, many Bitcoin miners offer financials that are deep in the red yet are expected to sharply rebound from the high-margin revenue that AI data center deals will provide. Layer-in that Bitcoin miners have a highly volatile trading history, and what you get is a high risk/high reward approach to participating in stocks that promises to solve AI’s largest bottleneck – which is that AI's power consumption is outpacing what the grid can supply.
For investors, the challenge in participating in this trend is two-fold – how to identify opportunities in the vast energy sector and how to manage volatility given the margins and cash in this sector is lumpy (to put it kindly). We recently covered Bitcoin miners in our analysis “Bitcoin Miners Addressing AI’s Near-term Time to Power Bottleneck with up to $50 Billion in Commitments"Bitcoin Miners Addressing AI’s Near-term Time to Power Bottleneck with up to $50 Billion in Commitments"
As stated in the analysis: “Instead of having to worry about the prolonged process of site selection, permitting, planning, and more before final power delivery, these neoclouds instead have near immediate access to the powered shell. While retrofitting for liquid cooling, networking and connectivity may still be necessary and pose some challenges, [Bitcoin] miners offer a fast time to operation and relatively cheaper capex costs for a hyperscale-size data center outfitted with tens to hundreds of thousands of GPUs.
Bitcoin miners can only meet a fraction of this growth, likely around several gigawatts in total. Yet their innate ability to deliver this power over the next 12 to 24 months, supporting up to hundreds of thousands of high-end GPUs in larger-scale facilities, is why miners are prime targets to meet hyperscalers and neoclouds’ immediate power needs.”
Below we look at Applied Digital in more detail, the first in a Series of Bitcoin Miners that we have planned for our premium members over the coming weeks.
Note, this is a momentum stock and we plan to adhere strictly to risk management.
Applied Digital Located in the Dakotas with PUE of 1.18
Applied Digital made a strong argument in a recent white paper that offering a more strategic location in the Dakotas can result in savings of up to $60 to $90 million per year by offering free cooling.
We’ve covered in the past the importance of liquid cooling, as AI/ML require massive amounts of data processing, and as future generations of CPUs and GPUs are released, these systems will exceed air cooling capacity. Liquid cooling solves throttling, which occurs when CPUs and GPUs overheat and are throttled back to avoid damage to the chip. In the case of high-performance computing, liquid cooling reduces total cost of ownership as air cooling requires air conditioning and server fans to run constantly.
Cooling data center servers is responsible for 40% of the data center energy consumption. According to Dell, enclosed DLC solutions can save up to 23% of energy compared to traditional air-cooled racks. McKinsey places this number at 27% savings when there is 75% liquid cooled and 25% air cooled servers.
What Applied Digital is proposing is that mechanical liquid cooling is not needed as much in cooler regions such as the Dakotas, stating their mining operation “uses ambient air to cool liquid directly – no mechanical refrigeration needed for significant portions of the year.”
This represents a PUE improvement from the industry average of 1.58 to Applied Digital’s stated 1.18, translating into estimated annual savings of $60–90 million. In the example below, the calculation highlights $84 million in savings, driven by a significantly lower cost per kilowatt-hour.
Source: Applied Digital white paper
When comparing the Dakotas to other regions, such as North Virginia and Texas, the Dakotas have surplus energy with 33% exported compared to these other power-constrained regions. The region also offers 220 days of free cooling, which drives the power savings described above.
Management discussed these benefits in the recent earnings call, stating: “This design seeks to achieve a projected PUE of 1.18 and near 0 water consumption intended to ensure exceptional efficiency and sustainability. We like this location for its abundant low-cost synergy, some of which is generated from stranded power with over 200 days of free natural cooling. We have calculated that 100-megawatt data center customer could save up to $2.7 billion over a 30-year period as compared to the current industry data centers in other regions.”
CoreWeave is Primary Customer; Potential Incoming Hyperscaler(s)
As it stands, CoreWeave is the primary customer for Applied Digital’s capacity. Last quarter, CRWV signed a 15-year lease agreement for 250 MW and an extension for 150 MW for the Polaris Forge 1 data center site located in Ellendale, North Dakota.
The roll-out is expected to look like this:
First 100-megawatt facility scheduled to be operational in Q4 of 2025
Second 150-megawatt facility coming online in mid-2026
Third 150-megawatt facility planned for 2027
There are two paths to revenue for this deal. The first is “fit-out” revenue where Applied Digital is paid to prepare the site. This revenue is expected to kick-in for the current quarter and next quarter, leading to a “significant sequential” increase in revenue. From there, the lease deal is worth $11 billion over a 15-year time frame, or about $733M per year averaged out.
Here is what was stated: “We expect revenue to increase significantly sequentially, beginning in the quarter ending for August 2025 due to the technical fit out of our first Polaris Forge 1 building. Note, our customer pays the cost of this fit-out with a small margin to the company. This fit-out revenue will largely be recognized in both the current fiscal quarter and as well as the quarter ending November 30, 2025. Now this is before the actual lease revenue for the facility begins to be recognized.”
CoreWeave has a few strategic advantages over hyperscalers when it comes to pursuing energy deals with hyperscalers. The first is the speed in which a smaller company like CoreWeave can negotiate, the second is the company’s willingness to acquire power that is located away from major metro hubs, and third, CoreWeave’s core business is supply fast and AI-optimized access to GPUs, whereas Big Tech has many priorities across their core segments.
However, with that said, the market will favor any miner that can sign a hyperscaler given CoreWeave’s financials are upside down and require creative ways to raise cash to finance their ambitions. When it comes to a steady, stable flow of cash for a Bitcoin miner (who they, themselves have high debt leverage), there is no better customer than a hyperscaler. You can think of CoreWeave signing lease deals with Bitcoin miners as somewhat of a house of cards – it works well when stocks are up, it does not work well if we go through a longer period of AI softness.
This is where Applied Digital stands out – the company has hinted they are in negotiations with two hyperscalers with one of the hyperscalers being in “advanced negotiations.” Management made sure to point out how rare it is for a Bitcoin miner to have a hyperscaler as a potential customer.
The crux of the issue for Bitcoin miners is the quality and longevity of these stocks as they will have to diversify beyond cash-strapped neoclouds to raise their credit ratings and offer investors visibility in terms of funding the construction of these massive multi-billion dollar projects. Therefore, I’m quoting in full the discussion on the high likelihood that Applied Digital not only secures a hyperscaler deal – but do so soon:
“Besides CoreWeave, we have completed the diligence and onboarding process with 2 other investment- grade North American hyperscalers […] We also expect to benefit from this competitive advantage as new entrants to the market confront time, money and effort it takes to overcome these industry syncretic barriers to entry for other players. We also, for us, we feel we are now in a position to do business with these companies in the future with a much shorter negotiating and contracting completion process. In fact, we are currently in various stages of negotiation with several investment-grade hyperscalers for large capacity campuses other than our Polaris Forge 1 campus with 1 of those negotiations being in an advanced stage.”
How Quickly can Applied Digital Add Capacity (and How Much):
Applied Digital has a few paths for announcing more deals. The first is that Polaris 1 offers 1GW of capacity and yet CoreWeave has contracted 400 MW of the available 1 GW.
Secondly, Applied Digital broke ground on a new $3 billion campus this month called Polaris 2 in Harwood, North Dakota. This new campus is expected to “scale beyond scheduled operations in 2026 and full capacity in early 2027.”
In terms of size for Polaris 2 and stated delivery for full capacity by 2027, the following was shared in the call indicating it could be at least 1 GW and will be completed in 12-14 months time: “Building on the momentum from these leases and the surging demand for AI infrastructure, we're actively marketing our multi- gigawatt pipeline to a diverse group of customers. […] As a result, we believe we've reduced our projected build times from 24 months to 12 to 14 months, allowing us to deliver on large-scale commitments faster and more efficiently than before.”
Note, the exact size and timeline was not officially provided yet the comment above helps to frame initial expectations.
The Value of 2 GW
If we assume Applied Digital will lease 2G by early 2027, and if we figure that each 400 MW is worth $73 million based on the CoreWeave deal, then a rough estimate of what Applied Digital's AI factories will be worth is $18.25M per 100MW. This means the total 2GW is roughly worth $365 million. This does not take into account the “fit-out” revenue, which we will get a better idea of in the upcoming quarter.
Given the 2027 estimates are for $501 million in revenue, taken at face value, one could argue Polaris 1 and Polaris 2 are priced in. This does not take into account additional sites, and if a hyperscaler pays more than what CoreWeave is paying. According to commentary on the call, APLD may be breaking ground on a third site: “We do expect to break ground and work has already started for that on 1 additional campus and potentially 2 before the end of this year.”
Project Costs and Financing
When we look at capex costs of $11-$13 million per 100 MW, the cash flow for Applied Digital should pay for the project costs. Financing partnerships and capital raises are central to funding APLD and its HPC buildout. For example, Macquarie has committed up to $5 billion, with $900 million already allocated to Ellendale and $4.1 billion available for future development. In addition, Applied raised roughly ~$875 million in FY25 though equity and debt financing, with another $268.9 million post-quarter.
Cash / Debt Position: At the end of Q4 FY25, cash stood at $120.9 million and debt at $688.2 million, yielding a Cash / Debt ratio of .18x. Including the post quarter raise, the ratio improves to 0.57x, providing significantly more liquidity runway. Still, this lags peers such as IREN (0.59x) and RIOT (0.38x), while being broadly in line with WULF (0.18x).
As you can see from the tables below, Applied’s Cash / Debt ratio has been extremely volatile. FY24 cash levels hovered around $30-40 million with minimal debt, spiked to $314 million in Q2 FY25 due to financing inflows, and declined again as capex surged in Q3 and Q4. This underscores the company’s reliance on external financing, followed by rapid deployment into HPC infrastructure.
Looking deeper into the Company’s investor base, another strong signal comes from Nvidia’s involvement. In September 2024, NVDA participated in APLD’s $160 million private placement financing, landing NVDA a ~3% stake in the Company valued around $63.7 million at the time. Importantly, Applied was already recognized as a Preferred NVIDIA Cloud Partner prior to this deal, meaning the investment represents more than just capital, it reinforces strategic alignment. For Applied, having Nvidia on the cap table provides credibility, potential preferential access to GPU’s, and validates the Company’s pivot into high-performance compute hosting. Relative to peers, few digital infrastructure firms can point to this level of backing, which helps APLD as a trusted partner for hyperscalers and AI-native cloud firms.
While historical financials illustrate the growing pains of a transition year, investors should focus on the Company’s financing position and go-forward revenue estimates, which accelerate meaningfully over the next four quarters. The $7 billion contracted HPC backlog implies a structural step-up in revenue visibility, while the post-quarter raise improves the cash/debt ratio from .18x to .57x, extending liquidity runway. These forward-looking metrics are more indicative of Applied’s trajectory than backward-looking margins which are distorted by non-cash charges and build-out spend. Ultimately, Applied’s ability to balance its liquidity runway with execution on HPC energization will determine whether the $7 billion in contracted backlog translates into the sharp revenue acceleration that analysts expect.
Revenue
Revenue optics are noisy due to Cloud Services fluctuations, but the underlying Hosting growth is much healthier than the FY headline suggests. The Q4 exit run-rate and $7 billion plus in HPC lease backlog indicate a more favorable forward profile.
Q4 Revenue came in at $38.0 million, representing 41% growth YoY and 4% growth QoQ. YoY growth reflects Ellendale operating at full capacity (vs. partial outages in Q4’24) and the transition away from related-party contracts, leaving a more third-party-driven base. QoQ growth of $1.5 million shows the strength of Hosting revenues amid Cloud Services decline, which was reclassified as “held for sale”.
FY2025 Revenue of $144.2 million represents 6% YoY growth or $7.6 million compared to $136.6 million in FY24. The growth this year looks weak compared to FY24, mainly due to the termination of legacy contracts in Q1‘25 and inconsistent Cloud Services revenue across the fiscal year.
While FY25 revenue printed slightly below consensus ($144.2 million vs. $148.0 million), the shortfall was isolated to Cloud Services, which has since been discontinued. Hosting was broadly in line, underscoring the durability of the core business. As you can see in the chart below, analysts expect top-line growth to accelerate sharply moving forward, with estimates calling for $250 million in FY26 and $502 million in FY2027. These strong forward estimates help explain why shares have traded less on historical misses and more on execution / timing of HPC energization.
Shifting our focus toward the segment breakdown, Hosting is the durable core business with stable recurring revenues between $35 – $38 million per quarter. As mentioned above, Q4 marked the first time both Jamestown (106 MW) and Ellendale (180 MW) facilities ran at stable full capacity.
Cloud Services drove growth early in FY25 during Q1 and Q2, shrank in Q3, and has now been re-classified as held for sale. These fluctuations are largely driven by GPU contract structure changes and management’s decision to wind down the unit.
HPC / AI Leases are not currently contributing to FY25 revenue but these are critical for the Company’s trajectory and should be viewed as the growth driver. Management highlighted that signed leases with Coreweave represent $7 billion in contracted revenue to be recognized over 15 years. For added context, this contract alone is roughly 49x FY25 revenue of $144.2 million. Suffice to say, the Company is undergoing transition among its key segments but the strong backlog implies a change in trajectory.
Margins: Underlying gross margin strength offset by OpEx, SG&A caution signal
Gross margin trend is positive but material swings in Opex and financing costs mask any improvement. Q4 spike in SG&A is a red flag on cost control that should be monitored going forward. The key takeaway here is that unit economics are improving at the gross level but cost discipline and financing overhands remain key risks.
Q4 gross profit of $7.8 million, down $7.6 million or 49% compared to Q3, but up $3.7 million or 90% compared to Q4’24. Gross margins of 20.5% vs. 15.2% in Q4’24 reflect 530 basis points of improvement, driven largely by the increase in revenue, along with better facility uptime and efficiency, plus mix shift away from lower-margin Cloud Services.
Q4 operating income of ($20.7) million, $1.7 million lower than Q3’25 and $18.0 million lower than Q4’24. Operating margin of (54.5%) is down substantially from (9.9%) in Q4’24 and heavily pressured by a 115% increase YoY in SG&A spend as the company expanded headcount, increased stock comp, and ramped corporate overhead associated with the HPC buildout.
Q4 Net income of ($26.6) million reflects sequential improvement of $30.1 million against Q3’25 and annual improvement of $8.7 million compared to Q4’24. Net Income Margin of (70.0%) is up from (131.3%) reported in Q4’24, driven by higher revenue and better gross margins, slightly offset by increase in opex mentioned above.
FY25 gross profit $42.7 million is up $12.8 million compared to FY24. This represents a gross margin of 29.6%, 770 bps of incremental progress compared to 21.9% seen in FY24. The full year improvement is largely driven by increased scale and less power-related disruptions.
FY25 operating loss of ($16.8) million reflects improvement of $16.1 million compared to ($32.9) million operating loss in FY24. This represents an Operating margin of (11.7%) compared to (24.0%) in FY24. Despite the heavy expenditures noted in Q4, margins improved on a full-year basis, indicating operating leverage.
FY25 Net Loss of ($161.0) million continues to expand versus ($74.0) million loss reported in FY24. This represents Net Margins of (112%), down compared to (54.2%) in FY24. Despite the top line growth and improving unit economics, net margins for the full year worsened due to $119 million in non-cash losses tied to debt conversions, warrants, and derivative liabilities.
EPS
Adjusted figures provide a more accurate view of operating performance, showing narrowing core losses, while GAAP EPS was heavily distorted by financing and derivative accounting. Investors should focus on Adjusted EPS, which shows sequential and YoY improvement, while GAAP EPS remains noisy until the financing structure stabilizes.
Q4 GAAP EPS of ($0.12) improved from ($0.28) in Q4’24, reflecting stronger revenue and gross profit at both Jamestown and Ellendale as uptime normalized. On an adjusted basis, EPS narrowed to ($.03), close to breakeven, as higher gross margin partially offset heavier SG&A tied to HPC build-out.
FY25 GAAP EPS of ($0.80) worsened from ($0.65) in FY24 due to $119 million in non-cash charges (debt conversions, warrant issuances, and derivative liabilities). Adjusted EP of ($.06) versus ($0.11) in FY24 highlights meaningful progress in narrowing core losses despite heavy investment and Cloud services volatility.
Cash Flow and Balance Sheet
FY25 was defined by record capital deployment into HPC infrastructure. While Q4 cash burn moderated sequentially, full year OCF and FCF deterioration underscore the scale of investment and reliance on external financing. Liquidity was extended through large capital raises but leverage ballooned. Sustained improvement in OCF and eventual FCF stabilization will hinge on timely energization and monetization of HPC leases.
Q4 Operating Cash Flow of ($15.7) million, up from ($37.2) million in Q3’24, as working capital normalized. Operating Cash Flow margin improved to (44.3%) versus (101.9%) in Q3.
Q4 Free Cash Flow of ($79.2) million, improved from ($261.5) million reported in Q3, with FCF margin narrowing to (208.4%) from (716.4%). Sequentially, lower capex spend drove the improvement.
FY25 Operating Cash Flow of ($115.4) million down significantly from $13.8 million in FY24, with OCF margin falling to (80.0%), down from 10.1% reported in FY24.
FY25 Free Cash Flow of ($797.0) million, down significantly from ($128.0) million in FY24. FCF Margin of (553.0%), down from (93.7%) in FY24, reflecting the heavy investment cycle for Polaris Forge and HPC facilities.
Cash of $120.9 million, up from $31.7 million in Q4’24 and up $46.7 million sequentially, was boosted by roughly $875 million in equity and debt financing. As noted by Management, this “does not include the additional $268.9 million in proceeds from our ATM and Series G preferred stock offering that occurred post quarter.”
Debt of Q4 was $688.2 million, up from $653.3 million in Q3’25 and a significant step up compared to $79.5 million as of Q4’24 (nearly 9x YoY increase).
Conclusion:
Applied Digital asserts they are unique due to being located in the Dakotas where free cooling is available for more than half the year. The company also points toward this region’s surplus of power as a strategic advantage compared to power constrained regions like Texas and North Virginia.
Most importantly, should Applied Digital announce a deal with a hyperscaler, then it will be the first among the miners. Should the deal materialize, it would indicate their operations are attractive on a competitive basis given hyperscalers have a slow, thorough process for site review.
We hold a small allocation in Applied Digital and plan to actively risk manage the position.
Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in APLD at the time of writing and may own stocks pictured in the charts.
GPU sales are surging at the moment, primarily from Big Tech’s $200 billion in capex for AI infrastructure services. Critical data center components, including networking, are required for GPU systems. There is a well-quoted discussion from Dell executives earlier this year, that by the end of the systems lifecycle, $2 to $3 will be spent on networking and storage for every $1 spent on GPUs. Granted, the effects of this 2-X market demand will be spread across many more players compared to GPUs. Yet, there is ample evidence that networking is sparking a remarkable growth trajectory of its own. For example, Nvidia’s InfiniBand has seen triple digit growth, Cisco has provided strong AI commentary, and Arista Networks’ view is that networking is mission-critical to improve GPU utilization.
Arista Networking is positioning itself as a pure-play in AI-driven networking, and management sees tailwinds to growth not only via Ethernet establishing itself as the go-to choice in networking, especially for AI training, but also stemming from the broader market opportunity arising from the massive shipment volumes of Nvidia’s GPUs.
Why Data Center Spend Is Accelerating
The current AI landscape is spearheaded by Big Tech. Microsoft and Amazon are touting multi-billion-dollar cloud revenue run rates from AI, Google sees a clear path to monetizing AI features, and Meta is aggressively investing in AI with some initial evidence it’s boosting average revenue per user (ARPU). What’s unfolding is an AI ‘arms race’, in that Big Tech, and a handful of startups including OpenAI, Mistral, Anthropic, and others, are competing to develop, deploy and commercialize the next cutting-edge AI model. This race also spreads over to who can develop the best AI assistants/Copilots, increase adoption of GenAI tools, and accelerate revenue growth in the cloud.
Nvidia CEO Jensen Huang explained exactly why this race is rapidly unfolding, and why Big Tech’s AI expenditures are increasing not only this year, but likely for the next few years: “time is really, really valuable to them. Let me give you an example of time being really valuable, why this idea of standing up a data center instantaneously is so valuable and getting this thing called time to train is so valuable. The reason for that is because the next company who reaches the next major plateau gets to announce a groundbreaking AI. And the second one after that gets to announce something that's 0.3% better. And so the question is, do you want to be repeatedly the company delivering groundbreaking AI or the company delivering 0.3% better? And that's the reason why this race, as in all technology races, the race is so important. And you're seeing this race across multiple companies because this is so vital to have technology leadership, for companies to trust the leadership and want to build on your platform and know that the platform that they're building on is going to get better and better.”
The conclusion is that Big Tech firms are snapping up Nvidia’s GPUs as fast as they reach the market, and this demand spills over into AI servers and networking components, as both are crucial for AI systems.
Networking Becoming Indispensable
Big Tech is deploying thousands to (soon) millions of GPUs and in-house AI accelerators, and networking is a mission-critical piece. Switches are crucial for communication inside the GPU clusters, allowing quick, efficient communication and transfer of data between each node, which is essential for parallel processing, and thus overall job completion time when it comes to training large-scale AI models and completing larger workloads.
A cohesive networking layer with high-quality switching technology can lower power consumption needs significantly and improve performance job completion times. As a result, the industry is going all-in on Ethernet, Nvidia included, in part due to its compatibility, cost and performance advantages, and security.
According to Arista, Ethernet has advantages over Nvidia’s InfiniBand: “AI workloads are placing greater demand on Ethernet, as they are both data and compute-intensive across thousands of processes today. Basically, AI at scale needs Ethernet at scale. AI workloads cannot tolerate the delays in the network, because the job can only be completed after all flows are successfully delivered to the GPU clusters. All it takes is one culprit of worst-case link to throttle an entire AI workload.”
Broadcom’s management seconded this, explaining that as companies scale GPU clusters, they are “going to have to use the best networking technology. And we believe that the best networking technology is Ethernet.” Broadcom’s Ram Velaga added that whether GPUs are “connected inside a data center or across data centers, you cannot get around the fact that you have to connect multiple GPUs. Once you accept the fact that it is a distributed computing problem and you need a network, then I would make a very strong case for you that the best network in the world, over multiple generations, over and again, has been Ethernet.”
Velaga used Meta as an example as to why networking (and Ethernet) is so important: when Meta is running “large workloads, anywhere between 18% to 57% of the time, the traffic is just sitting in the network. That means during this period of time, the GPUs are actually sitting idle. Now think about it. If on an average somebody is charging somebody between $20,000 to $30,000 per GPU and you've got 100,000 GPUs, you're talking about a $2 billion to $3 billion infrastructure. And if $2 billion to $3 billion infrastructure is sitting idle for 18% to 57% of the time, that's a lot of money, right?”
By creating more efficient lines of communication between GPUs, Ethernet can help accelerate job completion times, and in turn, allow more jobs to be completed on the clusters. Velaga touched upon the performance advantages of Ethernet versus Nvidia’s InfiniBand, noting that Meta tested both on a 24,000 GPU cluster and found that Ethernet provides up to 10% better performance at half the cost, which, when translated over to overall infrastructure costs, could equal hundreds of millions to billions saved.
Nvidia is also prioritizing Ethernet with its new Spectrum-X solution, despite seeing strong triple-digit networking revenue growth (accelerating from 94% YoY to 242% YoY in 4 quarters to over $3 billion in quarterly revenue) driven by InfiniBand. CEO Jensen Huang said Nvidia is “all-in on Ethernet” with an “exciting road map coming.” He added that “Spectrum-X is ramping in volume with multiple customers, including a massive 100,000 GPU cluster. Spectrum-X opens a brand-new market to NVIDIA networking and enables Ethernet only data centers to accommodate large-scale AI. We expect Spectrum-X to jump to a multibillion-dollar product line within a year.”
For more information on how Ethernet compares to InfiniBand, reference our analysis: “Broadcom: Networking/ASICs Giant and The Second Largest by AI Revenue” where we go through a side-by-side comparison including why Big Tech is pushing for Ethernet over Nvidia’s in-house InfiniBand.Broadcom: Networking/ASICs Giant and The Second Largest by AI Revenue” where we go through a side-by-side comparison including why Big Tech is pushing for Ethernet over Nvidia’s in-house InfiniBand.
Arista Confident in Ethernet Opportunity
Arista echoed much of Broadcom’s comments on Ethernet’s performance advantages versus InfiniBand, reiterating that Ethernet is “proving to offer at least 10% improvement of job completion performance across all packet sizes versus InfiniBand.”
Arista added that they are “witnessing an inflection of AI networking and expect this to continue throughout the year and decade” as Ethernet emerges as “a critical infrastructure across both front-end and back-end AI data centers.”
Per Arista’s Q1 call, we are “progressing well in four major AI Ethernet clusters that we won versus InfiniBand recently,” and in the four clusters, they are “migrating from trials to pilots, connecting thousands of GPUs this year, and we expect production in the range of 10K to 100K GPUs in 2025.”
Arista’s management reiterated the company will reach an AI target of $750 million in 2025. Barclays believes Arista “can top its guidance of $750M in AI-related back-end revenue for 2025,” commanding 18% market share in data center switching (versus 27% share for Nvidia and 22% for Cisco).
Arista also has a few core risks, particularly in that revenue is heavily concentrated in two major customers, Microsoft and Meta, accounting for 38% of company-wide revenue ($2.2 billion of Arista’s $5.8 billion in revenue in 2023). While we are seeing both companies spending quite aggressively in AI this year and next, revisions to capex plans intra-year (much like how we saw Meta increase its full year capex guide last quarter) can move Arista’s stock price.
Microsoft and Meta accounted for 39% of Arista’s revenue in 2023, down slightly from 42% in 2022 but up significantly from less than 25% in 2021. In dollar terms, the two are both billion-dollar customers, with revenue from Microsoft increasing more than 50% in 2023 and nearly 59% in 2022.
Both Microsoft and Meta increased 2024’s capex plans, with Microsoft’s Q1 capex rising 80% YoY to $14 billion, and full fiscal year capex up 50% YoY to $50 billion. Microsoft is reportedly seeking to triple its GPU supply this year to 1.8 million GPUs to support AI demand on Azure, and demand for networking components should rise hand in hand.
Meta boosted its full year capex range to $35-40 billion, pointing to 33% YoY growth and $4 billion more than previously anticipated, to build out AI infrastructure and support its internal AI roadmap. Meta’s Q1 capex was only $6.7 billion, implying that the bulk of this spend will hit in the second half of the year, possibly accelerating at a ~20% QoQ rate and exiting 2024 above the $11 billion range – hinting that Meta’s contributions to Arista’s revenue growth may not be felt in full force until the back half of 2024.
While the capex growth is positive, competition in the networking space is high, from Broadcom to Nvidia to Cisco to others. Cisco noted that it has been seeing strong momentum in Ethernet AI fabric deployment at three of the top four hyperscalers, possibly alongside Arista’s solutions, while Nvidia has recently released its Spectrum-X Ethernet solution which it expects to become a multibillion-dollar product line with a year. According to Nvidia, Spectrum-X delivers 1.6X better networking performance than traditional Ethernet.
Our previous Broadcom analysis points toward the Ethernet networking giant being second in AI revenue, primarily from AI networking revenue. Forward-looking, Broadcom is expected to end the year with $2.75 billion per quarter in AI revenue for $11B per year. Compare this to Nvidia’s networking revenue at $3.2 billion.
InfiniBand increases dependency on Nvidia, requires a new networking stack, and lags Ethernet on raw bandwidth. Improvements in Ethernet systems are expected to offer better load balancing and congestion control to help close the gap with InfiniBand’s low latency. Broadcom’s Jericho3-AI switch platform is the company’s AI fabric that competes with InfiniBand on AI training completion times, and it allows for more than 32,000 GPUS to be linked for a massive AI training system.
Regarding Arista, the company has partnered to offer a holistic solution, where a remote Arista-based AI agent will help customers optimize and manage their AI clusters with a single control point; however, investors should expect competition in the market to remain fierce as hyperscalers continue to build out data center infrastructure.
Turning to fiscal Q1’s earnings — Arista delivered a solid report, with revenue ahead of expectations as margins remained strong. However, headline revenue growth has decelerated rather quickly as Arista faced difficult comps in Q1. Despite the deceleration, the bottom line remains strong and in fact is strengthening.
In terms of AI revenue, management did not provide a figure for 2024, but its $750 million target for 2025 would represent close to 10% of total revenue, with consensus for FY2025 at $7.82 billion.
Revenue and EPS:
Arista reported revenue of $1.57 billion in Q1, representing YoY growth of 16.3% and beating expectations by nearly $24 million. This is down from 54% growth in the year-ago quarter. Growth is expected to decelerate more than 4 percentage points on a sequential basis in Q2.
GAAP EPS of $1.99 represented YoY growth of 44.2%, beating estimates by $0.40.
For Q2, Arista guided revenue between $1.62 billion and $1.65 billion, representing YoY growth of 12.1% at midpoint, a fifth consecutive quarter of decelerating revenue growth. However, Q2 is expected to mark the bottom, with analysts expecting growth to reaccelerate to the 14%+ range by Q4.
Margins:
Gross margin was 63.7% in Q1, down 120 bp QoQ from a six-year high in Q4 at 64.9%. Gross margin has been relatively stable in the 63% to 64% range aside from a dip to the 60% range in the first half of 2023.
Operating margin reached a record high at 42.0% in Q1, and represented a 50 bp QoQ and 610 bp YoY expansion. Operating margin has expanded steadily since 2021, increasing nearly 10 percentage points from the low 30% level.
Net margin was 40.6%, up 80 bp QoQ and 830 bp YoY. Arista’s bottom line strength should not be overlooked, especially as leverage improves down the line even with decelerating revenue growth. In the market’s top AI stocks at the moment, Arista has one of the strongest bottom lines outside of Nvidia.
Cash and Debt:
Arista reported cash and equivalents of $5.45 billion, and has zero debt.
Cash flow margins are strong — operating cash flow increased over 37% YoY to $514 million, for a 32.7% margin. Free cash flow also increased 37% YoY to $504 million, for a 32% margin.
How Arista Networks Ranks on AI Revenue:
Here’s a quick glance on the rankings for AI networking revenue, with Arista near the bottom of the list. Nvidia and Broadcom lead the sector, with Nvidia recently surpassing a $13 billion annual run rate in networking.
Nvidia is in first with $3.2 billion in networking revenue in fiscal Q1, after recently surpassing a $13 billion annual run rate in Q4. Nvidia is expecting its new Spectrum-X product to reach a multi-billion dollar run rate “within a year.”
Broadcom is in second place with AI revenue of $3.1 billion in fiscal Q2, projecting an annual run rate of more than $11 billion, or more than $2.75 billion per quarter. Based on management’s commentary, networking likely contributed upwards of $1 billion in the quarter, and could exit the year in the mid-$4 billion range.
Marvell reported approximately $500 million in AI revenue in the most recent quarter, with management eyeing a “a floor of $1.5 billion for AI revenue” for this fiscal year, with two-thirds coming from electro-optics and one-third from ASICs.
Juniper Networks reported $321.2 million in the AI enterprise segment, or 23.5% of revenue. Recently, it was announced that Juniper is being acquired by HPE.
Arista has not broken out AI revenue on a quarterly basis yet, but is targeting $750 million in AI revenue in 2025, which is equivalent to ~10% of revenue for that year.
Valuation:
Arista’s top line valuation has surpassed historical peaks. Bottom line strength and improved operating leverage are driving increased earnings power exiting 2024 with a bottom line valuation in line historically.
Arista currently trades at 19.4x sales and 17.3x forward sales, both above historical highs – in late 2021, Arista peaked at just under 17x sales, the same level where it pulled back from in early May following its post-earnings rally. Buying in the 8x sales range offers a higher probability for upside, although notably, AI stocks have not offered low entries over the past year.
On the bottom line, Arista trades slightly below 52x earnings and nearly 47x forward earnings, which on the surface is expensive, and above its 5-year average of 34x. Arista peaked at 60x earnings at its 17x sales peak multiple in late 2021, providing a tiny bit of breathing room for the bottom-line valuation on improved earnings power later this year and through next; however, downside risk is more prevalent as both the top and bottom-line valuations become stretched.
There are currently two counts that I see playing out in this final push higher. Both counts have us in the final swing higher of the larger 3rd wave, they only differ on how high this swing can go before we see a larger pullback:
Technical Analysis
By Knox Ridley
There are currently two counts that I see playing out in this final push higher. Both counts have us in the final swing higher of the larger 3rd wave, they only differ on how high this swing can go before we see a larger pullback:
Green – If we can breakout over $390 and hold this level, then the odds favor this path higher. This would target the $440 – $480 region directly. The final target should be between $490 – $520 before we see the larger 4th wave pullback.
Red – If we fail to breakout over $390, and instead see a breakdown below $335. This would signal that we are in the 4th wave drop, which I generally have targets between $260 – $200. If this plays out, as long as we hold $195, this would be a decent buying opportunity for the final 5th wave swing higher.
Conclusion
InfiniBand has been reporting up to 500% growth and more recently 300% growth, yet for the first time since the AI surge, Nvidia reported that networking declined sequentially this past quarter.
Despite Nvidia’s GPU moat being fully intact, its networking lead is in question. Gartner recently reported that by 2028, 80% of hyperscalers will “opportunistically” prefer Ethernet over proprietary technologies. According to Broadcom, this shift is happening quickly with management predicting that as soon as next year “all mega-scale GPU deployments will be on Ethernet.”
Arista Networks is a stock to watch in this space, primarily for its defensibility on the bottom line. The company sees $750 million in back-end AI revenue in 2025 stemming from growth among hyperscalers. Heavy revenue concentration in Meta and Microsoft is a core risk to watch, yet capex spend is accelerating for the time being, providing growth tailwinds for the networking industry as a whole. Next week, we will revisit another Ethernet networking play with more revenue and a stronger growth story for 2025, despite being weaker on the bottom line.
Damien Robbins, Equity Analyst at the I/O Fund, contributed to this article.
This quarter, Super Micro reported revenue of $3.85 billion, reflecting a staggering growth rate of 200.7% YoY. This technically missed estimates by 1.3%. Management increased guidance to between $5.1 billion to $5.5 billion, up from $4.9 billion, indicating year-over-year growth of 142.6% at the midpoint.
The company's GAAP EPS of $6.56, surpassed analyst expectations of $5.16. Next quarter, the expected GAAP EPS ranges from $7.20 to $8.05 compared to analyst expectations of $6.87 EPS.
What lies beneath this phenomenal growth rate is the need to raise cash to fund operations, which for Super Micro means buying excess inventory to prepare for future growth, especially as it relates to liquid cooling. Inventory of $4.1 billion amounts to 85% of revenue, whereas SMCI held closer to 70% of revenue in the past.
This quarter, the operating cash flow margin was (-39.5%) and the free cash flow margin was (-42%). This is a material change to the story given the current macro environment, and management did not indicate it will get better in the near-term as it’s related to components for liquid cooling systems. Per management: “We continue to face some supply chain challenges due to newer products that require new key components, especially, specifically, DLC related components, and believe this situation will gradually improve in the coming quarters.”
Revenue and EPS
For the third quarter of fiscal year 2024, Super Micro reported revenue of $3.85 billion, up $2.57 billion, or 200.7%, compared to $1.28 billion for the same quarter in the previous year. This growth was slightly under the projected midpoint based on management’s previous guidance of revenue in the range of $3.7 billion to $4.1 billion.
For Q4 FY24, management is guiding for revenue in the range of $5.1 billion to $5.5 billion, representing YoY growth of 142.6% at the midpoint. This compares to analyst estimates for $4.9 billion and growth of 124% going into the print.
There is a drop off in Q2 FY25, which is calendar year Dec 2024, and hopefully revisions flow through to increase this growth rate. It may seem far off, but this is a high beta stock that sees volatile price action based on signs of weakness or strength.
Management raised full year revenue growth to 109.3% for revenue of $14.9 billion. This is up from guidance for revenue of $14.5 billion last quarter.
SMCI also established full year GAAP EPS and non-GAAP EPS guidance for FY24. GAAP EPS is guided to $21.61 to $22.46, compared to GAAP EPS of $11.43 for FY23. This would mark a 92.9% increase YoY at the midpoint of guidance. Management expects non-GAAP EPS to be in the range of $23.29 and $24.09 for FY24. This would be YoY growth of 100.6% at the midpoint compared to FY23 non-GAAP EPS of $11.81.
GAAP EPS for March was $6.56 compared to analysts’ consensus of $5.16. This is 28.6% higher sequentially with $5.1 GAAP EPS in the previous quarter and is 328.8% growth from the year-ago quarter. Adjusted EPS was $6.65 for similar YoY and QoQ growth.
Looking forward, next quarter’s GAAP EPS is expected to be between $7.20 and $8.05 for over 122.3% growth from the year-ago quarter. Adjusted EPS of $8.02 at the midpoint will see similar YoY growth. This compares to analyst estimates of $7.16.
Margins
Gross margin was 15.5% in Q3 for gross profit of $597.4 million. The company is guiding to a lower gross margin next quarter. An analyst on the call implied it would be 13.5% to 14% next quarter.
Operating margin was 9.8% for operating profit of $378.3 million and adjusted OPM was 11.3%.
Net margin was 10.5% for net profits of $402.5 million. Adjusted net margin was 10.7%.
Adjusted gross margin was 15.6% for Q3, improved slightly QoQ from 15.5%, however, adjusted gross margin was down 210 bps YoY compared to 17.7% in the same quarter a year ago. On the adjusted gross margin declines, management stated it is due to product/customer mix and focus on market share gains.
Note: Margins on SMCI tend to be thinner than most semiconductors, which is a key topic of analysts’ focus during each earnings call. The CFO has stated the target margin is between 14% and 17%.
Cash Flow and Balance Sheet
Cash flow used in operations for Q3 was $1.5 billion compared to cash flow usage of $595 million during the previous quarter as the company grew inventory and accounts receivable for higher levels of business.
Cash flows from strong profitability was offset by higher Inventory, a large portion of which was received late in Q3, and higher Accounts Receivable from increasing revenues. The Q3 closing inventory was $4.1 billion, which increased by 67% QoQ from $2.5 billion in Q2 due to the purchase of key components. Capex was $93 million for Q3 resulting in negative free cash flow of $1.6 billion for the quarter.
On its balance sheet, the company reported $2.12 billion in cash and cash equivalents and $1.86 billion in debt, up from $726 million in cash and debt of $376 million in the previous quarter. Consequently, the net cash position stood at $260 million, declining from $350 million in the last quarter.
During the quarter, SMCI announced a $1.5 billion principal amount of convertible senior notes that will be due in 2029. The company also announced a public offering of common stock as SMCI raises capital to support operations, including purchases of inventory and other working capital needs, manufacturing capacity expansion and increased R&D investments.
Key Metrics:
Server and Storage Systems & Subsystems
Server and storage systems were $3.7 billion in revenue for growth of 218% YoY and was 96% of Q3 revenue.
Subsystems and Accessories were $152 million, up 27% YoY and was 4% of Q3 revenue.
Vertical Markets
OEM Appliance & Large DC: 50% of total revenues, down from 59% last quarter and up from 47% a year ago.
Organic (Enterprise & Channel), AI/ML: 49% of revenues, increasing from 40% of revenues last quarter, and slightly down from 50% of revenues a year ago.
5G, Telco & Edge/IoT: 1% of revenues, flat compared to last quarter and down from 3% of total revenues a year ago.
According to the CFO: “One existing CSP large data center customer represented 21% of Q3 revenues and one existing enterprise channel customer represented 17% of revenues.” This compares to last quarter’s customer concentration of 26% and 11%, respectively.
Inventory:
Inventory days increased to 92 days compared to 67 days in the previous quarter. The company’s Q3 closing inventory was $4.1 billion, which increased by 67% quarter-over-quarter from $2.5 billion in Q2 due to the purchase of key components.
This was asked about on the call and we detail it for you below. It’s also reflected in the steep, negative operating cash flow reported this quarter.
Geography:
All revenues were up by a wide margin QoQ.
United States was 70% of revenue, and increased 242% YoY and 3% QoQ.
Asia was 20% of revenue, and increased 257% YoY and 17% QoQ.
Europe was 7%, and increased 30% YoY and 3% QoQ
ROW was 3%, and was up 87% YoY and was down 11% QoQ
China accounted for 1% of total revenue.
Earnings Call:
Inventory increase:
The inventory days increasing doesn’t entirely explain the steep $4.1 billion in inventory. Rather, the company has to hold more inventory while waiting for key components related to liquid cooling. This is out of character for Super Micro to hold this much inventory and this will not be comfortable for the Street to accept given the company has diluted shareholders and raised debt in the past quarter.
Question
Aaron Rakers (Analysts)
Yes. I'll try and slip in 2 here, if I can. So I guess one of the just kind of housekeeping questions is a very significant increase in inventory this quarter. I know you said that it came in towards the end of the quarter. How do we think about the trajectory of inventory as the supply comes on? Do you expect inventory to stay at this level? Do you expect it to start to come down? I'm just kind of curious how we think that flow through kind of looks as you take on more supply […]
Answer
Charles Liang (Executives)
Two reasons we had to increase inventory: One is because Q4, I mean, June quarter, we will have a strong revenue growth; a second reason because we're preparing for high-volume liquid cooling. Again, we have more than 1,000 of 100k watt, I mean, liquid cooling rack we have to ship to customers in Q4. And liquid cooling as you know, is pretty new. So we had to prepare enough inventory so that we can deliver liquid cooling rack scale product to customer on time or with minimal lead time. So both factor, indeed, is a positive factor. And with our economic scale continuing to grow, indeed, our inventory average [ daily ], indeed, will slightly improve.
Answer
David Weigand (Executives)
Yes. So Aaron, my take on that is I hope that our inventory continues to grow because that means there's a reason behind it, so — and it's tied to sales.
–End quote
Here was another discussion around the inventory levels:
Question Nehal Chokshi (Analysts)
Congrats on a strong guide here. Talk about the guide here. Inventory increased $1.5 billion Q-over-Q. And Dave, as you mentioned, you'd like to see inventory increase. I do too because it's a strong indicator of things to come. And you guided June quarter to increase by $1.6 billion Q-over-Q. If I do this math, where I'm looking at the inventory at the quarter end and then the [ fourth ] quarter revenue, typically, it's around 60% to 70% of revenue. But with your March Q ending inventory and your current [ June, too, ] guidance, that equates to about 85% of projected revenue. So can you just explain what seems to be a little bit more usual inventory buildup given the revenue guidance range?
Answer David Weigand (Executives)
Sure. Absolutely. That's a fair question. So we actually got a substantial amount of inventory in the last week of the quarter, okay, which obviously, we're not going to be able to ship, but we took in $700 million in the last week of the quarter. So that's not something — that's something that has to do with when inventory arrives. And so we — it hurts our cash flow, but you know what, it doesn't matter, because we need that inventory for Q4 shipments.
Answer Charles Liang (Executives)
Yes. Again, 2 reasons, right? Q4, we will have a strong revenue, so we had to prepare for Q4. And also, I mean, liquid cooling, I mean, it's new. So we had to prepare enough safety inventory for liquid cooling demand for June quarter and September quarter as well. So that's another reason why we have a slightly higher inventory now.
Answer David Weigand (Executives)
Yes. And I want to add, Nehal, that, that's exactly why we did capital raises, too, is to prepare for these Q4 shipments, and — so that we could make those large purchases, and we hope to continue that.
–End Quote
Sequential Growth is the New Normal
My ears perked up on this comment, when the CEO was asked if they are capable of future sequential growth:
Answer Charles Liang (Executives)
Yes. As you know, traditionally, in the last 10 years, right, I mean, the September quarter and March quarter, always our soft quarter. But now with AI, we've been growing so strong. So we basically are able to grow sequentially. So although March and September be a little weak, but basically, because of strong AI growth and our market share growing, so the sequential growth will become the normal. And basically, I mean, we have even better technologies than before ever, and now economic scale become much bigger. Malaysia campus production will be ready by end of this calendar year, so we see a lot of positive factors to grow our business.
–End Quote
SMCI Likely to Raise more Capital
This is likely the comment that caused the stock to go down 10% AH despite the strong beat and raise. An analyst asked the CFO if he foresaw a need to raise more capital. Here was his reply:
Answer David Weigand (Executives)
Yes. So the way I would answer that is, is that I hope that I have — I need more capital, Jon, because that means that we're booking — that we're growing revenues even faster. So we've got capital adequate to get us through the current market, which means today. But in a week, that — we hope that, that changes, and we hope that we've got orders that require even more capital. So all I can say is I hope that — I'm hoping for the need for more capital.
Answer Charles Liang (Executives)
Yes. We believe our revenue will continue to grow strong. And that's why we need more capital to grow faster. If we grow 20%, 30%, we may have enough capital now, but it will grow much faster. Then for sure, we need more capital to grow stronger.
–End Quote
My comment: this is not what the market wants to hear right now, which is that you have to raise capital to fund growth. Supermicro is an incredible company situated perfectly between hyperscalers and the world’s best design companies. However, this is not the right macro environment to need to raise capital. Not even an AI bullet train can change those facts.
Conclusion:
There is no doubt, we rode a phenomenal wave with Supermicro over the past few months. The comment that sequential growth will be the new norm is music to our ears, as growth investors. However, we can’t fight the Fed. This is a good time to put the surf board down for a little while and let the next wave gather strength before we attempt Supermicro again.
In our pre-earnings writeup, I had stated: “it’ll be negative cash flow margins and/or dilution that penalizes the stock,” as well as: “The stock seems to be on a never-ending winning streak, however, what could be Super Micro’s Achilles heel is the cash issue — as the company must grow capacity to keep up with the revenue growth, yet to do so will require cash.”
If it were just about inventory, to where the shipments came in late in the quarter but was spoken for the following quarter, then that would not be an issue. From this report, the concern is the large appetite the company has to raise more cash to support growth. The tie-up on inventory for the direct liquid cooling components is an additional concern but it’s the primary issue around having to raise more cash that ultimately doesn’t meet our criteria at this time.
As you are aware, Super Micro is a high beta stock and we plan to adhere to our line in the sand. The good news is that we’ve made sizable profits and plan to put those to work at lower levels.
In our last write-up, we called Super Micro the “AI Bullet Train” due to neck-breaking growth rates. Last quarter, the company reported a notable 103% year-over-year growth, with revenues surging to $3.66 billion. Management's projections are for nearly double the growth, anticipating Q3 revenues to range between $3.7 billion and $4.1 billion, or growth of 204.7% year-over-year at the midpoint. Consensus estimates are for $3.92 billion, for 206.3% growth expected in tomorrow’s print.
Such predictions underscore a consistent upward revision in earnings guidance, reflecting anticipated strong growth for the remainder of the fiscal year and extending into 2025.
To understand how we got here, it was through multiple analyst revisions. For example, the March quarter started with estimates of $2.1 billion in August for growth of 45% but were revised by 160 points (!) to $3.92 billion for growth of 205.7%. The June quarter was seen upward analyst revisions increase by 92 points, and the September quarter’s numbers revised by 82.7 points in the span of one quarter!
Here's what that looks like:
The takeaway is that Super Micro’s bullet train-like price action is based on upward revisions, which is unique from earnings beat/raise or simply strong estimates. In fact, the stock was down (-12%) following its last earnings report and is now up 80% since that call with gains as high as 135%. This is a stock that does not rely on earnings pops, like most growth stocks, and rather, it requires a bit of tenacity as the intra-quarter changes have been quite profitable.
With that said, below is a deeper look into SuperMicro, including the underlying factors contributing to these impressive figures ahead of tomorrow’s Fiscal Q3 results and the risks that accompany this high beta stock.
Revenue and EPS
Last quarter, the company’s Q2 FY2024 revenue grew by 103% YoY to $3.66 billion. Management Q3 guidance is in the range of $3.7 billion to $4.1 billion, representing YoY growth of 204.7% at the mid-point. The consensus analysts’ estimate is $3.92 billion, representing a YoY growth of 206.3%. The guidance has increased consistently through recent months as strong growth is expected for the remaining quarters of the fiscal year and into 2025.
The drop off in Q2 FY25, which is calendar year Dec 2024, will be key to keep an eye on following this earnings report. It may seem far off but this is a high beta stock that gets slammed on any weakness. The opposite can also happen, which is that we see more revisions which supports the price action extending, as outlined in the introduction above.
Fiscal year estimates were revised upward over the past year by 87.8 points for growth of 105.5% and revenue of $14.6 billion. This is higher than the midpoint of management guidance for revenue of $14.5 billion, at the midpoint. Next fiscal year ending June 2025 has been revised upward 33 points to 44% growth. We will be watching these estimates closely as we manage our position intra-quarter throughout the next few months.
GAAP EPS for December was $5.10 compared to analysts’ consensus of $4.90. This is nearly 80% higher sequentially with $2.85 GAAP EPS in the previous quarter and is 62% growth from the year-ago quarter. Adjusted EPS was $5.59 for similar YoY and QoQ growth.
Looking forward, next quarter’s GAAP EPS is expected to be between $4.79 and $5.64 for over 240% growth from the year-ago quarter. Adjusted EPS of $5.60 at the midpoint will see similar YoY growth.
Note: Normally, we’d be hesitant to see the slowing growth on both top line and bottom line pictured above as moving from hypergrowth to average growth tends to cause a re-rating in valuation. However, we’d like to see if SMCI will continue its pattern of seeing upward revisions given the strength of the AI trend.
Margins
Gross margin was 15.4% in Q2 for gross profit of $564.4 million
Operating margin was 10.1% for operating profit of $371.5 million and adjusted OPM was 11.3%
Net margin was 8% for net profits of $295 million. Adjusted net margin was 9%
Adjusted gross margin was 15.5% for Q2, compared to 17% in Q1 and 18.8% in the same quarter a year ago. On the gross margin declines, management stated: “in order to take market share, we will take opportunities by being more competitive on pricing.”
The gross margins guide for Q3 is expected to be “slightly lower than Q2 levels.” This indicates another YoY and QoQ decline in gross margins if the management’s guide is correct. The goal is that margins will return to baseline once the company is operating at scale. However, it’s worth mentioning that stocks with thin margins tend to underperform in a Fed-driven market. This is one reason we will adhere to stops with SMCI.
Margins on SMCI tend to be thinner than most semiconductors, which is a key topic of analysts’ focus during each earnings call. The CFO has stated the target margin is between 14% and 17%.
Cash Flow and Balance Sheet
Operating cash flow reached (-$595) million, with a margin of (-16.2%). This performance contrasts with the positive margins of +12.8% and +9% reported in the September quarter and the June quarter, respectively.
The CFO explained that the cash outflow in operations for Q2, totaling (-$595) million, was a shift from the $271 million generated in the prior quarter. Despite robust profitability and an increased level of accounts payable, this was counterbalanced by a rise in inventory and accounts receivable, driven by preparations for Q3 and shipment timings in Q2.
Cash flow will be a primary focus on the call as any additional quarters that report negative free cash flow will force investors to price-in future stock dilution and cash raises. This line item can cause the stock to be re-rated should it continue to be weak.
Free cash flow was also negative at (-$610) million, representing a (-16.6%) margin, compared to positive margins of +12.7% last quarter and +8.4% in the corresponding quarter of the previous year.
On its balance sheet, the company reported $726 million in cash and $376 million in debt, up from $543 million in cash and net debt of $146 million in the previous quarter. Consequently, the net cash position stood at $350 million, slightly down from $397 million in the last quarter.
The company boosted its cash reserves with an equity offering. As stated by the CFO, the proceeds from the equity offering will be used to strengthen working capital, continued investments in R&D and expand global capacity.
Key Metrics:
In the latest quarterly financial update, the OEM Appliance and Large Data Center segment led the company's revenue streams, contributing $2.15 billion, which accounts for 59% of total revenue. This segment saw significant growth of 175% year-over-year and 83% quarter-over-quarter.
The Organic (Enterprise & Channel), AI/ML segment followed with $1.48 billion, making up 40% of the revenue and growing by 55% year-over-year, fueled by enterprise AI initiatives and CPU upgrade programs. The 5G, Telco, and Edge/IoT sectors, however, represented just 1% of revenue at $35 million.
In terms of the revenue mix, server and storage systems generated $3.4 billion and comprising 94% of the quarter's revenue, reflecting a year-over-year growth of 107%. Subsystems and Accessories contributed $229 million, accounting for the remaining 6% of revenue and marking a 61% increase from the previous year.
Inventory management improved, with inventory days reducing to 67 from 91 in the previous quarter, indicating a tightening of supply as noted by the management.
What to look for in the earnings report:
1) Declining margins are going to be a key focus of analysts.
Notably, Super Micro has weaker margins than Wall Street prefers and tend to be weaker amongst its peers. It’s no surprise when analysts pick up on this during the call and slide in a question or two on it.
Last quarter, there was a large revenue beat that did not flow through to a higher gross margin or operating margin, and in the following Q&A session, the CFO stated: “And so, at this time we are we are growing really quickly. And in order to do that and in order to take market share, we will take opportunities by being more competitive on pricing.” The CEO followed up with: “The good thing is that when we continue to grow our economies of scale, our operation margin indeed will be still able to keep in healthy position.”
Super Micro is primarily air cooled right now, yet liquid cooling is growing. Per the CEO in the opening remarks, we can expect major updates in the coming quarters on their progress: “By this June quarter, we will have high volume, dedicated capacity for manufacturing 100 kilowatt to 120 kilowatt racks with liquid-cooling capabilities, providing DLC, direct liquid cooling racks capacity up to 1,500 racks per month and our total rack production capacity will be up to 5,000 racks per month by then.” To read more on liquid cooling, reference our previous Super Micro analysis here.
3) Conservative Commentary by Management
This word, “conservative,” has been continually referenced by management in recent quarters. We hope to continue to hear the twelve-letter C-word from SMCI tomorrow evening!
David Weigand
“[…] And so really as Charles mentioned earlier, our only constraint is supply. However, the good news is, the supply is improving. And so, to your point, we have to be somewhat conservative, because we are constrained still by supply.”
Conclusion:
Despite facing declining gross margins, the firm's substantial year-over-year revenue and EPS growth underscore its product strength and positioning in a fiercely competitive environment. In case it’s not clear, Super Micro is an outlier and it all comes down to product differentiation, which you can read about here.
The stock seems to be on a never-ending winning streak, however, what could be Super Micro’s Achilles heel is the cash issue — as the company must grow capacity to keep up with the revenue growth, yet to do so will require cash.
Due to the high beta nature of Super Micro, I foresee us trying to ride this wave a few more times in the coming years. We will play this one with the understanding that volatility goes both ways, armed with the information that it’s the upward revisions that reward this stock (mainly intra-quarter), and it’ll be negative cash flow margins and/or dilution that penalizes the stock.
Overall, for our risk profile, entries in high beta stocks are accompanied by a strategy and with predetermined stops. You can read more about our line in the sand here along with upper price targets.
Chad Shoop, Equity Analyst for the I/O Fund, contributed to this analysis
Marvell’s management team did an excellent job of acquiring Inphi and executing. Typically, we avoid M&A for a year to allow the financials to merge, yet in this case, leaning into the acquisition was a good choice.
The Marvell management team’s execution skills are needed once again because Marvell has an opportunity to greatly increase its revenue and profits if management can execute in a new market one more time. The opportunity is a new architecture called CXL that disaggregates memory from the CPU. CXL is attracting a lot of attention at industry events, such as Hot Chips 2022, because it’s focused on optimizing one of the most expensive parts of the data center – which is memory.
Before we go into the 2023-2024 Marvell product road map, and why it’s key to the company’s future, I want to discuss the fiscal Q2 2023 earnings.
Fiscal Q2 2023 Earnings Overview
The market is concerned over Marvell’s data center guidance of 20% growth next quarter. This is a slowdown from the most recent quarter at 48% YoY growth and earlier quarters at >100% growth.
At an estimated $600 million, it will also mean a sequential decline both from Q2 and Q1, which were at $643M and $640M, respectively. Marvell stated it’s the on-premise business weighing on their cloud data center business and supply issues (more below).
Notably, Q2 of last year was an important moment for the company when 56% sequential data center growth grew from $277 million to $434 million in the span of three months following the close of the Inphi acquisition in April 2021. From there, the company has sustained Inphi’s already high growth levels for over a year.
The company is now at an annualized run rate of $6 billion, which the CEO reminded analysts, was the target for October of 2023. The company met the target originally provided at the October 2021 Investor Day one year earlier than expected. Notably, this was six months after Inphi was closed so M&A not a factor here.
Marvell’s Segment Overview:
The data center represents 42% of revenue at $643 million and grew 48% year-over-year.
The carrier infrastructure segment, which is wired and wireless and reflects 5G growth, reported 45% YoY to $285 million.
Enterprise networking grew handily at 53% YoY to $340 million and is expected to grow at 70% next quarter. We break this segment down below.
Consumer was down (1%) to $164 million and is expected to be down (10%) next quarter. Marvell has exposure to the storage market and this can weigh on the more robust segments.
Automotive was up 46% YoY to $84 million and is expected to be up 40% YoY next quarter. We also break down this segment below.
Marvell Financial Overview
Marvell was reporting negative top line revenue when we first covered it in 2019 and Marvell took another hit on revenue during Covid before accelerating to the 50%-74% revenue growth range.
The current quarter’s top line revenue in Q2 was at 41% which is a deceleration from Q1 with 74% revenue growth. The company guided for 29% year-over-year growth, which was a slight miss as analysts were expecting 30.3% growth in the fiscal Q3 quarter. The company reported EPS in line with adjusted EPS of $0.57. The guidance on EPS was a slight miss, however, at $0.59 reported versus $0.61 adjusted EPS estimated.
Semiconductors make a tougher investment as analysts can’t model too far into the future beyond what management teams provide. That is why there were many questions looking for help with how to factor in the “acceleration” in the data center the Marvell team is expecting in Q4 and what this will mean for CY2023.
An analyst asked if they can assume 10% QoQ in the data center for $1.7 billion overall revenue and the CEO said it sounded “a little on the high side.” This has led to analysts modeling $1.65 billion in revenue in Q4, for 22.5% growth. Therefore, despite a single-digit acceleration in the data center segment, there will still be a top line deceleration, if today’s forecast does not change.
The company’s margins and cash flow are a bright spot, and I believe this is being overlooked. If we get an acceleration in the data center into next year, then Marvell is fundamentally a much stronger company than it was during the previous data center streak.
On a GAAP basis, the gross margin was at 51% in the most recent quarter, up from 35% in the year ago quarter and up from 46% in FY2022. The company is guiding for the same GM of 51% next quarter.
The GAAP operating margin has improved quite a bit YoY to 8.3% in the current quarter compared to (25%) in the year ago quarter. This is also an improvement from Q1 with GAAP OM of 4.80%. The adjusted operating margin “hit a record” at 36.5% and is guided for 37% next quarter. Stock based compensation was at $139 million in the most recent quarter.
Cash flow is also improving with operating cash flow at $332 million, or 22% of revenue. This compares to $194 million last quarter and $819 million in FY2022. However, the company carries debt of $4.6 billion and has $617 million of cash on the balance sheet. This is a 1.8X net debt to EBITDA ratio.
Therefore, there has been substantial improvement yet Marvell does have a weaker debt profile than a company like AMD or Nvidia.
Source: YCharts
Note on Supply:
Marvell is aligned with AMD in that they believe supply chain issues will ease in Q4 and into 2023. Here is what Marvell said in the opening remarks:
“Therefore, for our overall data center end market, we project revenue in the third quarter to decline sequentially in the mid-single digits on a percentage basis. However, we expect our data center revenue in the fourth quarter to increase on a sequential basis, anticipating an improvement in supply and new product ramps in cloud.”
Here is what AMD said:
“The visibility with our customers, especially our large cloud customers’ second half of this year into next year is very good. And we’re planning really for the next four to six quarters, and that gives us good visibility” and later provided many references toward supply coming online in Q4, such as: “But overall, the 7% increase [in gross margin], I think, is very well supported given all of the new product ramps that we have going on in addition to some additional supply that’s coming in as we get into the fourth quarter.”
It never hurts to have two management teams agree on the larger broad-based issue. However, since those reports, we’ve seen analysts cast doubts on the effects of macro for the rest of the year: “[Mizuho analyst Rakesh] checks show hyperscale orders are seeing "pushbacks" but no cancels, with Q3 trending flat quarter-over-quarter and Q4 "potentially soft." Rakesh lowered estimates for AMD "with macro headwinds clouding the near-term outlook."
Marvell’s Products:
In six brief years, Marvell has pivoted away from consumer (storage) products as the revenue mix was previously 62% consumer/38% infrastructure to being 11% consumer/89% infrastructure today.
This was driven partly by hyperscalers building data center infrastructure and AI/ML driving the need for faster data speed. Inphi also contributed to this.
Data Center Segment
PAM Solutions:
Marvell offers 200-gig and 400-gig PAM-based electro-optics — and the company recently added 800-gig solutions. This market sees tailwinds from the need for more bandwidth as the electro-optics connect short distances and long distances to increase data rates. PAM4 has replaced NRZ data transmission with the benefit of doubling the bit rate.
Hyperscalers are going through an upgrade cycle that requires high bandwidth and port density. PAM4 connects networking ASICs and machines, like servers and AI machines. Digital-based PAM4 uses analog-to-digital converters to clean up the signal in the digital domain before converting it back to analog to transmit.
Artificial intelligence and machine learning drives demand for the 800-gig PAM to increase the speed of input-output and to process the data flows. This doubles the throughput (bandwidth) due to an 8x100Gpbs optical transceiver for inside and between AI clusters.
In the fiscal Q1 results ending in April, management had stated: “our first quarter results benefited from a ramp in volume shipments of our 800-gig PAM solutions at two large customers.” The company has also stated that their products will see increase demand with the release of more powerful CPUs.
COLORZ 400:
COLORZ allows regional data centers to be linked together in the same metro region to function as one single mega data center. COLORZ silicon photonics technology allows data centers in the same metropolitan region to function like a mega data center through a “network fabric.” This facilitates faster edge computing within an 80/120 km distance for 30-megawatt data centers as they will be linked together and function like a 120-megawatt data center.
“As artificial intelligence (AI), machine learning (ML) and high-performance computing (HPC) applications continue to drive greater bandwidth requirements, cloud-optimized 400G solutions are needed to support high-speed data center interconnections. These requirements can only be met through high bandwidth connectivity offered in a small, cost-effective form factor. The Marvell COLORZ II 400ZR enables cloud data centers the ability to increase the speed of data movement while keeping the power and cost low.”
Another press release stated the company shipped 100,000 units.
Here is what was said on the call about how/why the growth in the data center can continue:
Harlan Sur
Good afternoon. Thanks for taking my question. On the cloud optical connectivity business, this is both inside and between data centers, the upgrade cycles have been this really great multi-year tailwind for the team.
And if I look into next year, I believe that there's still at least one of the top four US hyperscale titans that's going to start the 400-gig PAM4 transition. You still have China CSPs that need to fire. You've got multiple customers on DR that's going to fire as well. Historically, like these transitions, I don't think have been impacted by a slowing macro demand environment. They're viewed as, I think, very strategic.
But is that how your cloud customers are thinking about these upgrades and your views on continued upgrade momentum in this segment for next year? And just relatedly, is the Innovium team on track to drive $150 million in revenues this year?
Matt Murphy
Hey. Thanks, Harlan. Yes, I think the first part of it, you got pretty well in terms of the transition on 200 and 400 gig PAM4 inside the data center. And then, the new ramps we're seeing in 400 gig ZR for DCI between data centers.
What I'd add on top of that is — which has been extremely strong and also, in some ways, a little bit of a constraint we've seen in terms of being able to keep up is, the demand on 800 gig, which is happening right now really around, obviously, very advanced AI workloads.
That is an area where, if we could obviously produce more material, we would be shipping it in Q3. So that's also a positive trend. So you've got sort of the transition going on all the way up to 800 gig, and that continues to look pretty good.
NOTE: Innovium is an acquisition that closed in 2021 and at time of acquisition was expected to add $150 million in revenue for CY2022/FY2023.
Compute Xpress Link (CXP): 2024-2025 Data Center Catalyst
Marvell is launching a new product line called CXL, which will improve how data centers add memory. Right now, a server must be opened to add DRAM and the DIMM slots are limited in number and don’t pass service history or bit-error history, which is needed by hyperscalers.
Memory pooling allows memory to scale independently from processors by taking memory for a task and then releasing the memory. The new fabric removes the need for local DRAM, which adds a bit of latency from 100ns to 140-160ns, however, there’s a possibility of adding a CXL accelerator to be more “cache coherent.”
The CXL switch will be used to accelerate protocol-level processing across ethernet, DPUs, SmartNICs and solid-state drive controllers (SSD).
What Marvell is proposing with CXL is a new server architecture to “dynamically assign memory resources between servers.” The result is boosted memory bandwidth and also the ability to enable memory pooling. The company sees a future where a new architecture will separate compute, memory and I/O racks with the interconnect being CXL.Partially-disaggregated racks are expected to deploy in 2024-2025.
Marvell is at the forefront of the shift toward “disaggregate memory from the CPU” because it currently supplies the optics that this new fabric will disrupt. Inphi is the leader in silicon optics, PAM-4, and the encoding of PAM-4 for PCIe 6.0.
2024 seems like a long ways off yet the market will be paying attention to this In Q2/Q3 2023.
“As you recall from our discussion last quarter, we see CXL as the next big evolution in cloud data centers that will enable us to increase our reach into the memory ecosystem and presents a multibillion-dollar SAM expansion opportunity for Marvell.
This includes a host of new products such as CXL expanders, cooling devices, switches and accelerators and the potential to embed CXL IP and a broad range of our data center products. Events and presentations at FMS strongly validated our excitement around CXL. This is the hottest topic at FMS with standing room-only presentations by many leading industry participants.
The Marvell booth, we demonstrated the industry's first CXL memory pooling solution, addressing the challenges related to memory scaling and cloud data centers. While the industry is still in the early stages of CXL adoption, we are working on closing significant opportunities right in front of us at key customers and envision a strong design win pipeline.”
Why Marvell for CXL?
There are a handful of companies going after the CXL opportunity. Marvell could be front runner as the company already works closely with memory OEMs by supplying HDD controllers, SSD controller and preamplifiers. The company also has an aggressive PCIe roadmap with the company shipping Gen 5 sockets whereas most SSD device are shipping Gen 4 solutions. Marvell is already investing in Gen 6, which in turn, attracts more Tier 1 memory OEMs.
Marvell acquired Tanzanite, a developer of advanced CXL technologies. The company plans to expand to CXL expanders, cooling devices, switches and accelerators.
The company has stated this will drive “a multibillion-dollar PAM expansion opportunity driven by CXL overtime.” (Note: Marvell is referring to PAM, their premiere product)
We will focus on this more next year. You can listen to a recent tech talk here on CXL. The presentation is located here. This is an article about Microsoft’s interest in CXL with a statement that “50% of their server costs are taken up by DRAM.”
Carrier Infrastructure:
The OCTEON processors and platform is an Arm-based compute architecture for embedded applications, such as wireless networking equipment including 5G, including switches, routers, firewalls and monitoring solutions.
The OCTEON DPU is used with SmartNICs and security accelerators with a 5nm design that delivers to the infrastructure industry the same processing node as consumer smart phones and high performance computing and shipped in 2021. The most recent release from last year was the OCTEON 10 DPU and Prestera carrier switches which combined consumes 50% less power than competitors (according to Marvell).
Marvell’s processors help 5G networks meet latency and bandwidth demand while also allowing the networks to upgrade as cellular standards evolve. Marvell also offers customized solutions, which is ideal for Tier 1 customers who can combine their IP with Marvell’s Arm v8 processors and accelerators.
Recently, Dell and Marvell partnered to develop a server-class accelerator card for 5G base stations based on Marvell’s arm-based OCTEON Fusion processor. The hardware accelerators deliver more processing power including processing solutions for smart radio heads to support massive MIMO antenna rays.
We wrote about MIMO a few years back in a reference guide: “Massive Multiple Input and Multiple Output (MIMO) sends the data through multiple data streams called layers, which increases parallelism and throughput. MIMO helps avoid lost signals with multipathing, which allows the base station to send multiple copies of the same signal for increased redundancy.
Note: The antenna array is one fundamental change to 5G infrastructure. The initial 5G rollout will use existing cell towers, however, newer, dedicated 5G network infrastructures will require many more antennas than used in previous generations. Read more.”
The distributed unit (DU) shares the load with the radio unit by running L1 functions on the RAN protocol. Marvell has been a proponent of OpenRAN with the O-RAN platform, which is an open protocol and open platform that allows Marvell’s hardware to be used with various software vendors. Facebook (Meta) is a partner with Facebook Connectivity.
DPU processors, or digital processing units, are gaining traction for 5G transport, 5G RAN intelligent controllers, edge computing and cloud data center workloads. These hardware accelerators enable high speed connectivity and can improve packet processing rates by 5X. DPUs are ideal for power sensitive edge applications. Marvell’s strength in DPUs is one reason it may be able to stave off competition, which in the narrow field of 5G base stations includes Qualcomm/HPE and Analog Devices. Beyond 5G, Marvell has other competitors for DPUs such as AMD/Pensando and Nvidia.
Regarding 5G, over 7 million of the Octeon processors have been used in 3G, 4G and 5G base stations with Tier 1 customers. In the past, we reported that Samsung and Nokia use Marvell, and supplying these particular companies was a tailwind when Huawei was blacklisted. More recently, Marvell has stated they have design wins with four of the top five global OEMs and next-tier OEMs building base station equipment. These design wins are based on the 5nm platform.
Marvell uses TSMC for the 5nm OCTEON DPUs and this is an advantage because Marvell has the 5nm now and is able to move quickly on a 3nm release.
Notably, 5G has been a long time coming but I do believe it will reward investors over the next few years. Technavio has a CAGR of 67% for 5G equipment through 2025. The growth trend of 5G/edge computing is not one that we plan to complacent on as it will provide the next leg up for substantial capex spending similar to data center capex spending.
Enterprise Networking:
Marvell sells ethernet switches and ethernet PHYs to IT managers and networking equipment manufacturers. The company uses DSP technology for CAT5e ethernet cables to supply data rates up to 5Gbps with support for CAT6 and CAT6a.
Management discussed on the call that the main driver for this market right now is wireless, specifically WiFi 6 as the wireless rate line is now faster than the wired rate. The call also pointed toward content per port going up in the transition to multi-gig. According to the CEO, “it's not like 10%, 20%, 30%. It's sort of multiples on a per port basis of where it was before.”
Increased enterprise share and content gains from wired and wireless enterprise networking drove 53% YoY revenue growth and 19% QoQ revenue growth.
Automotive:
Similar to the networking that Marvell supplies enterprises and the data center, Marvell also supplies auto manufacturers with ethernet PHY transceivers, camera bridges and switches for in-vehicle networks. This is used for things like collision detection, lane warnings, and autonomous driving.
Marvell believes Ethernet will be the backbone for connected and autonomous vehicles to connect the electronic control unit (ECUs), cameras, sensors, and central compute devices. The Ethernet device is called Brightlane.
ON Semi has partnered with Marvell on use cases such as pairing a standardized protocol, such Ethernet PHY, with ON’s portfolio of ultra-dynamic range image sensors.
Automotive was up 46% to $84 million, yet was down 6% sequentially. Management cited supply issues rather than demand. Marvell counts eight of the largest 10 OEMs worldwide and 36 OEMs total. The company believes revenue growth will be 40% next quarter.
Note on Consumer Market:
Marvell sells hard disc drives (HDD) and solid state disc (SSD) controllers. This is a weaker segment, declining 1% YoY and 8% sequentially to $164 million. For next quarter, Marvell expects revenue to be down 10% YoY and flat sequentially.
Conclusion:
There is a new, powerful trend on the way that is on par with the cloud computing trend. This trend of edge computing will rely on distributed computing rather than centralized processing. Both will exist and rely upon each other but edge computing will have a stronger growth trend when it breaks ground (by virtue of being new/rapidly expanding). Much of this will be in sync with the 5G buildout.
Marvell has the potential to be a strong stock during this buildout as the company provides the base station hardware, supports MIMO antenna rays, beamforming, and accelerates 5G transport and controllers which results in high-speed connectivity.
The company also provides electro-optics and silicon photonics for increased data rates and a network fabric for edge computing. The edge is defined as many things, but what all definitions can agree on, is that the edge needs superior connectivity/networking. Electro-optics, silicon photonics, DPUs, SmartNICs and ethernet in the data center are a warmup for Marvell supplying edge servers and edge devices. As this occurs, the demand for Marvell’s product suite will increase.
In addition to this, Marvell is thinking outside the box by focusing on restructuring memory while most companies are focused on more powerful chips. CXL drives down costs on DRAM and is likely to rapidly adopted by hyperscalers once it becomes available. There’s no guarantee that Marvell will be the one to win the contracts but it’s certainly a front runner.
MongoDB carries a bit of nostalgia for our team as it was one of the first stocks we covered after launching the premium site in July of 2019. At the time, there were concerns DocumentDB would rival Atlas yet Amazon had declared Atlas the segment winner at the open-source conference OSCON that month.
The company reported revenue of $285 million compared to estimates of $267 million for a $18 million beat. This represents growth of 57% and is the highest growth rate since Q3 2019. Particularly, it proves MongoDB can accelerate post-Covid which is rare among its peers.
The company had a sizable beat on adjusted EPS of $0.20 compared to estimates of ($0.10) EPS, which was a ($0.30) beat. The company’s adjusted operating income was $17.5 million compared to a loss of $2.8 million in the year ago quarter.
On a GAAP basis, the company reported EPS of ($1.14). Notably, GAAP losses increased this quarter to ($75.9) million compared to a GAAP loss of ($61.4) million in the year ago quarter. This is due to stock-based compensation expenses of $91 million with shares increasing from 61 million to 67 million over the past 12 months.
Gross margin expanded from 72% in the year ago period to 75% for gross profit of $214.3 million. The company stated this was due to increased efficiencies in Atlas. The company has $1.8 billion in cash and cash equivalents of which $456 million is cash. The company’s cash flow this quarter was $8.4 million, which is down from $16.8 million sequentially. Operating cash flow was $11.6 million compared to $22.3 million last quarter which management explained is partly because Q4 has more days than Q1 for consumption.
MongoDB is estimating a $30 to $35 million headwind. With that said, MongoDB reiterated guidance for the full year at $1.18 billion which would imply the company had expected to beat by $30 to $35 million. Management is also forecasting a $4 to $5 million headwind for next quarter yet was still able to raise guidance from $277 million expected next quarter to $279 million-$282 million provided as new guidance. The headwind of 1% sequential growth this quarter came from slower-than-expected customer growth in self-serve and mid-market channels in Europe yet could spread to impact all geographies.
The earnings guide for next quarter is for adjusted EPS of ($0.31) to ($0.28).
As we had covered three years ago, the MongoDB story centers around Atlas. This was the fourth quarter of over 80% growth and it now comprises 60% of revenue compared to 51% of revenue in the year-ago quarter.
There were ample questions about why MongoDB was able to weather the weakness in consumer-facing businesses better than Snowflake. The management feels they are more insulated because their consumption is tied to the value and usage of the applications and databases are not something that can be shut on, shut off or moderated by choice.
Here is the exact quote:
So the people are not using their application, something has gone wrong. So the more they use the application, the more value they’re seeing. So there’s a direct correlation between the value they get from the apps running on MongoDB and the value we get from those customers. Other software companies that you mentioned, I think are being forced to consider alternatives to be because there’s a trend where there’s a slight mismatch between price to value because as they suck in more data, it’s not completely clear how much incremental value that data is providing. So we don’t see that problem.
Probably the biggest contrast between Snowflake’s call and MongoDB’s call was that Snowflake noted a slowdown in a few of their biggest customers while MongoDB noted only a slowdown in their self-serve and mid-market. MongoDB also emphasized they are well diversified with six times more customers than Snowflake “tens and tens of thousands of customers” and due to representing more industries.
Conclusion:
We had stated the following:
“As stated above, MongoDB’s cash flow margin is what can keep the stock strong given stock based compensation is weighing on GAAP operating margin. We want a meet/beat on revenue, strong Atlas growth (bonus for acceleration) and we must continue to have a healthy, positive cash flow margin.
Analyst consensus has MongoDB reaching profitability on an adjusted basis by calendar year 2023.”
MongoDB proved they can become profitable on an adjusted basis in calendar year 2022 so that’s a plus. The company maintained its cash flow positive status. There were beats and raises alongside conservative guidance, which was really the ticket this quarter. It was easily the better report over Snowflake primarily because Snowflake has begun to concern analysts that the exposure to consumer could cause the company to become discretionary (more information is needed beyond one quarter). Meanwhile, MongoDB clearly illustrated this quarter its document databases are not discretionary.
In the packed 6-week calendar of earnings reports that comes from our tech universe, we like to note to our Members which ones we think are the strongest. To those who are new to our site, we don’t do news cycle-level earnings coverage as we are building positions rather than feeding a content machine. We think there is too much information coming at investors and it creates information overload. Therefore, this analysis is to say – “hey, you might want to take the time to look at AMD’s report because we think it was pretty special” – and we track dozens of earnings reports before we determine this.
AMD’s report was impressive on many accounts. Primarily, it’s because management is laying a solid foundation by leveraging general-purpose computing success for a workload specific product road map. AMD refers to this as compute differentiation and also adaptive computing in their decks, where the software defines the hardware’s workload rather than the other way around.
The company is optimizing CPUs and GPUs for cloud computing separately from technical computing workloads, for example. This means not only is AMD maximizing the large footprint of data centers right now but is looking at where data centers will be in the next 5-10 years, which will be optimized workloads, with CPUs and GPUs that are optimized to best serve cloud computing, high performance computing (HPC), gaming/Metaverse, and machine learning as unique workloads and by expanding AMD’s already strong product line in general-purpose CPUs and GPUs.
The $8 billion buyback is great news as it’s $4 billion higher than what the market had priced in. However, as Bradley discusses below, the Xilinx acquisition is a $35 billion all-stock transaction. He discusses what both mean for the stock in the near term and he also provides a health check on the financials.
I’m going to focus on how AMD continued to take server market share in this update as we haven’t revisited this for about six months and quite a bit has changed in favor of AMD investors. I’ll try to keep this analysis enjoyable as semis can sometimes be weighed down in technical jargon.
Regarding Xilinx, this company is a critical pillar to AMD’s future strategy so we will get to this in our next AMD update likely at the start of H2. Our goal with AMD is to figure out how big of a position the company can hold in our portfolio this year and in subsequent years. The Xilinx deal will play a big part in how we determine our allocation in future years as it has the potential to drive incremental gains in many segments, especially automotive/Embedded, plus perhaps 5G. Keep an eye out for that report in a few months. If you don’t want to wait that long, you can read what I wrote when the acquisition was first announced here.
A Trip Down Memory Lane …
How did AMD get here? I highly recommend every Member listen to our AMD webinar from July as we discuss the story of AMD’s “EPYC” comeback.
I’ll briefly summarize some of what was discussed in the webinar before we go into the current generation of the Zen architecture and the specific workloads that AMD plans to introduce to potentially take more market share from Intel and to also nibble at Nvidia’s near-monopoly in GPUs (I’m not worried about Nvidia, hence the word “nibble”).
The Zen architecture was introduced under Lisa Su in 2017. These processors are chipset free and fully integrated. Communication between CPUs is done through Infinity Fabric protocols. The result of the new architecture was more energy efficiency and the ability to execute more instructions per cycle. The Zen 1 architecture had 32 cores and 64 processing threads so more cores than Intel. There are 128 lanes of I/O for storage, networking and PCIe expansion. When you add two CPUs, Infinity Fabric is used as an interconnect to increase connectivity speed. In this case, there are 64 cores.
The first generation of the Zen architecture helped prove AMD still had a pulse and a heartbeat (however faint with 2% CPU market share) but it was between 2019-2020 when the company found its wings again and catapulted to 8% of the CPU market share. Today, it’s market share stands at an estimated 12%. The company is unlikely to 6X again but can it take the lead someday, is the question. We think it’s a possibility if management’s execution continues.
The company released the second generation of its Zen architecture and this is when AMD started to clearly outpace Intel in terms of computing power, memory and energy use – all at a lower cost.This was due to multi-chip modules that combine a 7nm with a 14nm to use the most advanced technology when and where it’s needed most by leveraging the more mature process node. The L1 cache and L2 cache locations in the core and across the core also helped the company beat Intel on memory bandwidth.
Intel was still producing a 14nm chip with a 10nm supposedly on the way.Essentially, AMD leapfrogged the incumbent with a product that is more power efficient and allows for more cores per chip. Because 7nm are twice as dense as 14nm, AMD was able to release a 64-core server chip and 128 threads rather than AMD’s previous 32-core server chip. Up until early 2019, Intel’s offering had been a 28-core server chip and 64 threads.
AMD’s products at the time were the EPYC Rome processors and the 7nm Radeon Instinct MI60 and MI50 accelerators that are built around the Zen-2 CPU microarchitecture. Here's the main thing about that point in time — Intel was expected to catch-up with a comparable 10nm release planned for Q2 or Q3 2020 called the Ice Lake Xeon Scalable. About four months before Intel’s expected release was when the I/O Fund covered AMD during the height of the pandemic sell-off. The company was “the one that got away” in 2019 and March of 2020 allowed us to revisit this.
If technical jargon around chips isn’t your thing, then this is probably the most important line from our original analysis in terms of AMD’s competitive prowess: “It’s estimated that for every $1.00 in Rome chip sales, Intel loses $2.25 on average in Intel Xeon SP sales. The savings are then deployed to buy more Rome chips, which can further depress Intel’s revenue.”$1.00 in Rome chip sales, Intel loses $2.25 on average in Intel Xeon SP sales. The savings are then deployed to buy more Rome chips, which can further depress Intel’s revenue.”
Going into July of 2020 earnings, we reminded our Members to watch for a AMD beat and an Intel miss, asthat would be the lynchpin that sets into motion the lead for AMD on product. This was not an earnings call, rather a “be alert” message as we were putting into place leading semiconductor allocations at that time. This was based off a few clues in Nokia’s earnings report that stated they were delayed due to a chip supplier. We knew they were referring to Intel. Two weeks later, we got exactly that –Intel stumbled by pushing out delivery on Ice Lake and AMD seized the moment by releasing Milan.
Image Source: 2021 Webinar, Guess Which Car is Intel? 🙂
The gains we have seen over the next past few years were put into motion with Intel and AMD’s Q2 2020 reports as it’s proven quite difficult for Intel to catchup from this delay as AMD answered by speeding up its product release cycle. The projections provided by management at AMD at the time we first covered the stock was for $14 billion in annual revenue by 2023. Instead, we will see $21.5 billion in revenue if we factor in the 31% revenue guide that was provided in the most recent earnings call – plus margins are expanding.
This is from product strength and that’s important to remember, there is nothing in the financials that could have predicted the nearly $6 billion beat on the top line over a four-year period. Arguably, Covid was a tailwind but we are likely to see the data center double on a quarterly basis this year for a much larger revenue contribution (in terms of dollar amount) than years prior.
Reviewing AMD in 2021
The Milan EPYC Series announced in August of 2020 was officially launched in March of 2021. The Milan is built on 7nm technology and has up to 64 cores and 128 threads with increased clocks compared to the Rome series. At the time of launch, Milan had a 100% advantage over Intel’s Sky Lake on server processor scores, according to Geekbench. The launch that was in question during the Milan release from Intel is called Ice Lake, and as the car race picture shows above, Intel drove into a brick wall as this release was two years delayed. Ice Lake would eventually launch with 40 cores up from 28 cores while AMD had 68 cores.
“We won’t rehash the delay, denial, and begrudging admittance cycle that is Ice-SP’s gestation, just be aware that it was a 2019 CPU and is now a mid-2021 CPU. We know it launches today and Intel is officially claiming, “We have shipped over 200,000 Ice Lake CPUs for revenue” and the shipping parts are the D-2 stepping. Since volume production started in mid-January and the throughput is 4-5 months, these parts are likely wafers pulled mid-production and restarted, real production volume is set for May delivery. Don’t take our word for it though, the largest OEM out there thinks so too.the largest OEM out there thinks so too.
As an aside lets do the math and assume those 200K Ice-SPs shipped in three months or about 66K CPUs/month. If the server market is about 30M CPUs/year, lets call it 32M for the sake of round numbers, that would be 8M/quarter for normal production. 200,000/32,000,000 = .025 or about 2.28 days worth of production. This is not a figure I would be mentioning in public if I was aiming to boost confidence.”
In my world, that throwdown by SemiAccurate is like a good Hollywood roast session.
In terms of how third-generation EPYC performed against third-generation Xeon processors, there are debates there. I mentioned in the webinar that Intel likely has a very large marketing department that can help make sure headlines are in its favor — although it is true that Ice Lake does boost performance for AI, high-performance computing and cloud workloads with eight channels of DDR-3200 memory per socket and 64 lanes of PCIe, up from six channels of DDR4-2933 and 48 lanes PCI Gen3.
The core count of the Milan Series is higher despite any transistor density improvements that Intel may have had from Ice Lake. Hyperscale cloud customers likely prefer AMD for adding more capacity and also due to virtual CPUs from AMD helping to drive down costs for the hyperscalers. Technically, there is a performance imbalance between AMD and Intel skewed in AMD’s favor.
In this case we don’t have to wade through a public relations campaign to figure out where Ice Lake stands like we did in 2020 as we now have the benefit of hindsight. We can clearly see AMD taking market share in server CPUs although losing ground in desktops and laptops (our thesis is server market share so that’s less important to us). Notably, overall CPU market share for AMD is up.
Most importantly, look at where AMD was when it launched the second generation of Zen (roughly 2%) to today (roughly 11%) market share – or nearly 6X from this major design win. Moving forward, Intel will need to deliver a 7nm chip – but by then Lisa Su will already be releasing a 5nm design. As the analysis points out, Intel needs to make up for lost time, meanwhile, Lisa Su is unlikely to allow that now that AMD has clawed its way back from a near-zero.
“Turning to our overall data center business. We made outstanding progress in the last year. We exited 2021 with data center revenue contributing a mid-20 percentage of overall revenue, and we expect 2022 to be another year of significant growth based on the strong customer demand signals for our current and next-generation products.”we expect 2022 to be another year of significant growth based on the strong customer demand signals for our current and next-generation products.”
The management also said this which pretty much sums up AMD’s level of confidence:
“Yes, Vivek. So look, we always expect the competitive environment to be very strong and very aggressive. And that's the way we plan our business. That being the case, I think we're very happy with the growth that we've seen in the business sort of last year.
And as we look forward, we see opportunities in both cloud and enterprise. On the cloud side, we're in 10 of the largest hyperscalers in the world are using AMD. As they get familiar with us over multiple generations, they're expanding the workloads that they're using AMD on. So we see that across internal and external workloads.On the cloud side, we're in 10 of the largest hyperscalers in the world are using AMD. As they get familiar with us over multiple generations, they're expanding the workloads that they're using AMD on. So we see that across internal and external workloads.
In the Enterprise segment, we doubled year-over-year here in 2021. We continue to add more field support to have more people get familiar with our architecture. We have very strong OEM relationships. So I feel very good about our server trajectory. And yes, it's very competitive out there. But we think the data center business is a secular growth business. And within that, we can grow significantly faster than the market.”And yes, it's very competitive out there. But we think the data center business is a secular growth business. And within that, we can grow significantly faster than the market.”
Notably, these comments are likely priced in at the moment as the 31% guide outpaces overall data center revenue growth in 2022, expected to be 11%. However, it helps to have context in terms of AMD’s confidence level right now.
Q4 Update and What to Look for in 2022
As discussed, AMD had a breakout year in terms of its position in the data center which helped drive top line growth in 2021 of 68% revenue growth for a record $16.4 billion. This was the company’s sixth consecutive quarter of 45%+ year-over-year growth. Mind you, this is not a $2 billion company putting up these growth numbers.
Q4 revenue grew 49% year-over-year and was up 12% sequentially to $4.8 billion.
With all the profitability concerns in the market, Lisa Su came out swinging on the bottom line, partially driven by higher average sales price (ASP) given the supply shortage. The company doubled operating income, net income and EPS over the past year and also recently announced $8 billion in buybacks. Operating cash flow was up 239% year-over-year. Cash flow was up 314% year-over-year for $3.2 billion. There is $3.6 billion on the balance sheet. One important caveat – as stated some of this is driven by higher average sales price (ASP) due to supply constraints which Bradley dissects below.
Q4 gross margin expanded 5% to a 50% GM and operating income doubled year-over-year. Operating income doubled to $1.3 billion, operating cash flow was up 48% year-over-year, and free cash flow was up 53% year-over-year. EPS was up 105% YoY on a GAAP basis and 77% on an adjusted basis to $0.92. (Again, big bottom line growth but will be tempered when supply shortages ease).
First quarter 2022 revenue is expected to be $5 billion, for an increase of 45% year-over-year. The sequential growth is expected to be driven by higher server and client revenue. Adjusted gross margin is expected to be 50.5%.
For FY2022, revenue is expected to be $21.5 billion, for an increase of 31% YoY with an adjusted gross margin of 51%. The company provided a statement at the Analyst Day that server will grow to contribute 30% of total revenue by 2023, implying increased market share.
We are primarily interested in the Data Center and this is bridged between two revenue segments – Computing and Graphics for GPUs, and Enterprise, Embedded and Semi-Custom for EPYC processors. Data center EPYC CPUs help drive AMD’s leading growth category, Enterprise, Embedded and Semi-Custom which was up 75% in revenue and up 17% sequentially. This segment’s operating income was up 213% YoY and up 40.5% sequentially.
GPUs helped drive the Computing and Graphics segment, which was up 32% YoY to $2.6 billion in revenue in Q4. This
includes more consumer facing products such as desktop and laptop processors. The company is focusing on more consumer-facing GPU products, such as the Radeon 6000 which grew double-digits sequentially, and will have the RDNA 2 to architecture powering gaming consoles and PCs. The company is also releasing a new mobile GPU and GPUs for lightweight gaming notebooks.
Regarding GPUs at the data center level, The Instinct MI200 accelerators power high-performance computing (HPC) and this helped drive data center graphics. The new Instinct accelerator outperforms the M100 with 383 TFLOPs compared to 185 TFLOPs. This is accomplished by coupling two CDNA2 dies with Infinity Fabric interconnects. AMD is starting to go more head-to-head with Nvidia on the A100 in terms of artificial intelligence applications and benchmarks.
Combined, data center EPYC CPUs and Instinct GPUs helped AMD cross the $1 billion revenue mark per quarter last year for data center revenue, and this could reach $1.5 billion per quarter by Q1 and may reach $2.3 billion per quarter by the end of this year. There are Trento EPYC CPUs and Aldebaran Instinct GPUs that are used in Frontier supercomputers but not as meaningful as the more commercial lines. Estimated revenue for HPC accelerators are in the $250 million range, per semiconductor analysts.
According to management, “revenue doubled year-over-year in this category and increased double-digits sequentially.” This is where AMD is gaining ground against Intel and is the primary growth we watch. The company was able to grow more than 100% year-over-year driven by cloud capex from companies such as AWS, Alibaba, Google, IBM and Microsoft Azure. Looking forward, Facebook is now a major customer for EPYC processors for their Metaverse workloads and new data center buildouts with Facebook also building data centers with Nvidia’s A100 GPUs.
Diversified Computing:
Milan-X EPYC with 3D V-Cache = Technical Computing
EPYC processors will get the boost that Ryzen gaming chips got last year from 3D stacked memory. The product is called “AMD 3D V-Cache” and will add cache capabilities in a vertical stack increasing the memory capacity from 256 MB to 768 MB by adding an additional 512 MB vertically. According to one report, the L3 cache (which boosts performance) can have up to four cache stacks per chip. According to AMD, this offers a "50 percent average uplift” across targeted workloads by offering 15X density increase, 200X interconnect density increase over 2D chiplets, and 3X energy efficiency. Ultimately, this means lower latency and improved performance.
Note: we covered 3D Stacking for Memory here in our Lam Reportcovered 3D Stacking for Memory here in our Lam Report
Microsoft published a report that showed 50% to 80% higher performance on complex simulations and workloads, such as electronic design automation (chip design), computational fluid dynamics and finite element analysis (FEA). The study also showed 42% to 51% lower memory latency compared to the previous generation of Milan with an amplification effect of up to 1.8X for effective memory bandwidth due to the workload performing as if it were being fed a higher bandwidth from DRAM.
Here's what AMD said on the call:
“Microsoft Azure previewed a new HPC instance, powered by our third-gen EPYC processors with 3D stack memory that delivers up to 80% more performance than currently available instances. Our differentiated 3D stacking technology further extends the leadership performance of EPYC processors and technical computing workloads like EDA, fluid dynamics, and complex simulations. We started volume production of EPYC processors with 3D stacked memory earlier this quarter in advance of OEM platform launches with all our major server partners.”
Genoa Series Shipping in 2022, Cloud-Native Specific Bergamo Close Behind in H1 2023
The fourth-generation of Zen architecture is the Genoa EPYC processors with 96 cores which will deliver the highest performing general-purpose compute. The Zen 4c core is made for cloud-native workloads due to its thread density and will be featured in the Bergamo server roadmap for the first half of 2023. This powerful combination of Zen 4 cores and power-efficient CPUs are tailored for cloud workloads. The Bergamo release will have up to 128 cores, an increase from the 96 cores in the 2022 Genoa series.
As we stated during Intel’s stumble, it’s likely we see a 5nm chip come from AMD in this time frame, and that’s exactly what the company plans to do with the Zen 4 and Zen 4c platform. The “c” stands for cloud computing. The Ryzen desktop processors will also leverage a 5nm Zen 4 core with the new AM5 socket; this was discussed at the CES 2022 presentation with expected launch in H2 2022.
Here is what the company said in the call about the upcoming lineup of Milan-X and Bergamo:
Chris CasoChris Caso
Yes. Thank you. Good evening. First question is, if you could give some indication of the strategy behind some of the processor variants that have come out, most recently Milan-X and Bergamo coming up. Do those variants represent incremental revenue to AMD? What's the strategy behind it? How does that help you, help the product line?First question is, if you could give some indication of the strategy behind some of the processor variants that have come out, most recently Milan-X and Bergamo coming up. Do those variants represent incremental revenue to AMD? What's the strategy behind it? How does that help you, help the product line?
Lisa SuLisa Su
Sure, Chris. Well, I think the strategy is, as we have gotten more scale in the business, we can invest more and we see ways to further differentiate our product portfolio. So I mean, I think Milan-X is really sort of the highest of the highest end. And we see that for technical computing and some of these EDA workloads that, that does give us a very differentiated product. And then we have the regular Milan product line. We'll have Genoa. And Bergamo is really optimized for cloud.
So I do believe it gives us more opportunity to expand from a market share and a footprint standpoint. And I think the broader statement, Chris, is that, the data center is so large. There are so many different workloads that you can optimize. Like, by doing these variants, we will actually get a better solution for the customer, give them better total cost of ownership and, hopefully, give us a larger footprint in that workload as well.So I do believe it gives us more opportunity to expand from a market share and a footprint standpoint. And I think the broader statement, Chris, is that, the data center is so large. There are so many different workloads that you can optimize. Like, by doing these variants, we will actually get a better solution for the customer, give them better total cost of ownership and, hopefully, give us a larger footprint in that workload as well.
Discussion on Average Selling Prices, Supply and Dilution Impact of Xilinx Acquisiton
By Bradley Cipriano
As mentioned above, Q4 sales increased 49% YoY to $4.8 billion while FY2021 sales increased 68% YoY to $16.4 billion. During the year, AMD’s largest segment, Compute and Graphics (57% of 2021 sales) increased 45% YoY to $9.4 billion. The rise in Compute sales was driven by a 57% YoY surge in average selling prices (ASP), offset with an 8% decline in volumes. AMD explained in its 10K that the rise in ASP was driven by the company’s focus on higher-end products, and a greater mix of its Ryzen, Radeon and AMD Instinct products. AMD further explained that the lower volumes were driven by its focus on premium products and a tight supply environment. The tight supply market also likely contributed to the ramp in selling prices.
It is noteworthy that 2021 was a unique year, and investors should expect that ASPs will likely normalize in the future. While we do not believe that ASPs will rise by 50%+ again in 2022, we do believe that volumes will rebound and supply chain constraints will ease. As shown in the chart below, AMD’s raw materials inventory balance is at multi-year lows, highlighting the limited supply of input materials in the current environment. On an absolute basis, Q4 raw materials declined 12% YoY to $82 million, a three-year low and inventory balances relative to sales were also at a five-year low in the most recent quarter. The low inventory levels highlight the strong demand for AMD's products, but may be a near term headwind to growth if raw materials inventories do not rebound.
Fab capacity is also constrained, and multiple companies such as TSMC, Texas Instruments and Intel have announced new fab constructions recently. To address this capacity issue, AMD has prepaid nearly $1 billion for long-term supply agreements (shown below). Lisu Su explained on the Q4 call that the firm’s focus is on securing long-term supply, rather than raising prices. However, she explained that prices will likely continue to rise given the strong demand in the current environment. This trend could lead to strong earnings and cash flows in the near future.
Picture 1. AMD Secures Long-Term Supply Capacity
While Computing volumes declined, Enterprise, Embedded and Semi-Custom sales surged 113% YoY to $7.1 billion, primarily due to higher volumes of EPYC server processors. As mentioned above, the growth in EPYC server processors is a secular tailwind that we expect will continue as AMD captures more data center market share.
AMD’s focus on premium products and its ramp in ASP and volumes led to strong growth in margins. Gross margins increased 300 bps YoY in 2021 to 48%, driven by a richer mix of EPYC, Radeon and Ryzen processor sales. However, margins may come under pressure going forward as ASPs ease following a normalization in supply chain constraints. Fortunately, this will likely be offset with a rise in volumes as inventories increase. AMD has secured long-term supply capacity which will help it meet the robust demand for its products, allowing it to continue to capture market share and grow earnings going forward. However, we will be closely monitoring the supply situation as semiconductors have historically been a cyclical industry (but that trend may be changing as well).
Near-term Impact of Xilinx Acquisition
In October 2020, AMD announced its intention to purchase Xilinx for $35 billion in an all-stock deal. Xilinx had 252 million in diluted shares before the acquisition closed on 2/14/22 and AMD issued 1.7234 shares per Xilinx shares, which resulted in ~430 million shares of AMD being issued. On the closing date of the transaction, AMD shares traded at $114.27, giving the transaction a value of about $50 billion. There may be some near term headwinds to AMD's stock price as holders of Xilinx sell AMD shares, however we do not expect a material impact to dilution for a couple of reasons. For one, outside of a few independent instances for non-employee Board of Director members, there was not an acceleration in the vesting of shares, so insider selling should not be expected to accelerate following the close of the transaction. AMD disclosed that “the [Xilinx share] awards generally remain subject to the same vesting and other terms and conditions that applied to the awards immediately prior to the [acquisition date]”. Importantly, insiders will not be selling 100% of their newly acquired AMD shares either. For example, Former Xilinx CEO Victor Peng will join AMD as president of the newly formed Adaptive and Embedded Computing Group. Mr. Peng owned 192,000 Xilinx shares, or about 1% of XLNX's float.
Further helping to offset the dilution is the announcement of an $8 billion share buyback on 2/24/22. This is on top of $1.2 billion of remaining buyback capacity after the close of the year. AMD will be able to purchase around 88 million shares at current prices, which reduces the potential dilution by ~20%. So long as insider selling is not above 20% over the next year, then we shouldn’t expect a material impact on the shares from the stock deal. Finally, AMD has capacity to ramp share repurchase going forward. The combined company should be producing north of $4 billion in annual FCF, with close to $6B in cash on the balance sheet, net of debt. AMD expects about $300 million in cost synergies as well following the deal, which should improve the capacity for share repurchases even further.
Conclusion:
By Beth Kindig
During the height of the high beta bubble (SPACs) and also when hypergrowth was in favor, it was counterintuitive to build a position in a semiconductor company. It can also be tough to rely on product over financials with semis, which is why we see fund managers still pushing Intel and analysts here.
However, from this stance on product, it will be easier for AMD to take market share from its current position than when it was at a 2% penetration. This is partly because Intel’s stumble came during *maybe* the most critical time for hyperscalers to expand due to “digital transformation.”
For the conclusion, I’m going to take a direct quote from CEO Lisa Su on the earnings call as to why I believe AMD can continue its lead in the data center and across other segments as she says it better than I can:
“And I think what you're seeing is growth in the model from the standpoint that we've always kind of said we're underrepresented in the business. When you look even today with all of our growth, we're still underrepresented in the business, whether you're talking about the server business or the PC business. And so we believe that our product strength and our customer engagements are such that we can grow significantly in this environment.”
The “underrepresented” she refers to in her comment is why I’ve called AMD “The Dark Horse” – which means an unexpected contender. We will need to rename AMD someday after the market is fully onto the product story as it was more fitting at 2% than at 11% market share and it certainly won’t apply if AMD continues on this trajectory.
Disclosure: The I/O Fund owns shares in AMD and has no plans to change their respective position within the next 72 hours. You can access the I/O Fund’s positions here. The above article expresses the opinions of the author, and the author did not receive compensation from any of the discussed companies.Disclosure: The I/O Fund owns shares in AMD and has no plans to change their respective position within the next 72 hours. You can access the I/O Fund’s positions herehere. The above article expresses the opinions of the author, and the author did not receive compensation from any of the discussed companies.
It would be impossible to look at the AMD-Xilinx acquisition without doing an in-depth breakdown of FPGAs. The chips are powerful yet are challenging to program, and therefore, adoption has been slower than originally forecasted around 2016-2018.
Around the time that I began writing on Nvidia, I was also covering Xilinx. At the time, the company was very promising as Microsoft was adopting FPGAs and the chips were slated to be a front-runner for 5G networks. This began to change, however, when Nokia announced they were moving away from Intel-Altera’s FPGAs after the poor results discussed in Q3 2019.
Below, I compare GPUs with FPGAs and ASICs to help break down the potential strengths for the AMD-Xilinx acquisition, especially why a predominantly CPU-company would acquire a FPGA-company. We discuss the headwinds for FPGAs and Xilinx and how AMD could potentially alleviate these.
For our purposes, we will call these chips data accelerator chips or AI chips. As you know, GPUs, ASICs/SoCs and FPGAs have many other uses (gaming, PCs, smartphones, electronics), but as tech growth investors, we are mainly interested in modern data centers and cloud IaaS infrastructure that can handle the increase of networking bandwidth and optimization of AI workloads. By accelerating the computing platform, these chips can power machine learning, deep learning and high-performance computing workloads. The leading chips will have quite the addressable market to capture and we want to be there when this happens.
The data center accelerator market was forecast to grow 49.47% CAGR between 2018-2023 from $1.4 billion in 2017 to $21.2 billion in 2023. At the time of the forecast, FPGAs were forecast to be the leading chip in terms of growth but this has not materialized (we cover this below). According to a more recent report in May of 2020, the global accelerator market will reach $38.9 billion with GPUs growing at a compounded rate of 47.1%
The growing need for cloud resources is driving a healthy market for hyperscale data centers with an expected increase of 60% between 2016 and 2021. We are also seeing substantial investments in next-generation data centers due to the pandemic, such as Alibaba’s announcement to invest RMB 200 billion in core technologies and future-oriented data centers.
When it comes to processors, the question is which chip will answer the demand. This is a question that has not been fully answered yet although GPUs are almost universal for general AI due to the ease of development for software engineers, ASICs are gaining popularity with Google and others who seek application-specific advantages, and FPGAs are continually forecast to lead the growth but sees roadblocks in the learning curve for hardware configuration.
The competition between FPGAs and ASICs could also be alleviated by combining an Arm-based processor with an FPGA. The Zynq-7000 SoC allows for dedicated hardware blocs to split-up non-critical tasks from critical tasks. I imagine AMD fully comprehends the strengths and weaknesses of FPGAs and is set to solve the developer adoption uptake should the acquisition go through.
Overview of GPUs, FPGAs and ASICs:
General Definitions:
GPU: graphics processing unit with a highly parallel structure compared to CPUs. When training deep learning neural networks, GPUs are up to 250 times faster than CPUs. When compared to FPGAs and ASICs, GPUs continue to lead due to the learning curve for software developers and not requiring changes to existing code. Notably, GPUs originated as graphics cards used in gaming but now lead in general AI processors. Nvidia is the major GPU player.
FPGA: field-programmable gate arrays that can be programmed electronically “in the field” post-manufacturing after the chip leaves the foundry. The chip is made up of configurable logic blocks and programmable interconnects that allow for the chips to be reprogrammed. FPGAs are preferred for prototyping or for instances where the design may evolve. Designing with FPGAs are low cost but can become expensive over time. Intel and Xilinx are the major FPGA players.
A few points to note:
• FPGAs increase real-time inference compared to CPUs and reduce latency compared to GPUs.
• FPGAs chips are also cheaper and also faster to bring to market than ASICs (although this is not an advantage for high production volumes — more below including a visualization of this).
• FPGAs are unique from ASICs and GPUs due to being customizable post-manufacturing. The “fieldprogrammable” piece is unique to FPGAs.
• There are new products being released all of the time that aim to get an advantage between ASICs and FPGAs, but for the most part, these two are similar in latency and somewhat similar in power efficiency except ASICs technically lead here due to being application-specific. The design needs will often determine the decision between ASICs and FPGAs and the production volume. Notably, FPGAs will often be used for prototyping before switching to ASICs.
• The drawback to FPGAs is the complexity in programming as software engineers are not as familiar with hardware-specific languages.
ASIC: application-specific integrated circuit customized for a specific application. If an ASIC has more than one processor core and/or combines various computer components, then it’s considered an SoC. You’ll hear these words used interchangeably (ASIC/SoC) on earnings calls.
ASICs are preferred for large production volumes as the cost for design can be in the millions of dollars but then averages out over time.
• Google is the perfect example of a company that uses ASICs as the company has many servers dedicated to solving specific problems.
• These chips, including Google’s TPU, can be designed for maximum efficiency by shifting the optimization and resource assignments to the CPU with the TPU/ASIC acting as a coprocessor for vertical instructions.
• Some companies may find ASICs to be too rigid and fixed. As of recently, Microsoft for example has preferred FPGAs as there is more flexibility in the design.
Expanding on these Definitions:
For most design purposes, FPGAs (Xilinx) are considered superior to GPUs (Nvidia) when it comes to power efficiency. They offer a higher amount of on-chip cache memory to help reduce the bottlenecks from external memory, and are flexible enough to be reconfigured for various data types, such as binary, ternary, and custom data types, whereas GPUs must be modified at the vendor level. With that said, Nvidia will likely leverage Mellanox to speed up GPUs and close the gap on latency performance with FPGAs and ASICs.
GPUs are programmed at the foundry and are restricted by Single Instruction Multiple Thread (SIMT), which provides an advantage over CPUs, but can also result in lower performance efficiency when enough parallels are not found for the workload.
Despite FPGAs resulting in faster high-performance computing, they are harder to program due to hardware circuit configurations compared to GPUs for machine learning, which require less engineering resources due to being programmed through software. FPGAs are generally run with high-level languages such as VHDL or Verilog. GPUs are also more cost efficient.
ASICs rival FPGAs on efficiency and power (and often beats FPGAs in these areas for specific workloads) and this is one reason why we’ve seen ASICs become more popular in recent years. The difference between these two is reconfigurability. This is a major advantage for end applications and workloads as the chip can be programmed “in the field” after it’s left the foundry. As discussed, the reconfigurability is what the acronym FPGA stands for – Field-programmable gate array. You can program the chip to be a microprocessor, graphics card or encryption unit.
ASICS are Application-specific Integrated Circuits and are designed to be application-specific for one purpose only. The circuitry cannot be changed because it is made up of permanent circuitry. You use ASICs every day in your smartphone, laptop and television.
ASICs have high “non-recurring engineering” costs (NRE) and are more expensive at the onset.
However, FPGAs come at an increased cost after a certain time period and have limited analog functionality. Therefore, FPGAs actually cost more overall because the cost of ASICs becomes lower with higher production.
FPGAs have limited analog functionality, such as Bluetooth and WiFi. This is why ASICs are the chosen chip for electronic devices. “Low power” is also a major advantage to ASICs which makes the chips ideal for specific battery-operated devices.
Here is a picture I provided in Marvell’s PDF. What this picture is showing is that it costs millions to begin with ASICs and less than $5000 to begin with FPGAs. However, over time, the cost of FPGAs exceeds that of ASICs.
FPGAs are used more for prototyping due to the reconfigurability and due to ASICs requiring more during the design process. ASICs take months longer to implement due to the manufacturing cycle, and as mentioned above, cost quite a bit at the onset. The R&D cycle for ASICs can become problematic when companies are competing neck-to-neck for market share.
FPGAs in Real Use Cases
We want to be clear that we are ultimately bullish on the AMD-Xilinx acquisition as we believe AMD has what it takes to bridge FPGAs. In the use cases below, you will see there are some mixed results with FPGAs competitively which is likely leading to Xilinx looking at an offer.
The Nvidia-Mellanox-Arm combination is a looming threat, as well, and if AMD can make FPGAs more accessible, then this will provide AMD with critical market share in data center acceleration/AI chips without having to compete with Nvidia head-on with GPUs.
Xilinx’s Segments
ABC: Automotive, Broadcast and Consumer
AIT: Aerospace and Defense, Industrial, Test and Measurement
DCG: Data Center Group
ISM: Industrial, Scientific & Medical
TME: Test, Measurement & Emulation
WWG: Wired and Wireless Group
Aerospace/Defense and Automotive:
Xilinx leads in aerospace/defense and automotive. These are industries where FPGAs have a clear advantage.
In May, Xilinx announced the first 20-nanometer (nm) space-grade FPGA to deliver machine-learning for space applications. This allows satellites to update in real-time, deliver video-on-demand, and perform compute “onthe-fly” to process complex algorithms.
Although autonomous driving is very new and the market is wide open, FPGAs beat GPUs in many regards for this application. This is likely part of Intel’s motivation in buying Altera. This space is constantly evolving but here is a recent quote from a product manager in the field, “Autonomous vehicles rely a great deal on machine learning, and every new vehicle in every new situation may contribute to the shared knowledge base,” said Tobias Welp, product manager at OneSpin Solutions.
FPGAs offer flexibility for many applications because both the hardware and the software can be reprogrammed. Reprogramming FPGAs when knowledge or algorithms are enhanced has the potential to keep autonomous driving in a state of continuous improvement.”
But there are tradeoffs. Verification in this case becomes a continuous process.“Every time the design changes, the full verification suite (static, formal, and simulation) must be run,” Welp said. “Formal equivalence checking also must be run to ensure that the FPGAs have no implementation errors, security vulnerabilities, or lurking hardware Trojans. Finally, the reprogrammed FPGAs must be extensively validated on test vehicles before updates are sent to the field.”
In regards to ASICs and how AVs are in constant flux, here is what Welp stated:
“When we started in this space and we were talking to automotive customers a year ago, everybody was going straight to Level 4 and Level 5 autonomous,” said Geoff Tate, CEO ofFlex Logix. “They were all going to do their own custom chips. They were all looking to license IP for inference acceleration. That’s changed dramatically. I don’t know of anybody who’s looking to do an ASIC in the automotive space right now. Everybody that was telling us they’re going do their own chips has changed to buying off-the-shelf chips, and almost all the major car companies are focused more on driver assist.”
Therefore, we see that FPGAs have a serious shot here at being the chosen chip for AV development.
Wired and Wireless Group:
Xilinx saw a significant slowdown in the wired and wireless group in the previous quarter due to supplying Huawei but some of this growth has returned. The Nokia-Intel FPGA flop in November 2019 has hurt the prospects for using FPGAs in 5G with Nokia turning more towards ASICs/Marvell.
Originally, Nokia stated that FPGAs seemed like the best choice because 5G standards were not developed yet and the flexibility was key. However, as we’ve illustrated in this analysis and covered in the Marvell PDF, ASICs cost less over time and this is becoming a priority for Nokia to protect their bottom line on the already-expensive 5G infrastructure overhaul. In an earnings call, Nokia’s CEO lamented that FPGAs were more costly than anticipated.
So, what’s happening right now is when he moved to 5G, we chose FPGA-based products. They give you flexibility, they give you time to market advantage, but then they’re expensive. And so, what we’re doing is, we’re moving to equity SoC-based products, which we’ll progressively start shipping during 2020. -Q3 2019 earnings call
Like I said earlier, I mean, the System on Chip strategy has been put into motion already a while ago, diversified our supplier base. We are increasing the investments purely because we want to increase even more the SoC penetration in our products and continue that. And of course, we know how to do System on Chip. Yes, we started with FPGA with 5G because it’s gave us that time to market catch-up advantage, but we do that in much of our portfolio with FP4 and PSE-3. So, we’re just replicating that in mobile. –Q3 2019 earnings call
And Tal, on the question regarding to System on Chip, so we are transitioning from FPGA to System on Chip and this is the metric that will give you an update and this is that we got to 10% of the 5G product by ReefShark System on Chip portfolio. We started ramping up volumes and that will get to 35% by the end of this year or greater than 35% and then 70% by the end of ‘21 and then this whole thing will be complete about 100% in 2022. -Q4 2019 earnings call, when an analyst asked for an update in moving from FPGA to SoC.
Despite Nokia’s decision, FPGA-proponents for 5G will argue that these chips are ideal for network infrastructure to prevent vendor lock-in and for futureproofing the network due to the ability to reprogram. In this way, FPGAs could reduce long-term operating expenses and reduce total cost of ownership.
You can access a full list of Xilinx’s segments here and how the company serves each of these markets, including Industrial, Medical and Video Processing.
High-Performance Supercomputing
FPGAs and high-performance computing (HPC) is an important part of data center acceleration that can benefit from easier programming options. This is because FPGAs are known to be pliable when involving interconnects and this is valuable for supercomputers.
This will not replace GPUs rather it will serve applications with heavy computations. FPGAs can optimize purposebuilt architecture and there is a high probability we will see the supercomputers of the future powered by a combination of CPUs/GPUs and FPGAs. As of now, Nvidia is the undisputed leader in the data center. In fact, as of May 2019, Nvidia was employed in 97.4% of cloud IaaS compute instance types with dedicated accelerators with combined Xilinx and Intel at 1.6%.
Liftr Insights shows a slightly better picture for FPGAs at about 5% for Xilinx and a little under 5% for Intel across Alibaba, AWS and Azure in March of 2020. The analysis firm puts Nvidia at 86% in this study.
In 2015, Intel acquired Altera in an all-cash transaction worth $16.7 billion. Altera was second to Xilinx as a leading provider of FPGAs. This acquisition occurred during the years that FPGAs were favored for data center growth over its counterparts yet Intel has not been able to penetrate data centers with this acquisition as originally estimated. This could be for two reasons: (1) FPGAs are more advanced and will rise in popularity after general AI is exhausted and more complexity is required by the market (2) ASICs are superior to FPGAs and are meeting the market demand with customizable that was once assumed would be met by field-programmable.
In 2018, Microsoft announced it would be phasing out Intel-Altera FPGAs for over half its servers in favor of Xilinx’s processors. This was confirmed again in October of 2019 by Microsoft at a Xilinx conference although no official update for two years now. At the time, I guessed this move by Microsoft was due to AI engineers preferring Xilinx over Intel, which I still believe to be the case when FPGAs are being considered.
It should be noted that AMD is already solid in the supercomputer category with Epyc CPUs powering many of the supercomputers in the top 500 list. Here’s a great write-up from Moor Insights on AMD’s partnership with HP’s supercomputer manufacturer Cray. The partnership is expected to launch the second Exascale system in the United States costing over $600 million. The Frontier Supercomputer is expected to put AMD on the map for AI accelerators and as a competitor to Nvidia.
Constantly Evolving:
Xilinx’s SDAccel IDE has attempted to provide software developers the same experience no matter the cloud provider (AWS, Alibaba, etc). The goal was to copy Nvidia’s CUDA platform to enable a larger ecosystem. The tool platform is called “Vitis” and is designed to provide accessibility for hardware developers and software developers. The first release of SDAccel supported deep learning frameworks, such as Caffe, MXNet and TensorFlow through Python APIs.
AMD backed Xilinx around this time for the Alveo model launch for machine learning, which was the first supported environment on Xilinx’s SDAccel IDE. Alveo was dubbed “the world’s fastest data center and accelerator cards” to increase real-time inference throughput by 20X compared to CPUs and 3X compared to GPUs. AMD offers Radeon Instinct accelerator cards built on Vega 7nm GPUs.
Xilinx also launched the “Versal” advanced computing acceleration platform. This is a fully softwareprogrammable heterogeneous compute platform that improves performance 20X over current FPGAs and 100X over CPUs (per Xilinx’s white paper). The SoC-like chips combine CPU cores, programmable logic and ASIC elements. This was around the time that Xilinx stopped referring to itself as a FPGA company and instead as a platform company with a focus on “whole application acceleration.”
The Versal series includes AI engines in the device series, such as Versal AI Edge, Versal AI Core and Versal AI RF. Xilinx aims to not only accelerate the AI portion of the task but to combine AI engines with DSP engines and also adaptive engines to accelerate the entire task, such as beamforming for 5G radar wireless communications or for smart controllers for storage systems in data centers.
Financials:
Advanced Micro Devices reported Q2 results on July 28th, beating comfortably on both the top and bottom lines. Revenue came in at $1.93B (+26% YoY), representing a beat of $70M above consensus estimates. Management attributed the revenue growth primarily to higher Computing and Graphics segment revenue.
Non-GAAP EPS grew 333% YoY to $0.18 per share in the quarter. Gross margin increased 3 percentage points YoY to 44%, primarily driven by Ryzen™ and EPYC™ processor sales. For the 3rd quarter and FY, management is calling for an acceleration of revenue growth to 42% YoY and 32% for FY 2020.
Xilinx reported Fiscal Q1 results on July 30th, reporting a slight miss on the top line and a slight beat on the bottom line. Revenue decelerated 14% YoY to just under $727M, missing consensus estimates by about $1M. Non-GAAP EPS decelerated 33% YoY to $0.65 per share, beating the consensus by half a cent. The company made no adjustments to its outlook and expects to record $755M in revenue in its next quarter.
At a listed acquisition price of $30B, Xilinx would be valued at 10x sales. Xilinx has 244.3M shares outstanding and the company is projected to deliver $3.54 in EPS in FY 2022, meaning they are on pace for $864M in net income in 2022.
At an acquisition price of $30B, AMD would need to issue 361M shares in an all-stock deal for Xilinx. In 2022, AMD is projected to deliver $2.20 in EPS. With $1.17B shares outstanding, the company is on pace for approximately $2.6B in net income for FY 2022.
In order for the deal to be accretive for AMD, the Xilinx business has to generate approximately $800M in 2022 annual net income. The Xilinx standalone business is projected to generate $864M in annual profits in 2022.
If the acquisition price exceeds $30B, an all-stock deal may become dilutive. At an acquisition price of $35B, AMD would need to issue 422M shares to acquire Xilinx. This would require the Xilinx business to generate $1B in 2022 net income for the deal to be accretive.
Conclusion:
On a technical level, workloads like machine learning, AI and 5G can benefit from a chip that is field-programmable and bridges the gap with customized chips that take too long to bring-to-market. Xilinx’s FPGAs allow algorithms to be adjusted for critical technologies and R&D processes.
This space is constantly evolving. FPGAs also have SOC-chips that Xilinx calls “all programmable SOCs.” Hard-silicon processor cores are being combined with FPGAs to compete with ASICs. In this case, an ARM Cortex A9 and Xilinx Zynq become dedicated hardware blocs to split up non-critical tasks from tasks requiring high-speed acceleration.
The question is can AMD do this for Xilinx versus Nvidia in key markets:
When it comes to efficiency, AMD is an unstoppable powerhouse. There are leaks that the 7nm Milan release will achieve a higher clock rate with performance increases of 10-20% between generations. This is virtually unheard of.
Lisa Su brought AMD from a $3 billion market cap to a $100 billion market cap in 5 years. As of now, we see Xilinx spread across too many segments and lacking focus. If AMD can popularize a platform for Xilinx/FPGAs that competes with CUDA and chooses the segments where FPGAs have the most promise, then we could see FPGAs finally live up to their true potential.
AMD under Lisa Su as CEO has risen 2000% over the past 5 years
I believe there will be quite a few negative opinions about AMD’s move with Xilinx. Analysts will say Nvidia has the undisputed throne, there is no overlap with AMD-Xilinx, that the acquisition is too expensive and dilutes shareholder value and that AMD does not have enough successful acquisitions under its belt to gamble on this combination.
Others may not see why AMD would acquire Xilinx, but I’ve been waiting for something to happen with FPGAs and this very well could be it. We had discussed AMD innovating past Intel prior to this happening in July. Our premium members were pretty happy about that call. On a similar note, I’ve been tracking Xilinx closely, waiting for a breakthrough of some sort.
Of course, I am a mega Nvidia bull and this will not change. This will not be a winner-takes-all market, rather a market that compounds quickly for the top handful of companies. There are a few CEOs I won’t be against and Lisa Su is one of them.
Grab some popcorn because it’s going to get pretty exciting between 2022-2025 as Su and Huang dual it out. My prediction is they will both take enough market share for my premium subs to remember these calls as some of my very best.
Our goal is to catch Micron for the 2021 rebound which is likely delayed a year from the anticipated 2020 rebound. This rebound should occur when high-end smartphones are released again and the automotive market comes back to help drive demand in embedded DRAM. Mobile and automotive are the hardest hit segments in 2020. Data center segment remains strong.
Due to the cyclical nature of memory and storage, Micron is likely to become a 1-2 year holding rather than a permanent buy-and-hold.
Product:
Micron is the only company in the world with a portfolio of DRAM, NAND, and 3D XPoint technologies. X100 is the fastest storage device in the world. The company has also entered into a new 3D XPoint wafer sale agreement with Intel that replaces the previous agreements.
In the most recent fiscal year, DRAM comprised two-thirds of Micron’s revenue and NAND one-third of revenue.
NAND memory saves data even when the power is removed, such as when a cell phone is turned off. DRAM only saves memory when a device has power but is much faster than NAND and lasts longer. Beyond mobile devices, NAND is found in traffic lights, digital advertising panels/displays, and anything with artificial intelligence that needs to store data.
As covered in the Lam Research report, NAND has been around since the 1980s but got a much-needed boost from 3D NAND, which stacks vertical chips. Historically, Micron focused on DRAM for PCs and servers an expanded into NAND over the past ten years.
One risk to Micron is the thin moat as competitors Samsung and SK Hynix outpace Micron in total memory/storage shipments. With little differentiation, these companies have pricing wars with Samsung generally considered the industry leader. Toshiba and Western Digital (SanDisk) are also competitors.
This is one reason Micron continues to invest in R&D in products such as 128-layer 3D NAND, 3D XPoint and also 1Z-nanometer DRAM.
“The Memory Guy” Jim Handy has a great write-up describing how Micron has improved its profitability in the DRAM market. His analysis points towards Micron holding a leadership position in 1Znm production over Samsung and Hynix. The new DRAM was introduced at CES and is geared towards the server and hyperscale markets.
One of the bull cases for Micron right now is DRAM and NAND pricing, which is high due to low inventory and previous capex cuts. There is low supply right now regardless of contracting demand. Prior to Covid-19, the market believed pricing had bottomed in 2019.
Micron is one of the most volatile semiconductor stocks with lows around $10 in 2016 and highs around $60 in 2018. Regarding valuation, the stock is trading at double its current PE ratio as 2019 and similar forward PE ratio as 2019. The issue here is any data center strength may not be able to offset the weakness in the mobile and automotive segment.
Historically, buying Micron at a price-to-book value of 1 has done well. The stock is currently trading at a price-tobook of 1.439.
Micron Financials:
In the most recent quarter ending in February, Micron’s revenue beat estimates yet fell 18% year-over-year to $4.80 billion. Revenue was down 7% from $5.14 billion quarter-over-quarter. TTM revenue was $19.6 billion with non-GAAP net income of $2.9 billion, or $2.54 EPS.
DRAM sales were down 11% sequentially and NAND sales were up 6% sequentially. DRAM was impacted by flat sales prices and lower bit shipments.
Earnings were also down YoY with Micron reporting GAAP net income of $405 million, or $0.36 EPS, compared to $1.62 billion, or $1.42 EPS in the year-ago quarter and $0.45 EPS last quarter. Non-GAAP income of $517 million or $0.45 per share beat estimates by $0.08 compared to $1.71 EPS.
Capital expenditures were $1.94 billion in Q2 2020. Management expects FY 2020 capex to be $7 to $8 billion. For fiscal Q2 ending in February, the company had cash and investments of $8.12 billion with a net cash position of $2.7 billion. The company has about $5 billion in long term debt. Recently, Micron drew on a $2.5 billion revolver to have cash on hand.
Margins are decreasing with gross margins of 28% in Q2 2020 compared to 49% in Q2 2019. Operating margins were at 9.2% in the most recent quarter compared to 33.5% in the year-ago quarter.
The median forecast for FY 2020 ending in August is $20.11 billion, down 14.7% year-over-year.
The median forecast for FY 2021 is $24.49 billion, up 21.74% year-over-year. Forward estimates for EPS of $4.90 for FY 2021 will represent an increase of 124% YoY.
QLC SSD bit shipments rose 60% sequentially in the 2Q FY2020. The company expects QLC SSD to grow in the 2H 2020.
The company began to deliver LP5 mobile DRAM products to customers including Xiaomi, which is using LP5 in its 5G-capable Mi smartphones in 8GB and 12GM configurations.
In the graphics market, GDDR6 bit shipments increased more than 40% q-o-q. In the new gaming consoles the company will deploy SSD’s in place of hard drives for the first time.
Effects of Covid-19:
Micron is more exposed than other semiconductors to consumer spending.
About 15% of Micron’s revenue comes from China, where there was weaker sell-through of consumer electronics and factory shutdowns in the fiscal second quarter ending in February. According to the most recent earnings call, some of this was offset by stronger data center demand due to increased gaming, e-commerce, and remote-work. Management expects this trend to continue globally.
Due to Covid-19, Micron expects to see lower demand for smartphones, consumer electronics, and automobiles than prior expectations. Anticipating changes to customer demand, Micron is moving supply from smartphones to service the strength in the data center markets for both DRAM and SSDs.
Some equipment companies have also indicated delays in equipment deliveries due to the impact of various government actions to combat COVID-19.
The Malaysian government issued lockdown orders on March 16 and Micron closed the manufacturing plants in Muar and Penang. Later, the Malaysian government declared semiconductor production as essential and after a few days the production resumed on a limited basis. In the earnings call, the company stated it’s using its global supply chain to mitigate production impact.
For the most part, analysts are cutting their forecasts for Micron, primarily due to Covid-19. Goldman Sachs, Piper Sandler, KeyBanc and Morgan Stanley have all lowered price targets.
Revenue Segments & Addressable Market:
Micron’s business composition is 64% DRAM, 32% NAND and 4% 3D XPoint memory.
Micron has four business units, which are reportable segments:
• Compute and Networking Business Unit (CNBU) — 41%
• Mobile Business Unit (MBU) — 26%,
• Storage Business Unit (SBU) — 18%
• Embedded Business Unit (EBU) — 15%
Micron has the following revenue segments. According to recent earnings reports from various semiconductor companies, mobile and automotive are exposed.
• Mobile — 25%
• Client and Graphics — 20%
• Enterprise and Cloud Server — 20%
• SSDs and other storage — 15%
• Automotive, Industrial and Consumer — 15%
Country 2019 Revenue in US$ Mil %
United States 12,451 53
Mainland China excl Hong Kong 3595 15
Taiwan 2,703 12
Hong Kong 1,614 7
Other Asia Pacific 1,032 4
Japan 958 4
Other 1,053 4
23,406 100
One of Micron’s strongest selling points is the addressable market of $83 billion for DRAM and $99 billion for 3D NAND by 2025. This is a combined addressable market of $182 billion.
Source: Micron Presentation
Future catalysts for NAND and DRAM include artificial intelligence and autonomous vehicles requiring data storage and memory capacities. In the long-term, the management believes it will benefit from secular growth in the industrial IoT market as 5G rolls out. Current markets include the data center and internet of things in addition to PCs and mobile smartphones
According to TrendForce, YMTC, a new competitor located in Wuhan, China, is set to compete with 128L products by the end of the year.
Technical Analysis
The above chart is a look at the weekly price pattern of Micron (MU). The larger the trend, the more important it is to the direction of the price. Since 2009, Micron has been trading within a leading diagonal pattern. This is a 5wave pattern that tracks along a trend channel (in gray). Each of the larger degree 5 waves (in red) are comprised of 3-waves (in blue).
According to this pattern, we are in the larger degree 4th wave (in red). Within this wave, we have completed the A and B wave. Therefore, we are in the middle of the final C-wave down. I will target the lower end of the trend channel, which we have not touched. There are a cluster of Fibonacci price levels around the trend channel between $34-$22.
The weekly RSI is also confirming that we are not yet in a renewed uptrend for MU. Until the RSI can break above the downward sloping trend line as well as break above 60, the momentum suggests the current uptrend off the March lows is a corrective move in a larger degree trend, which is pointing down.
It would be rare to see this larger degree pattern not follow the current trend. However, if price can break above the $61 level, which is confirmed by the weekly RSI, I will look at that level as a bullish move and a targeted entry to ride the new bull market in MU.
The daily chart shows this trend unfolding in real time. The uptrend’s structure off the March lows is overlapping and symmetrical. It further suggests weakness. This is also confirmed by the internals.
The volume is slowing down at current levels, suggesting that the participation at current prices is weakening. The Accumulation/Distribution line suggests that the smart money has not been buying into this uptrend, and in fact using it to unload shares. The MACD histogram and the MFI are showing notable weakness below the price as well.
All of this together further supports a topping pattern that is unfolding. If price can break below $41, this will confirm the target entries below.
