Arista Doubles Its AI Networking Guide as Memory Costs Test the Margin Story

Results vs. Consensus

Arista reported fiscal Q4 2025 (quarter ended December 31, 2025) after the close on February 12. The headline was an all-around beat on the two metrics that matter for a premium-multiple compounder: revenue growth and earnings. The quality was high, with the one blemish being a gross margin that continues to grind lower on cloud/AI mix.

Year-over-year comparison

Quarter-over-quarter comparison

Segment & Mix Performance

Arista does not report GAAP segments, but management frames the business along two axes that are essential to the thesis: customer sector (who is buying) and product category (what they are buying). Both were disclosed on an annual 2025 basis, and both matter for judging whether the AI concentration is a strength or a fragility.

Customer sector mix (FY2025)

Product category mix (FY2025)

2026 goals vs. 2025 results

Cloud & AI titans (48%)

This is the engine and the risk. The sector grew fastest, drove the international uptick (a stronger contribution from large global customers), and is the primary reason gross margin mix-shifted lower. Management reiterated it is in all four of its largest AI customers with Ethernet, three of which have now deployed a cumulative 100,000 GPUs each and are moving from pilots into production across multiple centers, with power as the gating constraint.

"With our increased visibility, we are now doubling from 2025 to 2026 to $3,250,000,000 in AI networking revenue." — Jayshree Ullal, Chairperson & CEO

Assessment: The doubling of the AI target is the single most important disclosure of the print and the reason we initiate constructively. The caveat is that this sector carries the lowest gross margin and the highest customer concentration, so its growth is dilutive to margin even as it is accretive to the multiple narrative.

Enterprise & financials (32%)

The quieter half of the business, and the one the bears zeroed in on. Enterprise is where campus, routing, and software attach live, and where Arista argues it is taking share from "legacy incumbents." Management cited a 93% net promoter score, 10,000+ cumulative customers, and particular traction in the $1M and $5–10M customer cohorts. But within the FY2026 guide, this is also where growth is implied to be muted after stripping out AI and campus.

"We now see the pace of acceptance and adoption accelerating in the enterprise customer base." — Jayshree Ullal, Chairperson & CEO

Assessment: Enterprise is the margin ballast and the diversification story, but it is not what the 2026 raise is built on. If the AI cohort ever stumbles, enterprise needs to be visibly re-accelerating to hold the multiple; right now it is steady rather than a growth driver.

AI & specialty providers (20%)

The recategorization here is substantive, not cosmetic. Oracle moved out of "cloud" into this bucket because its profile is now "dual personality" (OCI plus heavily AI-centric multitenant partnerships), and the sector now explicitly includes Apple, Oracle, and the emerging Neo Cloud names. Management drew a sharp line between healthy specialty providers pivoting to AI and a subset of "oil money converted into AI or crypto money converted into AI" that it is watching more carefully on customer health.

Assessment: This is where the next 10% customers most plausibly emerge, and also where the credit-quality tail risk sits. The candor about scrutinizing Neo Cloud balance sheets is a positive governance signal; it says Arista is chasing durable volume, not headline bookings.

Key Financial & Balance-Sheet KPIs

Two balance-sheet lines deserve to be read together: purchase commitments jumped $2.0B to $6.8B and deferred revenue rose to $5.4B. The first is Arista pre-positioning silicon and memory ahead of a supply squeeze; the second is revenue already sitting on the balance sheet awaiting customer acceptance. Both are, on balance, forward indicators of demand, but both also inject volatility into any quarter-to-quarter model.

Key Topics & Management Commentary

Overall Management Tone: Confident and, unusually for a networking company mid-supply-crisis, willing to raise the full year six months after an Analyst Day. Management was direct about the memory-cost problem rather than minimizing it, and repeatedly pushed back on attempts to extrapolate a precise full-year shape this early, framing the 25% guide as "a high dose of reality" rather than conservatism. The posture on demand was assured; the posture on 2026 margins and the non-AI core was deliberately non-committal.

1. AI Networking Guide Doubles to $3.25B

The centerpiece. Arista lifted its 2026 AI center networking target to $3.25B from the $2.75B set at the September Analyst Day, roughly doubling the 2025 result, which itself exceeded the $1.5B goal. Management framed AI as having "achieved production scale" on an all-Ethernet basis, spanning back-end training, front-end inference, and the emerging scale-across topology.

"Three years ago, we had no AI. We were staring at InfiniBand being deployed everywhere in the back end... Three years later, I am actually telling you we might do north of $3,000,000,000 this year and growing. That number definitely includes the front end." — Jayshree Ullal, Chairperson & CEO

Assessment: A raised AI target, disclosed with the tax and margin tailwinds stripped out, is the strongest signal in the print. Ethernet displacing InfiniBand in the back end is now a fact set, not a hope, and Arista is monetizing front-end attach on top of it.

2. The Memory & Silicon Cost Shock

The dominant risk of the call. Management described memory pricing as "an order of magnitude exponentially higher," called memory "the new gold for the AI and automotive sector," and flagged silicon fabrication cost increases flowing from a single leading-edge foundry. Arista absorbed much of this in 2025 but signaled it can no longer fully absorb it in 2026.

"In 2026, the situation has worsened significantly. We are having to smile and take it just about at any price we can get. And the prices are horrendous... we do believe there will be a one-time increase on select especially memory-intensive SKUs." — Jayshree Ullal, Chairperson & CEO

Assessment: This is the reason the stock's opening gap faded. The 62–64% full-year gross-margin range holds for now, but it holds because Arista is prepared to raise prices, and price increases into a demand-inelastic AI cycle are usually accepted. We treat this as a manageable-but-real margin overhang, not a thesis-breaker, and the single most important thing to track through 2026.

3. Operating Margin Steps Down to ~46%

Less discussed on the call but arguably the most concrete concession: the FY2026 operating-margin guide of ~46% is below the 48.2% Arista delivered in 2025. The company is choosing to invest through the cycle (R&D held near 11% of revenue) while absorbing component inflation.

"With our increased revenue guidance, we are now confident to raise the operating margin outlook to approximately 46% in 2026." — Chantelle Breithaupt, CFO

Assessment: "Raise to 46%" is technically a raise versus the Analyst Day framework, but on a year-over-year basis it is a ~200 bps step-down in profitability. For a stock priced on durable ~48% margins, this is the number the model has to absorb. It is defensible given the top-line acceleration, but it caps the earnings-growth rate below the revenue-growth rate in 2026.

4. Customer Concentration & the New 10% Customers

Arista had two greater-than-10% customers in 2025 (Customer A and Customer B), which together made up roughly 42% of revenue based on management's Q&A math (~26% and ~16%). The forward hook is that Arista expects "one, maybe even two additional 10% customers" in 2026 as the base diversifies.

"With our ever-increasing AI momentum, we anticipate a diversified customer base in 2026, including one, maybe even two additional 10% customers." — Jayshree Ullal, Chairperson & CEO

Assessment: Two customers at ~42% is a genuine concentration risk that the market has long tolerated because those customers are among the best-capitalized companies on earth. Adding a third and fourth 10% customer would be a clear positive, both for diversification and as evidence the AI TAM is broadening beyond the original titans.

5. New Customer Segmentation

Management restructured its sector reporting into Cloud & AI titans (48%), Enterprise & financials (32%), and AI & specialty providers (20%). The move of Oracle out of "cloud" and into "AI & specialty," alongside Apple and the Neo Clouds, is the tell: Arista sees an AI-centric customer archetype distinct from classic hyperscale.

"Especially with Oracle's AI, Acceleron, and multitenant partnerships that they have created, they have naturally got a dual personality... so the shift in their strategy made us shift the category and bifurcate the two." — Jayshree Ullal, Chairperson & CEO

Assessment: Re-segmentation just before a big guide can obscure comparability, but here it looks analytically honest rather than promotional. It gives investors a cleaner way to size the "pure AI" opportunity versus traditional cloud.

6. The "60% of the Business Grows ~Zero" Debate

The sharpest exchange of the call. A line of questioning argued that once you back out cloud and campus, the remaining ~60% of the business is being guided to roughly flat, versus 10–30% growth in prior years. Management refused to concede the point but also declined to promise better, leaning on the "it is February" argument.

"We are not guiding that our business is going to be flat... but, generally, when something is very fast-paced and growing, then other things grow less. And exactly whether it would be flat or grow double digits or single digits, Tal, it is February. I do not know what the rest of the year will be." — Jayshree Ullal, Chairperson & CEO

Assessment: This is the crux of the bear case and management did not fully rebut it. The most charitable read (and the one Arista's history supports) is conservatism: the full-year guide is deliberately loose in February and gets raised through the year. The less charitable read is that non-AI demand really is softening. We lean toward conservatism given the Q1 guide, but this is the number to watch.

7. Deferred Revenue Build & Acceptance-Criteria Opacity

Product deferred revenue rose again (+$469M QoQ; $5.4B total), a recurring source of analyst frustration because Arista will not guide it. New products, new customers, and new AI use cases carry customer-specific acceptance clauses of 12–18 months (some as short as six), so revenue timing is genuinely hard to predict.

"What goes into deferred is new product, new customers, new use cases. The great new use case is AI. The acceptance criteria for that for the larger deployments is 12 to 18 months... it is a net release of a balance, so it depends what comes in at that same quarter time." — Chantelle Breithaupt, CFO

Assessment: The build is a demand positive (it is future revenue already contracted), but the opacity is real and repeatedly frustrated the Street. Expect lumpy quarters where a large acceptance release either flatters or deflates a given period independent of underlying demand.

8. AMD & Open Multi-Vendor Momentum

A year ago Arista's AI deployments were "pretty much 99% NVIDIA." Management now sees 20–25% of deployments where AMD is the preferred accelerator, and argues Arista wins disproportionately in those open, best-of-breed configurations because AMD-centric customers want open standards rather than a single-vendor vertical stack.

"Today, when we look at our deployments, we see about 20%, maybe a little more, 20 to 25%, where AMD is becoming the preferred accelerator of choice. And in those scenarios, Arista is clearly preferred." — Jayshree Ullal, Chairperson & CEO

Assessment: Accelerator diversification is structurally good for Arista: the more heterogeneous the compute layer, the more valuable a vendor-neutral Ethernet fabric becomes. This is an underappreciated tailwind versus the "NVIDIA does its own networking" bear narrative.

9. Neo Cloud: Opportunity vs. Customer-Health Caution

Management drew a careful distinction within the Neo Cloud cohort. Some are healthy, well-capitalized providers pivoting into AI; others are speculative, thinly capitalized, and treated cautiously precisely because their business durability is unproven.

"There are a set of neo clouds that we watch more carefully, because some of them are oil money converted into AI or crypto money converted into AI. And over there, we are going to be much more careful." — Jayshree Ullal, Chairperson & CEO

Assessment: The willingness to say this out loud, rather than book every Neo Cloud dollar as equal, is a quality signal. It also implies Arista's AI backlog is somewhat self-selected toward creditworthy demand, which lowers the risk of a bookings air-pocket if the speculative tier retrenches.

10. Scale-Up / Scale-Out / Scale-Across & the ESUN Roadmap

On architecture, management positioned Arista across all three AI network domains. Today's deployments are mostly scale-out and scale-up on 800G; scale-across (connecting distributed data centers via DCI and coherent optics on the 7800 spine) is emerging as workloads outgrow a single building. Scale-up standardization hinges on the ESUN spec, expected at the earliest in Q4 2026, with production volume centered on 2027 and 1.6T.

"We will tie our scale up commitment greatly to availability of new products and a new ESUN spec, which we expect the earliest to be Q4 this year... the real production level will be in 2027. Primarily centered around not just 800 gig, but 1.6T." — Jayshree Ullal, Chairperson & CEO

Assessment: Scale-across is a genuine incremental TAM that plays to Arista's routing and DCI heritage, and the 7800 spine is a credible flagship for it. Scale-up is a 2027 story, so it is optionality on top of the 2026 guide rather than embedded in it.

11. Campus, Routing & the VeloCloud Adjacency

The adjacencies (campus plus routing) are ~18% of revenue combined. Campus is guided to $1.25B in 2026 after exceeding the $800M goal in 2025, anchored by cognitive wired/wireless, the VESPA and SWAG differentiators, and the July 2025 VeloCloud acquisition that extends a unified client-to-branch-to-campus stack. Routing added the 7800R4 spine (460 TB capacity) for DCI, AI spine, and multi-service use cases.

"Together with our recent VeloCloud acquisition in July 2025, we are driving that homogenous secure client to branch to campus solution with unified management domains." — Jayshree Ullal, Chairperson & CEO

Assessment: Campus and routing are the diversification hedge against AI concentration, and both are executing to plan. VeloCloud's service-renewal contribution also explains the noisy sequential move in the services/subscription mix (17.1% in Q4 versus an elevated 18.7% in Q3).

Guidance & Outlook

The shape of the guide is the story. Q1 at ~$2.6B implies roughly 30% YoY growth, yet the full year is guided to only 25%. That gap is the essence of Arista's playbook: set a full-year number in February that the company can beat and raise through the year. Management effectively confirmed this dynamic by refusing to characterize the back half beyond "it is February."

Implied ramp: Hitting ~$11.25B against a ~$2.6B Q1 requires roughly $8.65B across Q2–Q4, or an average of ~$2.88B per quarter, i.e., continued sequential growth off the Q1 base. That is achievable if AI acceptance releases land as expected; the swing factor is deferred-revenue timing, not end demand. Street at: consensus was tracking near the low-to-mid $11B range for FY2026 revenue pre-print and near ~$2.45B for Q1, so both the annual and quarterly guides came in above the Street. Guidance style: conservative by Arista's own multi-year pattern; the February full-year number has historically proven to be a floor.

Analyst Q&A Highlights

Why only 25% for the year when Q1 implies 30%?

The most-pressed question on the call: with the Q1 guide implying ~30% growth, several analysts probed why the full-year number sits at 25% and whether the gap reflects caution about new-customer ramps or supply. Management rejected the "cautious" framing and reframed the 25% as realism about what can physically ship this early in the year.

Q: "I do not want to sort of say that the 25% growth is not impressive, but since you are doing 30% is what the guidance is for Q1, maybe if I could understand what is maybe sort of leading to somewhat of a cautious terms of visibility for the rest of the year?"
— Samik Chatterjee, JPMorgan

A: "First, I do not think I am being cautious. I think I went all out to give you a high dose of reality... I am giving you the greatest visibility I can, fairly early in the year, on the reality of what we can ship, not what the demand might be. It might be a multiyear demand that ships over multiple years... we are also facing a law of large numbers."
— Jayshree Ullal, Chairperson & CEO

Assessment: Management wants investors to read 25% as a shippable floor, not a demand ceiling. Given the Q1 setup and Arista's track record, we agree, but the honest read is that visibility past mid-year is genuinely low and demand-to-revenue conversion hinges on acceptance timing.

Can you quantify the memory and gross-margin impact?

A recurring line of questioning tried to parameterize the memory headwind: how much revenue and margin risk is embedded, and how much was pre-purchased at lower prices. Management gave qualitative color rather than numbers, and the CFO defended the 62–64% range as deliberately set to absorb the pressure.

Q: "Can you help quantify sort of both the revenue impact and potential kind of gross margin impact embedded in your guide from the memory dynamics and the constraints?... give us a sense for what upside could be, in a perfect world."
— David Vogt, UBS

A: "We are comfortable in the guide, and that is why we... raised the numbers that we did... The range of 62 to 64, I think we are pleased to hold despite this kind of pressure coming into it. This has been our guide since September at our Analyst Day... whether it ends up being 62.5 versus 63.5 in the guide in that range, that is where we will continue to update you." — Chantelle Breithaupt, CFO

Assessment: Holding the September margin range through a worsening cost environment is a subtle positive; it says the raise is real, not funded by quietly guiding margins down. But the refusal to quantify leaves the downside tail undefined, which is why the market did not pay full price for the raise.

Is 60% of the business really guided to no growth?

The most confrontational exchange. An analyst dissected the guide to argue that ex-cloud and ex-campus, the residual ~60% of revenue is implied to grow roughly zero, versus 10–30% historically, and asked whether memory allocation was starving the non-AI business. Management pushed back on both the math and the allocation premise.

Q: "If I remove campus and I remove cloud... the rest of the business, which is 60% of the business, you guide it to grow zero. And in previous years... it was anywhere from 10% to 30% growth. So the question is, why are you guiding this way?... are there allocations here?"
— Tal Liani, Bank of America

A: "It is neither of the above. We do not allocate to our customers. It is first in, first served. And in fact, the enterprise customers get a very high sense of priority... We are confident that we could guide, six months after our Analyst Day, to a higher number. We do not know what the next four quarters will look like to the precision you are asking for." — Jayshree Ullal, Chairperson & CEO

Assessment: Management denied the flat-core interpretation without replacing it with a firmer number, which will keep this debate alive. The allocation denial is reassuring (enterprise is prioritized, not starved); the growth ambiguity is not. This is the bear case's strongest foothold.

How does product deferred revenue come off the balance sheet?

A recurring topic given the persistent build. Analysts wanted to know whether deferred releases in big chunks or gradually, and how much of the FY2026 guide depends on it. Management declined to guide deferred and reiterated that acceptance timing is customer-driven and unpredictable.

Q: "Does that come off in big chunks that we will see in different quarters in the future? Does it come off more gradually?... how much product deferred do you have in the full year revenue guidance to 25%?"
— George Notter, Wolfe Research

A: "It is super hard, George. It is when the acceptance criteria happens. If it happens December, it is a different situation. If it all happens in Q2, Q3, Q4, that is a different [situation]... Sorry that we are not able to be clairvoyant on that." — Jayshree Ullal, Chairperson & CEO

Assessment: The honest answer is that even management cannot forecast deferred timing precisely. Investors should size positions expecting lumpy prints and resist over-reading any single quarter's beat or light number as a demand signal.

How real is AMD accelerator attach?

Questioning on the competitive-accelerator landscape sought to size AMD momentum and whether new AMD product launches are a networking catalyst. Management quantified the shift from a year ago and tied Arista's win rate to open, multi-vendor architectures.

Q: "If you do not mind touching on AMD, with the launch, we are kind of hearing about you getting a lot of networking attached to the 450 type product or their new chips. Wondering if that is a catalyst or not as you go throughout the year."
— Benjamin Reitzes, Melius Research

A: "A year ago, it was pretty much 99% NVIDIA. Today... we see about 20%, maybe a little more, 20 to 25%, where AMD is becoming the preferred accelerator of choice. And in those scenarios, Arista is clearly preferred because they are building best-of-breed building blocks... and they want open standards as opposed to full-on vertical stack from one vendor." — Jayshree Ullal, Chairperson & CEO

Assessment: This is one of the most bullish disclosures on the call and got less airtime than it deserved. Accelerator heterogeneity is a direct tailwind to a vendor-neutral fabric, and it undercuts the thesis that NVIDIA's in-house networking caps Arista's AI TAM.

Where do the four large AI customers stand?

A question sought an update on the marquee AI cohort, specifically the fourth customer that had been slower to ramp and a fifth that had hit funding challenges. Management confirmed three have crossed 100,000 cumulative GPUs and the fourth is migrating off InfiniBand.

Q: "I was hoping to ask for an update on the Arista four large AI customers. I know that that fourth customer... was a bit slower to ramp to 100,000 GPUs... is there any indication that that fifth customer that ran into funding challenges might come back to you?"
— Adrienne Colby (for Atif Malik), Citi

A: "We are in all four customers deploying AI with Ethernet... Three of them have already deployed a cumulative of 100,000 GPUs, and are now growing from there... Our fourth customer is migrating from InfiniBand. So it is still below 100,000 GPUs at this time. But I fully expect them to get there this year." — Jayshree Ullal, Chairperson & CEO

Assessment: The fourth customer migrating off InfiniBand is a concrete Ethernet-displacement data point and a 2026 growth lever. Management declined to revisit the fifth customer, which we read as that opportunity remaining dormant.

Does the customer math imply a new near-10% hyperscaler?

Questioning tried to reconcile the sector percentages with the two named 10% customers to infer whether a new large customer is approaching the threshold, and probed the philosophy behind the new segmentation.

Q: "With cloud and AI at 48% of revenue and A and B... combined [42%], you have [6%] left over. Is that a hyperscale customer? Does it kind of imply that you have a new hyperscaler that is approaching 10%?"
— Michael Ng, Goldman Sachs

A: "On the cloud and AI titans, the way we classify that is it is significantly large scale customers with greater than a million servers, greater than 100,000 GPUs, an R&D focus on models and sometimes even their own XPUs... it is a very select few set of customers, less than five or about five." — Jayshree Ullal, Chairperson & CEO

Assessment: Management neither confirmed nor denied a new near-10% titan, but the framing of "about five" titans plus the promise of one-to-two new 10% customers points to a broadening base. That broadening is exactly what would de-risk the concentration critique over the next few quarters.

What They're NOT Saying

1. The dollar size of the memory hit: Management repeatedly declined to quantify the revenue or margin impact of memory/silicon inflation, offering only the 62–64% range. The magnitude of the potential price increase on memory-intensive SKUs is undefined, leaving the margin downside tail open.
2. Ex-AI/ex-campus growth: No explicit growth rate was given for the ~60% "core" business, and the direct question about whether it is guided to zero was deflected with "it is February." The omission is conspicuous given how pointedly it was asked.
3. Deferred revenue in the guide: Management would not say how much of the 25% FY2026 growth depends on deferred releases, nor whether the balance builds or bleeds down through the year. This is the single largest modeling unknown.
4. Named 10% customers: Arista continues to refer to "Customer A" and "Customer B" without identification, and would not confirm whether a specific new hyperscaler is approaching the 10% threshold.
5. The fifth AI customer: The customer that hit funding challenges was raised by an analyst and effectively passed over, implying that opportunity is not in the 2026 plan.

Market Reaction

• Pre-print setup: ANET closed at $135.12 entering the print, up 3.1% YTD, up 23.2% over the trailing 12 months, and near the top of its 52-week closing range ($64.37–$162.03). The options market implied a roughly 10% move.
• Reaction (T+1, Feb 13): The stock opened +7.9% at $145.82, traded as high as $148.77 (+10.1%), then faded to close +4.8% at $141.59. Volume was 21.8M shares, about 2.7x the 30-day average. The S&P 500 was flat on the session, so essentially all of the move was idiosyncratic.

The pattern is a partial "sell the news": a clean beat-and-raise gapped the stock up near the top of its implied move, then the intraday give-back reflected the memory-cost overhang and the ~46% operating-margin guide sitting below 2025's 48.2%. A +4.8% close on a ~10% implied move, into a stock that had already run 23% over the prior year, is the market paying for the raise but discounting it for margin risk and a full multiple. At the reaction close, ANET trades around 47x trailing and high-30s-to-40x forward non-GAAP earnings, a premium that requires the beat-and-raise cadence to continue.

Street Perspective

Debate: Is Arista's AI Ethernet share durable, or will white box, co-packaged optics, and NVIDIA in-house networking erode it?

Bull view: The bull case on the Street holds that Ethernet has decisively won the AI back-end standards battle (UEC 1.0, ESUN), that EOS plus CloudVision telemetry is a genuine software moat, and that accelerator diversification toward AMD structurally favors a vendor-neutral fabric. Three of four marquee customers at 100K+ GPUs and a fourth migrating off InfiniBand are cited as proof of displacement.

Bear view: Skeptics argue that hyperscalers increasingly design their own switches and could in-source, that co-packaged and coherent optics change the economics of the switch layer, and that a single-vendor AI stack from the dominant GPU supplier could re-bundle networking. Concentration in a few titans makes any one design-out material.

Our take: The bull case has the better of it for now. The AMD data point and the InfiniBand-to-Ethernet migration are hard evidence, and Arista's software layer is stickier than the "it's just a switch" bear framing allows. In-sourcing is a multi-year tail risk, not a 2026 one.

Debate: Is memory-cost inflation a manageable pass-through or a real margin reset?

Bull view: The optimistic camp notes Arista held its 62–64% gross-margin range through a worsening cost environment, has $6.8B of purchase commitments locking supply, and sells into an inelastic AI demand curve where price increases are absorbed. Peers in servers hit this first and passed it through.

Bear view: The bear camp points out that gross margin is already down 180 bps sequentially on mix alone, that the operating-margin guide stepped down to ~46%, and that management refused to quantify the hit, leaving the downside undefined into a year where costs are "an order of magnitude" higher.

Our take: Both are partly right. Arista will likely pass through most of the cost via SKU-level pricing, so gross margin probably holds inside the range, but the ~46% operating-margin guide is a real, if modest, profitability reset that caps 2026 EPS growth below revenue growth. Manageable, not immaterial.

Debate: Does a ~40x forward multiple leave room, given decelerating growth?

Bull view: A growing consensus view is that 25% guided growth is a conservative floor that becomes ~28–30% after the customary raises, that the AI TAM ($100B-plus) is early, and that a rare pure-play AI networking leader with 46% operating margins and a fortress balance sheet deserves a premium to the market and to networking peers.

Bear view: Some desks argue growth is decelerating (28.6% in 2025 to a guided 25%), that margins are compressing, and that at ~40x forward the stock is priced for perfection, leaving asymmetric downside if any single large customer's capex timing slips.

Our take: The multiple is full but not indefensible for this growth-and-margin profile, and the conservative guide gives estimates room to move up. We would not chase the stock on a gap-up, but we are comfortable owning it and adding on pullbacks, which is what the Outperform reflects.

Model Update Needed

Valuation impact: The revenue raise is partly offset by the tax normalization and the ~200 bps operating-margin step-down, so FY2026 EPS moves up modestly rather than dramatically. On a high-30s-to-40x forward multiple, we frame fair value as tracking earnings revisions higher through the year rather than re-rating; the Outperform is an earnings-compounding call, not a multiple-expansion call.

Thesis Scorecard Post-Earnings

This is initiation of coverage, so the scorecard establishes the standing thesis rather than grading a prior one.

Overall: The demand side of the thesis is confirmed emphatically (raised AI guide, Ethernet displacement, accelerator diversification); the cost and valuation side introduces real but manageable offsets. On balance the risk/reward favors ownership.

Action: Initiate Outperform. Build a position with room to add on volatility; the memory-cost trajectory and the ex-AI core growth rate are the two variables that would move us to Hold if they deteriorate.