Arista Networks: A Long-Term Study of the Data Center Switching Leader

Cost Structure and Dependencies

Fact: Arista runs a typical asset-light model: most manufacturing is outsourced to contract manufacturers such as Jabil, Sanmina, and Foxconn; the company itself stresses its reliance on a limited number of key component suppliers, and its switching silicon in particular depends primarily on Broadcom. At the end of 2025 the company had three principal contract manufacturing partners providing the bulk of electronics manufacturing services.

Fact: Customer concentration is high. In 2025 two end customers accounted for 26% and 16% of revenue respectively; in 2024 they were 20% and 15%; in 2023 they were 18% and 21%. The company did not name these customers in its 2025 annual report, but in its 2022 annual report it explicitly identified Microsoft and Meta Platforms, which together accounted for 42% of revenue that year. In addition, at the end of 2025 the top two distributors accounted for 52% of accounts receivable. This means that, however strong its technology and brand, Arista remains deeply exposed to the purchasing rhythm of a handful of mega-customers.

Can I understand this business

I believe I can understand it, though it is not "extremely simple." It is not a "selling water" kind of business, but rather the supply of critical infrastructure to hyperscale cloud, AI data centers, and enterprise networks through a software-defined high-performance network operating system, high-speed Ethernet switching and routing platforms, and support services. The value drivers are not mysterious: higher bandwidth, lower latency, better automation and observability, higher reliability, and lower total cost of ownership. But because the customers are hyperscalers, AI and cloud providers, and large enterprises, the project-based nature, testing and certification, supply chain, and competition over technology roadmaps complicate the financial rhythm. If the stock market closed for five years, then purely from the standpoint of "would I want to own the business," I would be happy to own Arista; but from the standpoint of "should I buy at today's price," I lean toward waiting.

Business understandability score: 4/5. The high-level logic is clear and the financial structure is not complicated; what is genuinely complex is customer behavior and the technology roadmap, not the business model itself.

Industry and Competitive Landscape

Fact: External industry data shows that the data center switching market is still in a growth phase, and that AI infrastructure is markedly accelerating it. IDC reports that the data center switching market grew 63% year over year in Q4 2025, with full-year Ethernet switching market revenue of about 55.1 billion dollars, up 31.5% year over year; Dell'Oro notes that Arista retained its overall Ethernet data center switching market leadership in 2025.

Fact: The competitive landscape is intense. In its 10-K Arista itself names Cisco, which has long dominated traditional data center and campus networking, with competitors also including HPE, Juniper, NVIDIA, Extreme, and white-box switch vendors; in AI back-end networking, Ethernet also competes with InfiniBand and NVLink. In October 2025 NVIDIA publicly announced that Meta and Oracle would adopt its Spectrum-X Ethernet switching solution. Inference: This is not a "good industry with no competition," but an industry where demand is excellent, technology advances quickly, and competition is brutal. Arista is a "strong company in a good industry," not a "near-invincible monopoly that can coast."

Fact: On share, Arista's slice of the overall global Ethernet switching market is smaller than Cisco's, but it carries heavy weight in the high-end data center Ethernet switching segments that matter. In its Q1 2025 figures IDC noted that about 90.9% of Arista's switch revenue came from the data center portion, giving it about 13.9% of overall global switching market revenue. Inference: Arista's true position is not "largest in the whole industry," but one of the strongest few in the high-performance Ethernet data center segments that matter most, and in certain cloud and AI scenarios a benchmark supplier.

Industry attractiveness score: 4/5. Demand trends up over the long run, but competition, technology substitution, and customer concentration keep it from a perfect score.

Moat and Management

Moat Breakdown

My Judgment

View: Arista's core moat is not a single point but a combination punch of "the EOS software stack, an engineering reputation, hyperscaler qualification relationships, a high-speed Ethernet product cadence, and an asset-light manufacturing model." Inference: This moat is broadly stable today and still shows signs of widening in enterprise networking and front-end / general data center networking; but on the hottest and most lucrative battlefield, AI back-end networking, the competition it faces is actually heating up, as NVIDIA enters with an integrated GPU-plus-networking approach and has already won public references at Meta and Oracle.

Inference: For a competitor to replicate Arista's current position, hardware alone is far from enough; what is genuinely hard to copy is the long-iterated software, the operational tooling, the quality validation system, and the large-customer qualification process. For a capable large vendor this is not "impossible," but building a truly mature, substitutable solution often takes more than three to five years, sustained R&D investment in the billions of dollars, and validation across multiple large-customer product cycles. This judgment is an inference, not a company-disclosed figure. It rests on the fact that the company explicitly states its large-customer sales cycles are long and its products must pass testing and certification, and that the industry's competition is highly technical.

Pricing Power, Inflation, and Downturns

Fact: Arista held a GAAP gross margin of 61.9% to 64.1% across 2023 to 2025, and still 61.9% in Q1 2026; this indicates a degree of differentiation and premium pricing. But it is equally a fact that the company explicitly disclosed a decline in gross margin from 63.7% to 61.9% in Q1 2026, driven mainly by a higher mix of large-customer sales at deeper discounts. View: This shows Arista has limited pricing power, but it is not a consumer brand that can raise prices at will; its pricing power is more "technical bargaining power" than "sentiment-driven brand premium."

Fact: The company has no interest-bearing debt and a very high net cash position, so even if the industry pulls back it is relatively well placed to ride through the cycle. Inference: In a downturn it would most likely remain profitable, but margins and cash flow would come under pressure from delayed customer projects, deeper discounts, and inventory and purchase commitments. In particular, as of Q1 2026 off-balance-sheet non-cancelable purchase commitments had risen to 8.9 billion dollars — a "preparatory move during a boom," which could also hurt cash efficiency in reverse if the cycle turns down.

Moat strength score: 4/5. Not a Coca-Cola-style moat, but quite rare within enterprise infrastructure.

Management and Capital Allocation

Fact: CEO Jayshree Ullal and CTO and President Kenneth Duda have long been in place; the 2026 proxy filing shows the CEO holds or controls about 29.3 million shares, Kenneth Duda about 3.519 million shares; the Bechtolsheim family trust holds about 183.8 million shares, or about 14.6%; all current executive officers and directors together hold about 33.47 million shares, or about 2.7%. At the broader level of directors, officers, and holders of more than 10% of shares combined, the company discloses combined influence of about 17.3%. This shows the core insiders and founder-related parties are not "without skin in the game."

Fact: Management's capital allocation is broadly disciplined. The company pays no dividend; in 2025 it completed the acquisition of the VeloCloud business for cash consideration of 300 million dollars, modest relative to its own cash and market cap; in 2024 and 2025 it repurchased about 424 million dollars and 1.603 billion dollars of stock respectively, with shares outstanding at the end of 2025 falling from 1.2613 billion to 1.2565 billion, indicating that buybacks have begun to outpace part of the equity-compensation dilution, though overall it looks more like "offsetting dilution plus modest cash return" than aggressive buybacks at depressed valuations.

Fact: On governance, strengths and blemishes coexist. The strengths are that the company has share ownership guidelines and a clawback policy, and that in 2025 the CEO voluntarily forwent new equity grants, which the company discloses is related to dilution management. The blemish is that in 2025 the advisory say-on-pay vote on executive compensation passed with only about 62% support, which is not pretty; the company subsequently conducted supplemental shareholder outreach, engaging 18 shareholders representing more than 30% of shares outstanding. View: This shows management does not ignore shareholder voices, but pay governance has not reached a "top-grade" level of shareholder friendliness.

View: My overall assessment of management is "capable, technically driven, long-term oriented, but not a perfect capital-allocation textbook." Its strength lies in product and strategic execution, not in manufacturing per-share value growth through financial engineering. The VeloCloud acquisition is still too early to prove value creation; and while the buybacks are not unreasonable, they hardly amount to "buying heavily into clear undervaluation."

Management and capital allocation score: 3.5/5.

Financial Quality

Key Financial Metrics

• TTM YoY is an approximate inference on a trailing basis. ** Understood on a comparable basis after the subsequent stock split; a strictly consistent historical comparison should refer to the company's split disclosure.

Table note: 2021–2022 data comes mainly from the 2022 annual report; 2023–2025 from the 2025 annual report; TTM is computed on a trailing basis using the 2025 annual report, the 2025Q1 10-Q, and the 2026Q1 10-Q, and is an inference. For 2022 capex and operating margin, I did not obtain the original same-page table in the verified materials at hand, so I clearly mark them "needs additional."

How to Read These Numbers

Fact: Arista's revenue grew from 2.948 billion dollars in 2021 to 9.006 billion dollars in 2025, roughly tripling over four years; net income rose faster, from 841 million to 3.511 billion; the GAAP operating margin improved from 38.5% to 42.8% across 2023 to 2025, reflecting scale effects and an improving product mix. View: This is not a "high growth but no profit" story, but the rarer "high growth and high profitability."

Fact: Cash flow quality is strong overall, but not equally pretty every year. In 2022 operating cash flow was only 493 million dollars, well below net income of 1.352 billion dollars, which the company attributes mainly to a significant increase in working capital needs that year; whereas operating cash flow across 2023 to 2025 was 2.034 billion, 3.708 billion, and 4.372 billion dollars respectively, and in Q1 2026 alone operating cash flow reached 1.694 billion dollars. Inference: Arista's profits are broadly real, but cash flow is affected by inventory, deferred revenue and deferred costs from acceptance terms, and tax timing, so a final judgment should not rest on a single quarter's or year's FCF alone.

Fact: In 2025 operating cash flow exceeded net income, with reasons including a 2.452 billion dollar increase in deferred revenue; in Q1 2026 cash flow was again boosted by an 826 million dollar increase in deferred revenue and a 353 million dollar increase in taxes payable. At the same time, increases in "other assets," accounts receivable, and inventory in 2025 reflect that product deferral and supply chain preparation are also consuming cash. View: This means Arista's free cash flow is very strong, but part of its "exceptional strength" is a growth-driven dividend of customer prepayments and deferred recognition, and should not be treated entirely as perpetually distributable cash.

Fact: Capex is extremely light. In 2025 related spending including intangible asset purchases was about 119.5 million dollars, only about 1.3% of revenue; on a TTM basis about 146 million dollars, or about 1.5% of revenue. Inference: Arista's growth does not depend on heavy capital investment, one of the hallmarks of an exceptionally high-quality business model.

Fact: The balance sheet is extremely sound. At the end of 2025 total assets were 19.449 billion dollars and total liabilities 7.078 billion dollars, the largest liability being deferred revenue rather than interest-bearing debt; at the end of Q1 2026 cash and marketable securities were about 12.35 billion dollars, while purchase commitments had risen to 8.9 billion dollars. Inference: Solvency risk is barely an issue; what truly deserves watching is purchase commitment and inventory risk, not a balance-sheet blowup.

Fact: Over the past three years the company has consistently recognized inventory write-downs, of 131.6 million, 267.2 million, and 234.4 million dollars in 2025, 2024, and 2023 respectively. This is not a bad thing in itself; if anything it shows management is recognizing part of the inventory risk in a timely way. On top of that, the company's 2025 internal controls received an unqualified opinion from EY. View: I currently see no obvious signs of financial fraud, aggressive profit recognition, or material accounting manipulation, but deferred revenue, deferred costs, and large-customer acceptance terms do raise the bar for reading the statements.

On Returns

Fact: On a simpler, verifiable basis, Arista's ROE was about 31% in 2025, about 33% in 2024, and about 35% in 2023; ROA was about 21% in 2025. Inference: If the enormous net cash and deferred revenue were stripped out, Arista's true return on operating capital would be even higher, high enough to distort the traditional ROIC measure. I therefore prefer to judge capital quality by "high margins plus very low capex plus net cash plus rapidly accreting retained earnings," rather than placing faith in a single ROIC formula.

Owner Earnings and Intrinsic Value

A Conservative Estimate of Owner Earnings

Fact: On a TTM basis to 2026Q1, Arista's trailing net income was about 3.721 billion dollars; trailing depreciation and amortization about 82 million dollars; trailing operating cash flow about 5.424 billion dollars; trailing capex about 146 million dollars. If you accept reported free cash flow in full, TTM FCF is roughly 5.278 billion dollars.

But I would not treat 5.28 billion dollars directly as owner earnings. There are three reasons:

First, stock-based compensation is a real cost. TTM stock-based compensation was about 467 million dollars, and not treating it as a cost is the same as handing dilution to management and employees for free. Second, in the 2025 and 2026Q1 cash flows, deferred revenue and tax timing made a clear positive contribution to CFO, and this part does not equal an infinitely stable, repeatable "true dividend capacity." Third, Arista's business captures a cash advantage from customer prepayments and deferred recognition during high-growth periods, but if growth slows, that cash advantage converges.

Conservative Owner Earnings Estimate

View: I would put Arista's current real, distributable annualized owner earnings in the 3 billion to 3.4 billion dollar range, rather than optimistically reading it as the 5 billion-plus of surface FCF. This is conservative, but it fits the standpoint of a "long-term business owner." On that basis, at the current market cap of about 196.2 billion dollars, ANET corresponds to roughly 57 to 65 times conservative owner earnings. That is not cheap by any "conservative value investor" standard.

Valuation Method One: Owner Earnings Discounting

What follows is not "forecasting the share price" but examining, under different assumptions, whether buying the entire company today is worthwhile.

A note on assumptions: These valuations already embed the premise that "the company is of very high quality, but the valuation cannot be infinitely expensive"; if you switch to more optimistic FCF or a lower discount rate, the valuation rises markedly. View: For a balanced, conservatively inclined investor, the current price of around 156 dollars is already near or slightly above the upper bound of my optimistic scenario. That means you are not buying "cheap," but buying "the possibility of continued upside surprise."

Valuation Method Two: Relative Valuation

Fact: Arista's current P/E is close to NVIDIA's and far above Cisco's and HPE's. Inference: The market is not assigning ANET a "networking-equipment stock valuation" but a "high-quality AI infrastructure growth stock valuation." If growth over the next three to five years cannot stay markedly above that of mature networking vendors, that valuation is fragile.

Supplementary Relative Measures

Fact: By my trailing financial estimates, ANET currently corresponds to about 37 times P/FCF, about 43 times EV/EBITDA, and more than 14 times P/B. A note: These are inferred calculations based on the 2026Q1 balance sheet and trailing profit and cash flow; the aim is not to create a sense of decimal precision but to show that the valuation is notably high. At the same time, because of HPE's basis changes after acquiring Juniper and the substantial differences in business structure between Cisco and Arista, I have not forced a consistent EV/EBITDA/PB for every peer in this report, so as not to "mislead with precision."

Valuation Method Three: Asset or Liquidation Value

Fact: At the end of 2026Q1, Arista's cash and marketable securities were about 12.35 billion dollars, with almost no interest-bearing debt; at the end of 2025 book goodwill was about 416 million dollars, mainly from the VeloCloud acquisition. View: This company is clearly not attractive on liquidation value; its value comes almost entirely from its ability to keep generating high-return cash flow in the future. Book net assets explain little of it, but net cash provides a meaningful risk buffer.

Price Ranges

View: If you hold to "leave at least a 25% to 30% margin of safety," then I would want to buy at no higher than around 90 to 115 dollars. At today's price of about 156 dollars, a fresh purchase does not fit that discipline.

Margin of Safety and Risk

Is the Margin of Safety Sufficient

My answer is: not sufficient.

Fact: ANET's current trailing P/E is about 48.6x; on my more conservative owner earnings, the implied multiple is higher still; meanwhile the U.S. 10-year Treasury yield was about 4.57% on 2026-05-21. Inference: Arista's current earnings yield and free cash flow yield are both below the risk-free rate, so the investment case requires high growth and high margins to be sustained for many years to come. In other words, in buying it today, you are buying not current cash returns but continued future upside surprise.

There are three most fragile assumptions in the valuation. First, whether AI-related networking demand can stay in a multi-year boom rather than going through the typical overbuild-and-digestion cycle. Second, whether the two largest customers will still be willing to direct purchases mainly to Arista rather than spreading them across NVIDIA, Cisco, HPE Networking, and others. Third, whether gross margin can hold above 60% under large-customer discounts and component cost pressure. If any one of these assumptions is clearly broken, the current valuation could compress.

If growth falls short of expectations, does the investment still hold? View: For "business quality," it still holds; for "the investment return at the current price," not necessarily. Arista is not the kind of company that swings to a loss the moment growth slows, but it is the kind of company whose valuation elasticity disappears first the moment growth slows. For a stock at a 48x P/E, the greatest risk is not that the business fails tomorrow, but that results over the next five years are decent while shareholder returns are merely mediocre.

If margins decline, does it still hold? Partly. Because the company has net cash, extremely light capex, and a very high baseline margin, even if margins fall from the 42% GAAP operating margin back toward 30%, it would still be a high-quality business. But the fact is that if margins pull back while the valuation multiple compresses at the same time, a buyer at the current price could face a long stretch of "good company, bad returns."

The Most Important Risk List

The Strongest Bear Case

The bear case would put it this way: Everyone knows Arista is "good," and the share price has long since fully reflected it. The market is pricing it as a "core beneficiary of AI infrastructure," but the reality is that Arista will not be the only beneficiary of AI networking; NVIDIA can even package GPUs, networking, and software to sell to customers; Meta and Oracle have publicly adopted NVIDIA's Spectrum-X; once large customers reallocate orders, Arista's revenue and margins would come under pressure simultaneously. On top of that, with the current yield below Treasuries, investors are effectively prepaying many years of certainty at a very high price. The moment future growth shifts from "very fast" to "decent," that is enough to cause permanent capital loss — not bankruptcy of the company, but buying high and not recovering for years.

Which Facts Would Overturn the Investment Judgment

If the following facts appeared, I would conclude that the original "high-quality business" judgment needs a clear downgrade: One, the loss of a major customer leading to a concentrated revenue decline that cannot be made up by new customers. Two, gross margin falling below 58% to 60% for years with no recovery in sight. Three, cloud and AI Ethernet share being eroded by NVIDIA, Cisco, and HPE over the long run, with Arista no longer the technology and execution leader. Four, management beginning to make high-premium, asset-heavy acquisitions far from the core business. Five, stock buybacks and compensation incentives evolving over the long run into a cycle of "high-valuation dilution and high-valuation buybacks." These thresholds are a view and tracking framework, not company disclosure.

Checklist and Final Investment Conclusion

Comparison with Other Opportunities

The strongest peer competitor: Looking only at traditional network infrastructure, Cisco remains the larger, more complete incumbent leader; looking at AI back-end networking, the most dangerous integrated competitor is in fact NVIDIA. After completing the Juniper acquisition, HPE's networking business has also moved up a level in completeness. View: Arista's advantage is "a purer, higher-quality, more focused software-driven network platform"; the rivals' advantage is "greater size, more bundling ability, and a broader product stack."

Comparison with the S&P 500: As of 2026-05-22, the S&P 500 closed at about 7473.47. The S&P represents broadly diversified large-cap U.S. equity risk; Arista is a high-quality but highly concentrated single-technology-infrastructure bet. View: At the current price, ANET is not clearly superior to buying the index, because you take on higher customer concentration, technology substitution, and valuation compression risk for a single company. Absent a meaningful price advantage, I would rather direct new money to the index or keep it in high-grade bonds and cash while waiting.

Comparison with the risk-free yield: The current U.S. 10-year Treasury yield is about 4.57%. ANET's current P/E corresponds to an earnings yield of about 2.1%, and even on surface FCF the FCF yield is only low single digits. Inference: Only if you believe it can still grow at high double digits for many years and keep an outstanding moat does today's price make sense. Otherwise the opportunity cost is not low.

If you could hold only five assets, does it qualify for the portfolio? The business quality qualifies; the current price does not. That is, it belongs on the list, but now is not the moment I would buy it heavily.

Investment Checklist

The conclusions of this checklist are drawn from a synthesis of all the facts and inferences above.

Open Questions and Limitations

This report's most important conclusion does not depend on the following gaps, but they are worth continuing to track: One, after HPE's acquisition of Juniper, a fully consistent EV/EBITDA, P/B, and ROIC comparison with Arista is disturbed by accounting-basis differences; two, for certain 2022 historical line items such as capex and operating margin, I did not cross-verify them page by page in this round of high-confidence materials, and so did not force in figures; three, management's higher-level full-year 2026 guidance and AI revenue targets have been updated across various conferences and transcripts, but in this report I have prioritized the most certain parts in the 10-K, 10-Q, and official press releases.

Final Judgment

【Final Rating】 Watch

【One-Sentence Investment Thesis】 Arista is an excellent business worth tracking over the long run, even one to aspire to own, but buying at the current price leaves an insufficient margin of safety for a conservative value investor.

【Core Bull Case】 Arista has the scarce combination of "high growth, high margins, high cash generation, and zero interest-bearing debt." Software capabilities such as EOS, CloudVision, and NetDL mean it is not simply a hardware company selling metal boxes. In Ethernet data center switching, the company remains in an industry-leading position, benefiting from the expansion of AI infrastructure. Low capex and high net cash make its growth quality clearly better than that of most infrastructure companies. A strong corporate culture and execution have, over history, long converted scale growth into profit and cash.

【Core Bear Case】 The current valuation is high and has prepaid a great deal of future high growth. Concentration in the two largest customers is too high, and a change in a single customer's orders is enough to disturb the statements. The threat from competitors such as NVIDIA in AI networking is rising. Cash flow is very strong, but a substantial part comes from deferred revenue and timing factors and cannot simply be extrapolated linearly. Governance quality is good but not perfect, and pay alignment still needs continued observation.

【Key Assumptions】 Arista can keep its lead in cloud and AI Ethernet networking. The top two customers will not suffer a long-lasting, material loss. The GAAP gross margin center can hold at roughly 60%+. AI-related networking demand will not go through a severe bubble burst in the near term. Management will not make aggressive, off-core acquisitions using high-valuation stock or cash.

【Fair Buy Price】 90 to 115 dollars. The basis is the conservative owner earnings discount, the fair value range under the neutral scenario, and retaining a margin of safety of at least 25% to 30%.

【Target Holding Period】 More than 10 years. For this kind of company, compounding value comes mainly from long-term share, software embedding, and retained earnings growth, not from one or two years of valuation swings.

【Expected Annualized Return】 Conservative scenario: 0% to 3%; Neutral scenario: 4% to 8%; Optimistic scenario: 8% to 12%. These returns are inferred from the current price, the conservative owner earnings basis, and an assumption of future valuation reversion; they are not a promise of returns.

【Maximum Loss Risk】 If AI networking capex cools, core customers shift orders to competitors, margins move down, and the valuation multiple reverts from high-growth-stock levels to mature-networking-equipment-stock levels, a permanent capital loss of around 50% or even deeper from the current level is not impossible. This is not a balance-sheet risk but the risk of "a high-quality business bought at a high valuation."

【Tracking Metrics】 Changes in the share of revenue from Cloud and AI Titans. The combined share of revenue from the top two customers. The center of GAAP gross margin and operating margin. Changes in deferred revenue and remaining performance obligations. Inventory, purchase commitments, and inventory write-downs. Dependence on Broadcom supply and alternative paths. The net share-reduction effect of buybacks relative to SBC. VeloCloud integration and the share of software and service revenue. Public customer progress for NVIDIA / Cisco / HPE in AI Ethernet. The share of deferred revenue's contribution to full-year operating cash flow.

【Signals That Trigger Re-Evaluation】 An abnormal decline in the revenue share and order cadence of a single major customer. Gross margin clearly below 60% for several consecutive quarters. AI back-end networking customers publicly shifting to NVIDIA or other solutions and forming a trend. Purchase commitments and inventory continuing to rise sharply while revenue growth clearly slows. The company making an off-core, highly priced acquisition with unclear synergies. Management beginning to use high-valuation buybacks to mask SBC dilution rather than genuinely accreting per-share value.

【Final Recommendation】 Soberly put, Arista deserves respect, but at present it does not deserve impulse. If you already hold it at a low cost, I would not argue for selling mechanically just because it "looks expensive," because the business fundamentals are genuinely excellent. But if you are a new buyer who holds to "long-term value plus a conservative margin of safety," I would choose to keep watching, waiting for a better price or for results to keep delivering so the valuation digests itself naturally. The hardest thing with a high-quality company is restraint; and for ANET today, I believe restraint is worth more than excitement.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.