ANET: The $11.4 Billion Supply Bet Behind the AI Surge

The 60-second read

Five things to know

1. Growth accelerated. Q3 revenue rose 58.5% and beat the company's $3.3B guide by 10.8%.
2. Supply risk expanded. Non-cancellable purchases reached $11.4B versus $6.8B at year-end, with inventory also rising.
3. EPS needs a bridge. A $134.5M strategic-investment gain helps explain why GAAP EPS exceeds adjusted EPS.
4. Cash has timing support. Deferred revenue provided $1.467B of nine-month cash; that contribution cannot be assumed perpetual.
5. Value remains below price. Three explicit routes support $150 current fair value versus the $185.02 reference close.

Three scenarios, one probability-weighted number

Illustrative November 2027 price values: operating DCF value rolls forward one year; existing cash earns zero in this conservative bridge.

Signposts: what would change our mind

Thresholds are analyst monitoring rules, not company guidance. Next results date is not issuer-confirmed; use the actual filed report.

The AI surge now has an $11.4 billion supply commitment

Arista has converted accelerating artificial-intelligence demand into unusually strong sales and operating profit. The less visible development is the amount of supply it has committed to purchase. Non-cancellable commitments reached $11.4 billion at September 30, compared with $6.8 billion at December 31. Inventory also rose sharply. Securing supply can protect growth when components are scarce, but it makes the balance sheet more exposed to a change in customer ordering. That tension is the focus of this quarter's analysis.

Our rating is HOLD, conviction 3/5, with high uncertainty and a central current fair value of $150 per share. The reference price is the November 10, 2026 close of $185.02. Three valuation routes produce approximately $145, $158 and $151, rounded into a $150 anchor and a practical central band of $140–$160. The scenarios are much wider because growth, cash conversion and terminal economics interact. A great operating result does not automatically create a margin of safety at the quoted price.

The release arrived after the November 3 regular session. The stock closed at $169.91 before those results and $178.48 at the next close, a computed increase of 5.04%. It subsequently reached $185.02 by November 10, 8.89% above the pre-release close. These are dated observations, not proof that the earnings release caused every subsequent move. The intervening path was volatile, including a $159.16 close on November 5. A single post-release return would conceal that uncertainty.

The earlier Q2 episode used the angle that extraordinary growth still required an aggressive long-term revenue outcome to justify the price. This packet updates the growth base and focuses on supply commitments, conversion and accounting quality. It does not repeat the old valuation unchanged after a substantially stronger print. Nor does it imply the previous value was a promise about the next quarter's stock price. The current model is a new, explicit estimate built from the September financial statements.

A large beat with another sequential increase guided

Revenue was $3.6576 billion, rising 58.5% from the year-earlier quarter and 20.5% from the preceding quarter. That exceeded the company's $3.3 billion guidance by 10.8%. FMP's saved revenue estimate is $3.334747 billion, implying a 9.7% revenue surprise. The comparison with management's own guidance is particularly useful because both values refer directly to the same revenue definition and fiscal period.

GAAP operating income reached $1.6424 billion and the operating margin was 44.9%, compared with 42.4% a year earlier. Adjusted operating income was $1.7958 billion, a 49.1% margin. Gross margin declined year over year, so the operating improvement came from stronger expense leverage rather than an improvement in every part of the income statement. This distinction matters when projecting what can persist at a much larger revenue base.

GAAP diluted EPS was $1.22, above adjusted EPS of $1.20. That unusual ordering is explained by a strategic-investment gain and tax adjustments, discussed in the profit bridge below. The saved provider reports actual EPS of $1.20 and expected EPS of $1.08. Its actual value matches the issuer's adjusted result, but the estimate's detailed accounting definition is not independently established; we therefore avoid presenting the percentage difference as a fully reconciled like-for-like earnings surprise.

For Q4, management guides revenue of $3.9–$4.0 billion, adjusted operating margin of 48–49%, and adjusted diluted EPS of $1.26–$1.30. The revenue midpoint implies another 8.0% sequential increase. Adding that midpoint to nine-month revenue produces a computed FY2026 revenue base of $13.3523 billion. This is a bridge from reported results and quarterly guidance, not a separately issued full-year revenue target. The forward adjusted measures cannot be reconciled to GAAP without unreasonable effort, according to the company, and should remain labeled adjusted throughout.

One operating system, one segment, two revenue streams

Arista reports one operating segment. Its product and service categories reveal different economics, but the company does not disclose independent operating profit for each. We calculate category gross profit from the explicit revenue and cost-of-revenue lines; allocating research, sales or corporate overhead to invent segment operating margins would overstate the available evidence.

Product revenue was $3.190 billion, up 66.9%. Service revenue was $467.6 million, up 17.9%. Product represented 87.2% of total revenue. Computed product gross margin was 60.7%, while service gross margin was 83.6%. The high service margin supports the attraction of support and software relationships, but service is a smaller share of the rapidly growing total. The combined gross margin can therefore move with customer and product mix even if individual contracts remain economically attractive.

The platform spans data-center, AI, campus and routing applications through the EOS operating stack. New rack-scale architectures and scale-across networking address important technical opportunities. The release describes capabilities and partnerships; it does not provide a separate audited AI revenue figure or a standalone profitability forecast for each new platform. We do not convert a list of partners into committed revenue. Customer qualification cycles, competing architectures and deployment schedules determine whether announced capabilities become durable sales.

Concentration remains central. The current 10-Q says two customers represented 16% and 26% of total revenue in 2025; those are annual figures, not Q3 customer shares. Large customers can create exceptional volume and also demand pricing concessions, change deployment timing or shift spending between AI and other infrastructure. The filing specifically describes variability from qualification and spending cycles. Broad product ambitions are therefore not the same as diversified realized revenue. The signpost is growth that continues while margins and cash conversion hold, not the number of new platforms named in a press release.

Why reported EPS exceeds adjusted EPS

The operating bridge is straightforward. GAAP operating profit of $1.6424 billion plus stock compensation of $138.1 million and acquired-intangible amortization of $15.3 million produces adjusted operating profit of $1.7958 billion. Stock compensation is an economic expense even though the company removes it from its adjusted performance measure. The model uses GAAP economics and does not assume that every adjustment disappears as revenue grows.

Below operating profit, interest income of $137.3 million and other income of $135.8 million produced total other income of $273.1 million. Pretax income was $1.9155 billion; tax expense of $355.1 million left net income of $1.5604 billion. The effective tax rate was 18.5%. These figures explain why net earnings grew faster than operating profit, without requiring an unsupported claim that all of the improvement came from sales execution.

The adjusted-net-income bridge adds stock compensation and amortization, then removes a $134.5 million gain on strategic investments, $40.0 million of tax benefits on stock awards, and $8.5 million of tax effects on exclusions. The result is $1.5308 billion of adjusted net income. Rounded per-share amounts are $1.22 GAAP and $1.20 adjusted. Removing a gain can outweigh the addition of expenses; a lower adjusted EPS is not an arithmetic error.

Prior-period adjusted figures were recast for a change in the treatment of tax effects associated with stock awards, effective in Q4 2025. The current release says that the change increased comparative Q3 2025 adjusted net income by $17.1 million and adjusted EPS by two cents. We use the current release's recast comparator. A chart assembled from older unrecast releases without adjustment could manufacture a misleading growth rate. The GAAP history is less affected by that presentation issue, while the full raw releases remain available for an audit.

Cash is strong, but growth consumes more working capital

Nine-month operating cash flow was $3.9774 billion against net income of $3.7962 billion, or 1.05 times accounting profit. Subtracting $184.9 million of cash capital expenditure produces a simple free-cash-flow proxy of $3.7925 billion. This proxy is our calculation, not an additional company-defined adjusted metric. It excludes acquisitions and purchases of marketable securities, which have different analytical meanings from recurring operating investment.

Deferred revenue provided $1.4671 billion of cash in the nine-month cash-flow bridge. Inventory consumed $962.8 million, receivables consumed $440.9 million and other assets consumed $604.2 million. Customer financing is valuable, but its growth should not be projected as a perpetual extra margin. As a diagnostic, removing the deferred-revenue cash contribution from the simple FCF proxy leaves $2.3254 billion. That is not normalized free cash flow: it changes one component while leaving other working-capital movements intact. It shows how sensitive the headline is to timing.

Subtracting nine-month stock compensation of $379.4 million from the simple FCF proxy gives $3.4131 billion. This is another diagnostic rather than an instruction to subtract SBC twice in a valuation already based on GAAP earnings. Compensation ultimately affects owners through dilution or the cash required to offset it. The diluted weighted-average share count was 1,278.2 million, compared with 1,276.6 million a year earlier. There were no share repurchases during the first nine months of 2026.

End-balance receivable days are approximately 58.6 and inventory days approximately 221.7, using ninety-two days and current-quarter revenue or total cost of revenue. These are rough proxies, not the company's reported collection cycle or inventory-turnover methodology. Inventory includes strategic stock for future demand; rapid growth and timing can distort the ratio. The correct response is to monitor conversion and excess-inventory provisions alongside growth. Current disclosure controls were assessed effective and the filing reports no material quarter change in internal controls; this supports reporting reliability without eliminating business risk.

Liquidity protects the company; commitments change the risk

Cash and equivalents of $2.2881 billion plus marketable securities of $12.0709 billion total $14.359 billion. The balance sheet has no financial borrowing line. On the valuation's 1,278.2 million diluted-share denominator, cash and securities equal $11.23 per share. The market-cap figure reported by the quote provider uses a different, more current basic-share count; we therefore identify the valuation denominator explicitly rather than silently mixing it with a weighted-average count.

This liquidity is substantial, but the company also carries $6.8395 billion of deferred revenue and $11.4 billion of non-cancellable purchase commitments outside the balance sheet. Of the latter, $9.3 billion has expected receipt dates within twelve months and $2.1 billion later. Supplier deposits were $152.8 million, up from $53.0 million at year-end. The filing links elevated commitments to new-product ramps, AI-network demand and tighter memory and silicon supplies. That is a business explanation, not assurance that every committed component will be sold at the intended margin.

Purchase commitments are not financial debt and should not simply be subtracted from equity value as though they were an additional loan with no associated inventory. Normal purchases are already reflected in projected operating costs and cash conversion. Conversely, calling all cash excess would be too casual when inventory and deposits need funding. Our cash-margin assumptions explicitly allow conversion below GAAP after-tax operating margins, while the bear case tests weaker demand and lower margins. A sudden obsolete-component charge would require a separate update.

Remaining performance obligations total $8.6 billion, with about 91% expected to become revenue within two years. RPO and purchase commitments have different definitions, timing, margins and cancellation features. Netting one against the other would create a misleading coverage ratio. Deferred revenue also includes service and support obligations. We add liquid assets in the model while retaining fulfillment costs in operating forecasts; subtracting the full deferred balance again would double count much of that obligation.

Three routes and the growth already embedded in the quote

Our primary route is a ten-year owner-cash DCF. The starting FY2026 revenue base is $13.3523 billion. Annual growth then fades from 28%, 25% and 22% to 20%, 18%, 15%, 13%, 11%, 9% and 7%. Revenue reaches approximately $62.1 billion in 2036. This is already a demanding expansion, not a forecast that the present growth rate collapses immediately. The owner-cash margin begins at 32.5% and declines to 30.7%, after stock compensation, taxes and normalized reinvestment. Those margins are assumptions rather than company guidance.

We discount the annual cash flows at 10.5%, apply 3.5% terminal growth, add $14.359 billion of cash and securities once, and divide by 1,278.2 million diluted shares. The result is $145.09 per share. Interest earned on the cash pile is excluded from the operating cash forecast, so adding cash does not also capitalize the same interest stream. The model does not add strategic investment gains as recurring operating profit. It uses constant shares rather than giving free credit for future repurchases.

Route two applies 25 times assumed FY2027 GAAP operating income to the same revenue base growing 28%, at a 44% operating margin, then adds liquid assets. It yields approximately $158 per share. This is an enterprise-value multiple cross-check, with lease costs retained in the operating economics. Route three assumes a 35% after-tax net operating margin for FY2027, adds annualized after-tax interest income, divides by diluted shares and applies 30 times earnings. It yields approximately $151. Cash is not added again in this EPS route because its interest income is already capitalized. Multiples and margin choices are explicit analyst assumptions.

At $185.02, diluted equity value in our model is $236.493 billion; subtracting liquid assets gives $222.134 billion of operating enterprise value. Under the base discount rate and unchanged growth pattern, the entire owner-cash stream would need to be roughly 30% larger to justify that quote. Equivalently the opening cash margin would need to be above 42% with the same proportional path. This is a sensitivity, not a prediction or a unique reverse-DCF solution: stronger growth, lower required return or better terminal economics can also support the price.

We round the three routes to a $150 central anchor, with a $140–$160 central band. Our separate illustrative November 2027 price scenarios are bear $85, base $160 and bull $305, weighted 30%, 50% and 20%, producing $166.50. We roll each operating DCF value forward one year at its required return, retain the interim owner cash and keep existing liquid assets at their current value. This assumes no distributions, repurchases or interest earned on the initial cash pile. The rounded present DCF anchors remain $75, $145 and $280, so future scenario targets must not be mistaken for current fair value. High uncertainty calls for a meaningful margin of safety; $120 is an entry-watch level at 20% below the central anchor. A price reaching that level would still require a current check that the operating thesis survived.

A more cautious value than the target consensus

The saved FMP target-consensus snapshot is $242.43, with a $164 low and $305 high. The ratings snapshot contains 41 Buys, 12 Holds and no Sells. These are provider aggregates and not a vote that any particular valuation model is correct. Aggregates can mix different update dates, horizons and accounting assumptions; the apparent precision of an average does not remove those differences.

Dated post-release target records include Truist at $240, Morgan Stanley at $230, UBS at $291 and Melius at $305, all recorded November 4. We identify those records as provider-distributed analyst news, not as original research reports we have read in full. They show that optimism persisted after the release. They do not tell us precisely how each analyst handles purchase commitments, cash conversion or terminal margins. A target-price comparison should not imply we reconstructed proprietary models from a headline.

Our $150 current-value anchor is therefore materially more cautious than the $242.43 consensus target. Some difference can reflect horizon: a twelve-month target may roll forward earnings while our DCF is current value. Much of the difference also reflects the premium paid for durable growth. We explicitly model revenue reaching roughly $62 billion in a decade and still require a better entry price. That makes the disagreement about the price of growth and its cash conversion, not a claim that Arista's products lack demand.

The saved annual consensus dataset includes revenue and earnings paths, analyst counts and a broad range. We retain those inputs so later revisions can be measured, but do not relabel the vendor's earnings estimates as GAAP or adjusted without a confirmed basis. The next useful comparison is whether estimates rise because shipment volumes and margins improve together, or whether higher sales are offset by more capital tied up in components. The latter can produce a strong revenue story without an equally strong change in owner value.

Execution earns credit; capital commitments need follow-through

Management deserves credit for exceeding its own revenue outlook while sustaining very high operating profitability. The $3.3 billion Q3 guide was surpassed by $357.6 million. The next quarter's revenue midpoint is higher again. The proper score is strong recent execution, not an invented multi-year percentage of quarters beaten. A single observed beat does not establish that guidance will always be conservative.

Research and development expense increased to $486.9 million from $326.0 million. Sales and marketing rose only modestly to $156.6 million, while general and administrative expense was $39.8 million. The spending mix supports an engineering-led business interpretation. It also means that new products must earn a return on a growing development budget. We do not assume higher research spending automatically creates an enduring technological lead; the test is customer adoption at attractive margins.

Capital allocation is less aggressive in repurchases than the cash balance might suggest. The company bought no shares in the first nine months of 2026, with $817.9 million remaining under the authorization at quarter-end. Preserving cash can be rational when supply commitments and operating investment are rising. It also avoids treating a repurchase authorization as a guaranteed source of per-share accretion. Investors should distinguish the authority to buy from actual completed purchases.

The five questions we would ask are operationally specific. How much of the purchase-commitment increase relates to customer-backed deployments versus general supply protection? What portion can be rescheduled if delivery timing changes? How do extended customer payment terms affect cash conversion? What margin should investors expect as new rack-scale products ramp? And what evidence would cause management to reduce inventory rather than keep adding supply? The public filing answers parts of those questions but does not quantify every exposure. We retain the uncertainty instead of filling the gaps with a reassuring backlog narrative.

The bull case is real and the price is demanding

The highest-ranked risk is a mismatch between supplier commitments and the timing of large customer deployments. Its likelihood is medium and potential impact high. A delay need not mean the long-term market has disappeared; it can still leave inventory, deposits and cash tied up for longer than expected. The balance sheet provides resilience, but equity value can fall before a liquidity problem emerges. A good company can deliver disappointing investment returns when expectations and timing are demanding.

Customer concentration and architecture competition are closely related risks. Large cloud customers have bargaining power and engineering capabilities of their own. Competing networking approaches, merchant-silicon dependence and qualification cycles can change mix or pricing. We rate the potential impact high without inventing a probability that a named customer will leave. The filing identifies concentration and supply-chain dependencies; it does not establish a current loss of a major customer.

The strongest bull argument is that open Ethernet networking keeps gaining relevance across AI deployments, allowing Arista to grow faster for longer while retaining its software and operational advantages. In the bull DCF, growth fades more slowly, cash margins remain higher, the required return falls to 9.5% and terminal growth is 4%. Those combined assumptions produce about $280. The bull case is possible, but it is not simply today's quarter repeated indefinitely. Each assumption must be earned.

The bear DCF slows growth more quickly, begins with a 30% owner-cash margin that fades to 27.3%, uses an 11.5% required return and 3% terminal growth. It produces about $76, rounded to $75. This case does not require the business to stop growing. It requires that growth become less valuable when cash conversion and the valuation multiple normalize. Our HOLD stance acknowledges that upside exists while declining to pay for the more favorable path without a sufficient cushion.

The next result must connect growth to cash

The next quarterly report is the main checkpoint. The provider calendar currently lists February 11, 2027, but that date is a planning lead rather than a confirmed issuer announcement. We will check official investor-relations scheduling before relying on the day. The operating tests are already defined: Q4 revenue against $3.9–$4.0 billion, adjusted operating margin against 48–49%, and adjusted EPS against $1.26–$1.30. A guide beat matters more when the cash and inventory evidence improves with it.

The year-end balance sheet should show whether the larger supply position is converting into deliveries without sacrificing margin. We will compare inventory to the current $3.2098 billion, non-cancellable commitments to $11.4 billion and cash plus securities to $14.359 billion. A lower inventory balance is not automatically good if it reflects a shortage that constrains sales; a higher balance is not automatically bad if shipments and customer demand support it. The relationships among these figures are the investment signal.

Product milestones and customer qualifications are additional catalysts, but an announcement alone is not evidence of revenue scale. We look for disclosed sales, margin and cash effects when available. The company may also provide more detail about how new platforms affect supply commitments or payment terms. Until then, the model uses a range and a normalized cash margin rather than treating every technical claim as a new revenue line.

These checkpoints make the HOLD call falsifiable. Sustained growth above our model with healthy conversion would raise fair value even if the price never visits $120. Conversely, weaker conversion or excess-inventory charges could lower value while revenue remains impressive. Price and business evidence must be reconsidered together; neither a target hit nor a share-price decline is, by itself, proof that the original operating thesis succeeded or failed.

Source map, definitions and reproducible assumptions

All current-quarter financial figures originate in the November 3 earnings release filed with the SEC and the September 30 Form 10-Q. The release tables are headed three months and nine months ended September 30, with 2026 before 2025. Income-statement dollars are in millions except per-share amounts. Keeping those headers explicit prevents a prior-year/current-year inversion and prevents a nine-month cash-flow number from being presented as a single-quarter result.

The retained data package includes nine quarters of provider income statements, balance sheets and cash flows; the matching earnings-release history; three annual cash-flow records and annual filings; daily prices; quote, targets, rating counts, earnings expectations and annual estimates. Current issuer figures control where mappings differ. In particular, provider EBIT and EBITDA include below-operating items in some rows, and its zero interest-income mapping for Q3 conflicts with the issuer's $137.3 million. We do not use those vendor fields to value operations. Historical adjusted-tax comparability is disclosed rather than silently repaired with guessed numbers.

Free cash flow here means operating cash less cash purchases of property and equipment unless a diagnostic is explicitly labeled otherwise. Owner cash in the DCF is a forward economic assumption after stock compensation, taxes and normalized reinvestment; it is not the current reported FCF ratio extrapolated unchanged. Remaining performance obligations measure future revenue under the company's contract accounting. Non-cancellable purchase commitments describe supplier obligations. Neither is equivalent to profit, and their different horizons make direct subtraction inappropriate.

Per-share values are ordinary dollars, while financial tables and cash-flow charts use millions of dollars. The DCF and scenario probabilities are our judgments, computed from saved inputs; they are not management guidance or a consensus prediction. Historical returns use dated closing prices and do not establish causality. This packet is general research with a stated uncertainty range. It does not know an individual investor's objectives, liquidity needs, taxes, leverage or concentration limits.