Arista Networks ANET

> ## ⚑ EARNINGS BANNER — REPORTED THIS MORNING

> Second-quarter 2026 results were released via 8-K/Exhibit 99.1 on 2026-08-04, the day of this dive. Revenue was $3.036 billion — the first $3 billion quarter — up 37.7% year on year, with GAAP diluted EPS of $0.95 and non-GAAP of $1.02 against a $0.89 consensus. The stock closed at its 52-week high, up 3.06%. Third-quarter guidance is approximately $3.3 billion of revenue at a 48-49% non-GAAP operating margin. THE VENDOR PAYLOAD DOES NOT CONTAIN THIS QUARTER — inc_q ends at 2026-03-31 and earn_cal records the actual as null — so every June-quarter figure in this dive comes from the 8-K itself, and every consensus estimate in the file predates the print.

Reference table

1. The quarter, reported this morning

From the 8-K/Exhibit 99.1 released 2026-08-04:

The single most important line is the margin, and it needs to be read against the rest of this batch. Arista's operating margin EXPANDED — 70 basis points on GAAP and 110 on non-GAAP — while revenue grew 37.7%. Dell Technologies, growing revenue 87.5% into the same artificial-intelligence build, saw gross margin fall 340 basis points because it assembles from bought-in components. The divergence is the whole quality argument: Arista is selling something it prices, not something it assembles.

Management framing, half-weighted per house rule but quoted because both statements are checkable against the numbers above:

> "As we deliver our first $3 billion quarter in Q2 2026, it is clear that our Arista 2.0 platform strategy is compelling. Customers see networking as the central nervous system for infrastructure from the client to campus to data and AI centers." — Jayshree Ullal, Chairperson and Chief Executive

> "Our second quarter 2026 reflects strong, broad-based growth, with revenue up 37.7% and EPS up 39.7%." — Chantelle Breithaupt, Chief Financial Officer

Guidance, issued in the same release:

> "For the third quarter of 2026, we expect: Revenue of approximately $3.3 billion; Non-GAAP operating margin of 48 - 49%."

$3.3 billion is an 8.7% sequential step on a quarter that already grew 12.1% sequentially, at a margin only 90-190 basis points below the one just delivered.

The product disclosure in the same release is the substance behind the margin, and we quote it because it is the only description of the competitive position available in this entire file:

> "Arista introduced the 7060XE7 Series, a next-generation 1.6-terabit portfolio for AI fabrics — The new Etherlink platforms deliver up to 100 Tbps of system bandwidth and support Linear Pluggable Optics, cutting interconnect power consumption by approximately 60% compared to traditional pluggable optics."

> "Arista detailed new innovations for scale-up, scale-out, and scale-across AI fabrics — The company outlined techniques including multi-planar leaf-spine designs, the open Multipath Reliable Connection (MRC) protocol, and segment routing over IPv6 (SRv6), aimed at maximizing AI cluster utilization and resilience."

Power and cluster utilisation are the two binding constraints on artificial-intelligence infrastructure economics, and this company is selling directly into both. A 60% reduction in interconnect power consumption is a quantified, checkable claim about the value proposition, and it is the kind of thing that supports a 49.9% operating margin.

The release also records recognition as a Leader in the 2026 Gartner Magic Quadrant for Enterprise Wired and Wireless LAN, which points at a materially larger and more fragmented customer base than the hyperscalers — the enterprise campus. That is optionality this file cannot size.

2. The record — six beats and an accelerating rate

Six consecutive beats, with the largest of them the most recent, and a year-on-year growth rate that has accelerated for three quarters running. Sequential growth has also accelerated in each of the last three quarters: 4.7%, 7.8%, 8.9%, 12.1%.

Annual context — five years of compounding at high and rising margins:

Revenue compounded 31.2% a year for five years while net margin rose from 27.4% to 39.0%. That combination — accelerating scale with expanding margin — is what separates a franchise from a cycle, and it is the strongest single argument for owning this business at some price.

Revenue mix, from seg_prod, which ties exactly: FY2025 Product $7.577B (84.1%) and Service $1.429B (15.9%), summing to $9.006 billion. Service grew 27.7% and Product 28.8% — the recurring share is holding rather than growing, which is a mild negative for revenue quality.

Geography, from seg_geo, which also ties exactly: FY2025 Americas $7.122B (79.1%), EMEA $1.070B (11.9%), Asia Pacific $813M (9.0%). A 79.1% single-region concentration is the highest in this batch and it is the structural counterpart of a customer base concentrated in large United States technology buyers. This file does not disclose the customer mix and that absence is a genuine gap in the analysis.

3. Balance sheet — zero debt, and the largest net-cash error in this batch

From bal_a at 2025-12-31:

Correction 1 — netDebt omits $8.779 billion of short-term investments. This is the largest omission of that class in this batch. The vendor's figure of −$1.964 billion nets zero debt against cash and equivalents only. The correct net cash position is $10.743 billion — 5.5 times the vendor's figure and 4.5% of market capitalisation.

Correction 2 — the enterprise value inherits it. enterpriseValueTTM of $237.145 billion is market capitalisation less $2.790 billion of net cash. Corrected: $239.935B − $10.743B = $229.19 billion, which is $7.95 billion, or 3.5%, lower.

Correction 3 — every debt and coverage ratio is UNDEFINED, not zero. debtToEquityRatioTTM, debtToAssetsRatioTTM, debtToCapitalRatioTTM, interestCoverageRatioTTM, debtServiceCoverageRatioTTM and operatingCashFlowCoverageRatioTTM all read 0. A zero interest-coverage ratio on a company with no interest expense is an undefined ratio, not poor coverage — and the vendor's composite rating assigns debtToEquityScore of 1 out of 5 on the strength of it. That sub-score is rejected: a company with zero debt is not a leverage risk.

The genuine balance-sheet flag is inventory. daysOfInventoryOutstandingTTM reads 245.4 days — the highest of any name in this batch — on a $2.247 billion balance that rose 22.5% year on year, inside a 255.0-day cash conversion cycle. For a company whose products are configured to order and whose customers place large, lumpy orders, holding eight months of inventory is a deliberate supply-chain choice and it is also $2.2 billion of capital committed against demand nobody controls.

4. Cash flow and capital returns

A 47-52% free-cash-flow margin is extraordinary and it is the direct consequence of a business model with essentially no capital intensity: capital expenditure of $119.5 million on $9.006 billion of revenue is 1.3%. capexToRevenueTTM reads 1.50%. Arista converts roughly half of every revenue dollar into free cash.

Trailing free cash flow of approximately $5.278 billion against a market capitalisation of $239.93 billion is a 2.20% free-cash-flow yield. Thin — but it grew 28% in the last fiscal year and the corrected net-cash balance of $10.743 billion is already 4.5% of market capitalisation, growing at roughly $2.6 billion a year net of buybacks.

Capital returns are modest and there is no dividend. FY2025 repurchases of $1.603 billion are 0.67% of market capitalisation, up from $423.6 million in FY2024. Total shareholder yield is therefore approximately 0.67%, all buyback, and cash is accumulating faster than it is being returned. Stock-based compensation of $439.2 million is 4.9% of revenue and diluted shares have been essentially flat at 1,271-1,283 million across three years, so the buyback is roughly offsetting dilution rather than shrinking the count.

5. Valuation — priced in or room?

At $190.55 (market cap $239.93B, ~1,259.2M shares, corrected enterprise value $229.19B):

Estimate coverage is good on FY2026 and FY2027 (16-21 analysts) and thin on FY2028 EPS (SIX) and FY2029 EPS (FOUR). Neither of the latter informs any conclusion.

The consensus in this file is stale by hours and demonstrably too low. First-half revenue was $5.745 billion ($2.709B + $3.036B) and the third quarter is guided to approximately $3.3 billion. Against a FY2026 consensus of $11.617 billion, that leaves $2.572 billion implied for the fourth quarter — a 22% sequential DECLINE from the guided third quarter, on a business that has grown sequentially in every quarter in this file. That is not a forecast; it is an estimate that has not been updated. Our own FY2026 revenue estimate is approximately $12.5 billion and our FY2027 estimate approximately $16.0 billion, both labelled as ours.

Per the data contract we tested est.ebitAvg and est.ebitdaAvg. The arithmetic screens pass — ebitAvg below ebitdaAvg in every year, no negatives — but the gap between them is implausibly small: FY2026 ebitdaAvg of $7.300 billion against ebitAvg of $7.289 billion implies depreciation and amortisation of $11 million, against an actual FY2025 figure of $72.6 million. The EBITDA rows understate depreciation by roughly 85% and are used for nothing.

A separate and more serious estimate defect: the FY2023 row is corrupt. It reports epsAvg of $7.888 and netIncomeAvg of $10.199 billion against actual FY2023 diluted EPS of $1.65 and net income of $2.087 billion — overstatements of 4.8x and 4.9x respectively. The FY2022 row is correct ($1.096 estimated against $1.07 actual), so this is not a systematic split-adjustment failure; it is a single corrupt row. It is discarded entirely and no conclusion touches it.

Peer context. The vendor peer set is a market-capitalisation grouping rather than a comparable set — Applied Materials, Amphenol, AppLovin, Intuit, Lam Research and ServiceNow are not networking companies. No genuine networking comparable is supplied, and no peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The relevant observation within this batch is that Arista's $239.93 billion market capitalisation is 77% of Dell Technologies' $310.32 billion on 7.9% of Dell's revenue — the market is paying roughly ten times as much per dollar of sales for the switch vendor as for the server assembler, and the 49.9% versus 8.3% operating-margin difference is why.

5a. What today's price assumes (the inversion)

At $190.55 — 42.7x FY2027 consensus and 21.7x trailing sales — the price embeds:

• Revenue reaches at least $14.4 billion in FY2027 and $17.4 billion in FY2028. (Consensus; 21 and 19 analysts, both struck before this morning's print.) On our own arithmetic the FY2027 figure is nearer $16 billion, so this bullet is the one we think is most likely to be beaten. The nearest check is the guided third quarter of approximately $3.3 billion.
• Non-GAAP operating margin holds near 48-50%. (Company guidance for the third quarter.) This is by far the most fragile assumption in the price, because it is the entire difference between Arista and a commodity switch vendor. Every other artificial-intelligence hardware name in this batch has seen margin compress as the mix shifted; Arista's has expanded. If that reverses, the multiple has no support.
• The market keeps paying more than 35x forward earnings. (Our number.) At 25x FY2027 consensus the stock is $112; at 50x it is $223. The multiple is worth $111 — 58% of the current price — on unchanged estimates.
• The customer concentration does not bite. (Not disclosed in this file; our inference.) 79.1% of revenue is in the Americas and the buyers of large-scale data-centre fabric are a small number of very large organisations whose capital-expenditure decisions are annual and revisable. This file contains no customer-concentration disclosure and its absence is a gap we flag rather than fill.
• The $2.247 billion inventory balance converts to revenue rather than to charges. (Filing-confirmed balance; our assumption.) 245.4 days is the highest in this batch and rose 22.5% year on year.

5b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+22.4%, from FY2027E $4.468 to FY2028E $5.469, and materially more on our own estimates) + multiple drift (COMPRESSION, from 42.7x to roughly 36x, −16%) + shareholder yield (+0.67%) ≈ +5% to +8%.

Our base of $195 assumes substantial multiple compression offset by earnings that we think will come in above consensus. The compression is assumed for the ordinary reason: consensus itself has revenue growth decelerating from 29% (FY2026) to 23.9% (FY2027) to 21.2% (FY2028), and a multiple appropriate to 38% growth is not appropriate to 21%.

Sensitivity, stated openly, on the FY2027 consensus EPS of $4.468: at 25x the shares are $112; at 35x, $156; at 43.6x, $195; at 50x, $223; at 59x, $265. On our own FY2027 estimate of roughly $16 billion of revenue at a 39% net margin — approximately $4.90 of GAAP EPS — the same multiples give $123, $172, $214, $245 and $289.

The bull case at $265 requires no re-rating on our own numbers — only that revenue lands where we think it will and the market keeps paying roughly today's multiple. The bear case at $120 requires no growth failure at all — only a return to 25-27x forward earnings, which is where high-quality networking hardware has historically traded.

5c. Variant perception (where we differ, what would surprise)

• We differ from the consensus estimates in this file and we think they are simply stale. FY2026 revenue of $11.617 billion implies a fourth quarter of $2.572 billion against a guided third quarter of approximately $3.3 billion — a 22% sequential decline that nobody is forecasting. Our FY2026 estimate is approximately $12.5 billion and our FY2027 approximately $16.0 billion, both labelled as ours. Watchable event: consensus revisions over the next two weeks. If FY2027 consensus does not move above $15 billion, the market is disagreeing with the guide.
• We think the street's price targets are also stale and this is unusually easy to demonstrate. The consensus target of $186.67 sits 2.0% BELOW the closing price, with a median of $187.50 and a high of $200 — on a distribution of 39 buy ratings, 13 holds and ZERO sells. A universally positive rating set with an average target below the price is not a bearish signal; it is an un-updated one, and a wave of increases is close to mechanical.
• We think the margin divergence within this batch is the most important cross-name observation available and it is entirely quantitative. Arista's operating margin EXPANDED 70-110 basis points while revenue grew 37.7%. Dell Technologies' gross margin FELL 340 basis points while revenue grew 87.5%. Both sell into the same build. The difference is that one prices a product and the other assembles components, and it is the entire justification for a tenfold difference in price-to-sales. Watchable number: the non-GAAP operating margin in the next release against the guided 48-49%. Any print below 47% attacks the premium directly.
• We have no qualitative check on any of this and we state it as a limitation, not a footnote. The knowledge base contains zero claims about Arista Networks. For every other name in this batch we could test our reading against independent expert claims — and on several, that testing changed the conclusion. Here there is nothing. The dive rests entirely on one 8-K, one 10-K, two 10-Qs and a vendor payload that is missing the most recent quarter.
• Positive surprise that would force a re-rate: third-quarter revenue above the guided $3.3 billion with non-GAAP operating margin at or above 49%; the enterprise campus business — flagged in this morning's release via the Gartner recognition — being sized separately for the first time; or international revenue growing materially faster than the Americas, which would reduce the concentration discount.
• Negative surprise that would break the thesis: non-GAAP operating margin below 47%; any sequential revenue decline; an inventory charge against the $2.247 billion balance at 245 days; disclosure of customer concentration above roughly 20% for any single buyer combined with an order pause; or a competitive announcement that puts the AI-fabric position at risk. At 60x trailing earnings on a 52-week-high close, none of these produces a proportionate decline.

Synthos fair values

All three anchors are multiples of the FY2027 consensus GAAP diluted EPS of $4.468 (16 analysts), cross-checked against our own FY2027 estimate of roughly $4.90 and against enterprise value to sales.

• Bear ~$120 — 26.9x FY2027E consensus, or 24.5x our own FY2027 estimate. Cross-checks: 37.8x trailing GAAP; 12.9x FY2027E revenue on rebuilt enterprise value; 18% below the 200-day average of $146.31. The scenario: the non-GAAP operating margin compresses below 45% as the artificial-intelligence fabric mix grows, a large customer pauses, the $2.247 billion inventory balance produces a charge, and the multiple returns to where high-quality networking hardware has historically traded. −37.0%. Note this requires no revenue miss against consensus — only margin compression and a normal multiple.
• Base ~$195 — 43.6x FY2027E consensus, or 39.8x our own estimate. Cross-checks: 61.5x trailing GAAP; 56.4x trailing non-GAAP; 15.3x our FY2027 revenue estimate on rebuilt enterprise value. Sensitivity: 35x consensus gives $156, 50x gives $223, and the street's $186.67 implies 41.8x. The scenario: the guide holds, FY2027 revenue beats the current consensus by roughly 11%, non-GAAP operating margin stays in the high forties, and the multiple compresses modestly as growth decelerates from 38% toward 21%. +2.3%, plus 0.67% of shareholder yield.
• Bull ~$265 — 59.3x FY2027E consensus, or 54.1x our own estimate — i.e. essentially today's multiple, HELD. This bull case requires no re-rating whatsoever, only that revenue lands where we think it will. The scenario: the sequential acceleration continues, the enterprise campus opportunity begins to contribute visibly, non-GAAP operating margin holds at 49-50%, and consensus is revised up through the year. +39.1%.

Base is 2.3% above spot; asymmetry roughly 1.06:1 (37.0% down, 39.1% up). That is barely above parity and it is the direct output of a 52-week-high close. What makes this a Watch rather than a Buy is that the base case is spot, the entry is the highest price of the year, and there is no independent qualitative corroboration of any kind — the knowledge base is empty.

6. Knowledge base — zero claims

Raw entity hits: 0. Text matches: 0. Discarded: 0.

The Synthos knowledge base contains no claims naming Arista Networks. Searches covered ANET, Arista and Arista Networks against the entity field of all 51,928 distilled claims, plus a text sweep on the token arista across claim text and categories. Nothing returned.

This is the only name in this batch of which that is true, and it is worth setting against the others: Palantir has 83 entity claims, Chevron 26, Merck 24, Goldman Sachs 28, Dell 29. Arista has none — despite being a $239.93 billion company at the centre of the artificial-intelligence infrastructure build that dominates every one of those other lanes.

Two readings, and we do not choose between them. The first is that networking is simply not a subject the sources this knowledge base ingests discuss — the discourse concentrates on accelerator silicon, model developers, power and, latterly, servers, and the fabric between them is invisible to it. The second is that the coverage universe has a gap. Either way, the practical consequence for this dive is the same and it is significant: there is no independent qualitative check on the competitive position, the customer concentration, the durability of a 49.9% operating margin, or the competitive threat from accelerator vendors moving into the fabric.

On several names in this batch, the knowledge base changed the conclusion — it supplied the valuation framework on Goldman Sachs, the bear precedent on Dell, the biologic-erosion qualification on Merck, and the only structural bear argument on Texas Instruments. Here it supplies nothing, and the conviction contribution is exactly zero. We report this as a finding and we let it weigh on the verdict: a name with no independent corroboration, trading at 60x trailing earnings at its 52-week high, is a Watch rather than a Buy on that ground alone.

7. Data integrity — what we rejected from the vendor file and why

Eight findings, and the first is the most consequential thing a reader can know about this file.

1. The vendor payload does NOT contain the June 2026 quarter — the quarter reported this morning. inc_q ends at 2026-03-31, and the earn_cal entry for 2026-08-04 records epsActual: null and revenueActual: null. Every June-quarter figure in this dive — revenue of $3.036 billion, GAAP and non-GAAP operating margins of 45.4% and 49.9%, GAAP and non-GAAP diluted EPS of $0.95 and $1.02, and the third-quarter guidance — comes from the 8-K/Exhibit 99.1 filed 2026-08-04. Consequently every trailing metric in the payload — priceToEarningsRatioTTM of 64.4x, revenuePerShareTTM, netIncomePerShareTTM, freeCashFlowYieldTTM — is struck on a window ending 2026-03-31 and is NOT trailing twelve months to the most recent quarter. We recompute all of them.

2. netDebt omits $8.779 billion of short-term investments — the largest omission of that class in this batch. The vendor reports −$1.964 billion, netting zero debt against cash and equivalents only. The correct net cash position is $10.743 billion — 5.5 times larger. Consequential correction: enterpriseValueTTM from $237.145 billion to $229.19 billion, and evToSalesTTM from 24.42x (on a stale trailing revenue base) to 21.74x on the correct one.

3. Six debt and coverage ratios read 0 and are UNDEFINED, not zero — rejected. debtToEquityRatioTTM, debtToAssetsRatioTTM, debtToCapitalRatioTTM, longTermDebtToCapitalRatioTTM, interestCoverageRatioTTM and debtServiceCoverageRatioTTM all read 0 because Arista has no debt and no interest expense. The vendor's composite rating assigns debtToEquityScore of 1 out of 5 on the strength of these — a company with zero debt scored as maximum leverage risk. That sub-score is rejected outright. priceToBookScore of 1 at 17.8x book is a correct signal and is accepted; returnOnEquityScore and returnOnAssetsScore of 5 are also correct at 30.6% and 17.2%.

4. The FY2023 estimate row is corrupt by roughly 4.8x — DISCARDED. It reports epsAvg of $7.888 and netIncomeAvg of $10.199 billion against actual FY2023 diluted EPS of $1.65 and net income of $2.087 billion. The FY2022 row is correct ($1.096 estimated against $1.07 actual), so this is a single corrupt row rather than a systematic split-adjustment failure. It is discarded entirely and no conclusion in this dive touches it.

5. est.ebitdaAvg and est.ebitAvg differ by roughly $11 million in every forward year — the EBITDA rows understate depreciation by ~85% and are NOT USED. FY2026 ebitdaAvg of $7.300 billion against ebitAvg of $7.289 billion implies depreciation and amortisation of $11 million, against an actual FY2025 figure of $72.6 million. The arithmetic screens in the data contract pass — EBIT below EBITDA, no negatives — but the rows are economically wrong and every conclusion here runs on epsAvg and revenueAvg.

6. The FY2026 revenue consensus is arithmetically inconsistent with the guidance issued this morning — flagged and replaced with our own estimate. First-half revenue was $5.745 billion and the third quarter is guided to approximately $3.3 billion. A FY2026 consensus of $11.617 billion therefore implies a fourth quarter of $2.572 billion — a 22% sequential decline from the guided third quarter, on a business that has grown sequentially in every quarter in this file. The estimates predate the print by hours. We use our own FY2026 estimate of approximately $12.5 billion and FY2027 of approximately $16.0 billion, labelled as ours throughout.

7. The street's price targets predate the print and sit BELOW the price — flagged. Consensus $186.67, median $187.50, high $200, low $164 — against a closing price of $190.55 — on a distribution of 39 buy, 13 hold and ZERO sell. A universally positive rating set with an average target below spot is not a signal about value; it is a signal that the targets have not been updated.

8. quote.yearHigh and tech.hi52 disagree, and both are informative — we use tech. The quote block reports a 52-week high of $194.3499, which is the intraday high of the current session; the computed block reports $190.36, which is the closing high. tech.pct_from_hi reads exactly 0.00% and tech.max_dd_from_peak also reads 0.00%, correctly identifying a 52-week-high CLOSE and a coincidence of the 52-week and multi-year peaks — per the data contract this is NOT a defect. quote.yearLow of $114.52 against tech.lo52 of $116.13 differ by 1.4%; we use tech for consistency with the moving averages. Note also tech.last of $190.36 against quote.price of $190.55 — a 10-basis-point difference resolved in favour of the quote for price and tech for everything computed from the close series.

Not defects, correctly reported and independently confirmed: the seg_prod block ties to reported revenue exactly for every year (FY2025 Product $7.5769B plus Service $1.4288B equals $9.0057B); the seg_geo block also ties exactly (Americas $7.1221B, EMEA $1.0703B, Asia Pacific $813.3M); the implied share count of approximately 1,259.2 million is consistent with the 1,273.8 million diluted weighted average with no share-class complication; capexToRevenueTTM of 1.50% correctly reflects the asset-light model; and returnOnInvestedCapitalTTM of 22.4% is consistent with the invested-capital base.

Non-equity tripwire — checked and passed. ANET is common stock, NYSE-listed, one class. Beta 1.615 — the highest in this batch; a 52-week range of $116.13 to $190.36, a 64% spread; no dividend; volume of 10.79M shares (roughly $2.06B of turnover). This is common equity.

8. Technicals

• Price $190.55. tech.pct_from_hi = 0.00% — the stock CLOSED at its 52-week high, having traded to $194.35 intraday. +63.9% above the 52-week low of $116.13. Position within the annual range: 100th percentile.
• Moving averages: +13.2% above a 50-day average of $168.31; +30.2% above a 200-day average of $146.31. Both rising.
• RSI 64.8 and MACD +3.36 — firm, not extended. An RSI below 65 on a stock closing at its 52-week high after a 61.9% twelve-month advance indicates a steady climb rather than a spike.
• Maximum drawdown from peak over the trailing year: 0.00% — the current price IS the peak.
• Relative performance: 3-month +10.28% vs SPY +7.59% and QQQ +7.67%; 6-month +37.57% vs SPY +11.09% and QQQ +15.71%; 12-month +61.91% vs SPY +24.26% and QQQ +30.80%. A 38-point twelve-month outperformance on a 1.615-beta stock.
• Sentiment: 39 buy, 13 hold, ZERO sell across 52 analysts — 75% buy and no sells, the most uniformly positive distribution in this batch. Consensus target $186.67 — 2.0% BELOW spot — median $187.50, high $200 (+5.0%), low $164 (−13.9%). A $36 spread between high and low is the NARROWEST in this batch, and combined with an average target below the price it is unambiguous evidence that the targets predate this morning's print.

Today's move

ANET closed 2026-08-04 at $190.55, up 3.06% or $5.66 from a $184.89 close. It opened at $187.70, traded $187.01 to $194.35, and closed 1.9% below the day's high on 10.79M shares against a 9.74M average. The second-quarter release was published the same morning, beating consensus EPS by 14.6% and revenue by 7.3%, with third-quarter guidance of approximately $3.3 billion. The session also saw a broad semiconductor and technology-hardware rally, so part of the move is sector.

The honest read. A 3.06% response to a 14.6% earnings beat and a strong guide is a measured reaction, not a melt-up — and it still leaves the stock at its highest close of the year. There is nothing wrong with the setup; there is simply no discount in it.

9. Insiders — an officer trimming into strength

The most recent transaction in the file is an S-Sale by Kenneth Duda, director and President and Chief Technology Officer, of 6,062 shares at $169.8977 on 2026-07-20, held indirectly, leaving 1,041,106 shares.

The reading. 6,062 shares against a retained position of 1,041,106 is 0.58% of the holding — a trim, executed fifteen days before a print that took the stock to $190.55, roughly 12% above the sale price. The retained stake is worth approximately $198.4 million at today's close. This is a routine, immaterial disposal by a founder-executive with an enormous unhedged position, and it carries no signal.

What the file contains is therefore: zero open-market purchases and one 0.58% officer trim. Neutral, and reported as neutral. The next Form 4 cycle, following a 52-week-high close, will be more informative than this one.

10. Verdict, kill-criteria and flip conditions

Watch.

What is genuinely exceptional, and every figure is from a filing released this morning: revenue of $3.036 billion — the first $3 billion quarter — up 37.7% year on year and 12.1% sequentially; GAAP operating margin of 45.4% and non-GAAP of 49.9%, BOTH HIGHER than a year earlier, while every comparable artificial-intelligence hardware name in this batch has seen margin compress; GAAP diluted EPS from $0.70 to $0.95 and non-GAAP from $0.73 to $1.02; a 14.6% earnings beat, the sixth consecutive beat in this file and the largest of them; third-quarter guidance of approximately $3.3 billion at a 48-49% non-GAAP operating margin; ZERO debt and $10.743 billion of net cash — 4.5% of market capitalisation; a 47-52% free-cash-flow margin on capital expenditure of 1.5% of revenue; a 22.4% return on invested capital; five-year revenue compounding of 31.2% a year with net margin rising from 27.4% to 39.0%; and a product position — 1.6-terabit AI fabrics, optics cutting interconnect power approximately 60%, protocol work aimed at maximising cluster utilisation — in the two binding constraints of the artificial-intelligence build.

What prevents a Buy: 60.1x trailing GAAP earnings, 55.1x trailing non-GAAP, 42.7x FY2027 consensus and 21.7x trailing sales on a 52-week-high close after a 61.9% twelve-month advance; a base fair value of $195 that is only 2.3% above spot, with 1.06:1 asymmetry — barely above parity; 79.1% of revenue in one region with no customer-concentration disclosure anywhere in this file; 245.4 days of inventory, the highest in this batch, on a balance that rose 22.5% year on year; a beta of 1.615, the highest here; a shareholder yield of 0.67%, effectively nil; and — uniquely among these twelve names — a knowledge base containing zero claims, so there is no independent qualitative check on any of the above.

The distinction that matters. This is not a valuation argument about business quality. Arista is, on the numbers in this file, the highest-margin growth business in this batch and possibly the best-positioned. The argument is that a 52-week-high close on the day of the print, at 60x trailing earnings, with our own fair value 2.3% above the price and no outside view to test it against, is not an entry — it is a decision to pay for everything in advance. In the Synthos frame, an excellent business at fair-value-equals-price with zero conviction lane is a Watch by construction.

Pre-registered KILL criteria — what would take this to Avoid:

• Non-GAAP operating margin below 47% in any quarter, against guidance of 48-49%. This is the entire premium — it is what separates Arista from a commodity switch vendor and from Dell's 17.7% gross margin.
• Any sequential revenue decline, breaking a run of growth in every quarter in this file.
• An inventory charge against the $2.247 billion balance at 245.4 days.
• Disclosure of customer concentration above roughly 20% for a single buyer combined with an order pause.
• Third-quarter revenue below the guided $3.3 billion, i.e. a miss against the company's own number issued hours ago.
• Americas revenue declining, which at 79.1% of the total would be a demand event rather than a mix event.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $150 — 33.6x FY2027 consensus, 30.6x our own estimate, near the 200-day average of $146.31 — at which the base case is +30% and the asymmetry becomes genuinely attractive. This is 21% below today's close.
• A hold of the 50-day average at $168.31 on any consolidation, combined with FY2027 consensus revenue revising above $15 billion — which would confirm the sell-side has incorporated the guide and that our base case is conservative.
• Two further quarters of non-GAAP operating margin at or above 49% as the artificial-intelligence fabric mix grows, which would convert the margin premium from cyclical to structural.
• The enterprise campus business sized separately for the first time. This morning's Gartner recognition points at a much larger and more fragmented customer base that would materially reduce the concentration risk, and it is currently unquantified.
• Any independent knowledge-base coverage appearing at all. At present there is none, and that alone caps the position size regardless of the arithmetic.

Where ANET fits in the Synthos Framework Portfolio. The AI-infrastructure sleeve, at 0% today with a 2% target on a fill near $150. On batch overlap, this is the highest-quality expression of the same trade Dell and Palantir express: Arista sells the fabric, Dell assembles the racks, Palantir sells the application layer, and Texas Instruments supplies analogue content into the edge of it. The margin comparison is the cleanest available statement of where the value accrues: Arista 49.9% non-GAAP operating margin on 37.7% growth; Dell 8.3% operating margin on 87.5% growth. Owning both is one trade at two points in the stack, and Arista is the point at which pricing power exists. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $190.55.

Single biggest risk: the multiple, compounded by concentration we cannot see. At 60.1x trailing GAAP earnings and 21.7x trailing sales, on a 52-week-high close, with 79.1% of revenue in the Americas and no customer-concentration disclosure in this file at all, the price assumes both that growth persists and that the buyer base is durable. This dive can verify the first from the filings and cannot verify the second from anything. A 10% revenue disappointment at this multiple does not produce a 10% decline; it produces a de-rating on top of one, which is what the $120 bear case is — and that bear case requires no revenue miss, only margin compression and a return to a normal multiple.

Most fragile assumption in the price: that a 49.9% non-GAAP operating margin survives the artificial-intelligence mix shift. This is the single number that justifies paying ten times as much per dollar of sales for Arista as for Dell. It has held and improved so far — 48.8% to 49.9% year on year, with the third quarter guided to 48-49%. But every comparable name in this batch that has scaled into the same build has seen margin compress: Dell's gross margin fell 340 basis points in a year and management guides to more. If Arista's non-GAAP operating margin settles at 44% rather than 49%, FY2027 EPS is roughly 11% lower and the multiple the market pays is lower still — the combination puts the shares nearer $140 than $195 with revenue exactly where consensus expects it. That is the specific mechanism by which an investor loses 27% here while every growth number in this dive proves correct.

Provenance & disclosures

• Traceability: ZERO knowledge-base claims name Arista Networks. Zero entity matches and zero text matches across all 51,928 distilled claims. Searches covered ANET, Arista and Arista Networks against the entity field, plus a text sweep on the token arista across claim text and categories. Nothing returned. This is the only name in this batch with no coverage at all — against Palantir's 83 entity claims, Dell's 29, Goldman Sachs's 28, Chevron's 26 and Merck's 24 — despite Arista being a $239.93 billion company at the centre of the same artificial-intelligence infrastructure build those lanes discuss. The consequence is stated rather than papered over: there is no independent qualitative check in this dive on the competitive position, the customer concentration, the durability of a 49.9% operating margin, or the competitive threat from accelerator vendors moving into the fabric. Breadth 0, conviction contribution zero, and the absence is a direct input to the Watch verdict.
• Data as-of: second-quarter 2026 results — revenue, GAAP and non-GAAP operating margins, GAAP and non-GAAP diluted EPS, and third-quarter guidance — from the 8-K/Exhibit 99.1 filed 2026-08-04, the day of this dive; fiscal 2025 and the first quarter of 2026 from the vendor payload, cross-checked against the 10-K filed 2026-02-17 and the 10-Q filed 2026-05-06 · estimates 2026-08-04 (and stale by hours — see below) · prices 2026-08-04, quote timestamp 1785873601 = 2026-08-04T20:00:01Z ($190.55, +3.06%, a 52-week-high close; 50-DMA $168.31; 200-DMA $146.31; RSI 64.8; MACD +3.36) · knowledge-base claims 2026-08-04. Arista's fiscal year is the calendar year. All figures come from the Synthos vendor data file for ANET or from the SEC filings in the ANET archive; no figure comes from memory, recall or external retrieval.
• Estimate and target staleness — stated prominently because it changes the arithmetic. Second-quarter results were released on the morning of 2026-08-04. The evidence that every estimate and target in this file predates it is direct and arithmetic: the FY2026 revenue consensus of $11.617 billion, set against first-half actual revenue of $5.745 billion and third-quarter guidance of approximately $3.3 billion, implies a fourth quarter of $2.572 billion — a 22% sequential DECLINE that nobody forecasts. And the consensus price target of $186.67 sits 2.0% BELOW the closing price on a distribution of 39 buy, 13 hold and ZERO sell ratings. Our fair values therefore use our own FY2026 and FY2027 revenue estimates of approximately $12.5 billion and $16.0 billion, labelled as ours at every occurrence, cross-checked against the stale consensus.
• Filing archive contents: 10-K filed 2026-02-17 (fiscal 2025, tables preserved); 10-Q filed 2025-11-05 (September 2025 quarter); 10-Q filed 2026-05-06 (March 2026 quarter); 8-K filed 2026-05-05 (Q1 results); 8-K filed 2026-06-02; 8-K filed 2026-08-04 carrying the full Exhibit 99.1 second-quarter release with the financial highlights, guidance, management commentary and product disclosures used throughout this dive. Tables were available on all substantive documents.
• Where the filings corrected or extended the vendor (detailed in Section 7): the vendor payload does NOT contain the June 2026 quarter at all — inc_q ends at 2026-03-31 and earn_cal records the actual as null — so every June figure here comes from the 8-K, and every trailing metric in the payload is struck on the wrong window and has been recomputed; netDebt corrected from −$1.964 billion to $10.743 billion of net cash for the omission of $8.779 billion of short-term investments — the largest omission of that class in this batch — taking enterprise value from $237.145 billion to $229.19 billion and enterprise-value-to-sales from 24.42x to 21.74x; six debt and coverage ratios identified as UNDEFINED rather than zero, with the vendor's debtToEquityScore of 1 out of 5 rejected outright for a company with no debt; the FY2023 estimate row DISCARDED as corrupt (epsAvg $7.888 and netIncomeAvg $10.199 billion against actual $1.65 and $2.087 billion — overstatements of 4.8x and 4.9x — while the FY2022 row is correct); est.ebitdaAvg and est.ebitAvg found to differ by only ~$11 million against actual depreciation of $72.6 million, understating it by roughly 85% and rendering the rows unusable; the FY2026 revenue consensus shown to be arithmetically inconsistent with guidance issued hours earlier and replaced with our own estimate; and the street's price targets flagged as predating the print. Where vendor and filing AGREED — worth recording: the seg_prod block ties to reported revenue exactly for every year, the seg_geo block likewise, the implied share count is consistent with the diluted weighted average with no share-class complication, and tech.pct_from_hi of exactly 0.00% correctly identifies a 52-week-high close (with tech.max_dd_from_peak also at 0.00%, a legitimate coincidence of the 52-week and multi-year peaks rather than a defect).
• Basis note: Arista reports both GAAP and non-GAAP results; the non-GAAP measure excludes stock-based compensation and related items. The June quarter delivered GAAP diluted EPS of $0.95 and non-GAAP of $1.02 — a 7.4% gap, which is narrow for a company with stock-based compensation at 4.9% of revenue. The earn_cal consensus of $0.89 is on the non-GAAP basis, as is the guided 48-49% operating margin. The est block appears to be GAAP-based — the FY2024 estimate of $2.192 against a delivered GAAP $2.23 is a close match while non-GAAP would have been materially higher. Trailing multiples are shown on both bases (60.1x GAAP, 55.1x non-GAAP); forward multiples use epsAvg and are therefore GAAP, and are labelled as such.
• Estimate coverage: 20 analysts on FY2026 revenue and 16 on FY2026 EPS; 21 and 16 on FY2027; 19 and SIX on FY2028; 10 and FOUR on FY2029. Neither the FY2028 nor the FY2029 EPS row informs any conclusion. est.ebitAvg and est.ebitdaAvg are not used; the FY2023 row is discarded entirely.
• Peer note: the vendor peer set is a market-capitalisation grouping rather than a comparable set — Applied Materials, Amphenol, AppLovin, Intuit, Lam Research and ServiceNow are not networking companies, and no genuine networking comparable is supplied at all. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The relevant within-batch observation is that Arista's $239.93 billion market capitalisation is 77% of Dell Technologies' $310.32 billion on 7.9% of Dell's revenue — roughly ten times the price per dollar of sales — and the 49.9% versus 8.3% operating-margin difference is the reason.
• Fair-value caveat: the $120 / $195 / $265 anchors are multiples of the FY2027 consensus GAAP diluted EPS of $4.468 — 26.9x, 43.6x and 59.3x — cross-checked against our own FY2027 revenue estimate of approximately $16.0 billion, which is 11% above the stale consensus and is labelled as ours. Sensitivity on the consensus figure: 25x gives $112, 35x gives $156, 50x gives $223; the street's $186.67 implies 41.8x. Stated arithmetic, not a discounted cash flow. The bear case requires NO revenue miss — only margin compression and a return to a normal multiple. The bull case requires no re-rating on our own numbers, only that revenue lands where we think it will. And no case incorporates any independent qualitative view, because none exists: the knowledge base is empty.
• Timing: second-quarter 2026 results were released on the morning of 2026-08-04, the day of this dive, beating consensus EPS by 14.6% and revenue by 7.3%, with third-quarter guidance of approximately $3.3 billion at a 48-49% non-GAAP operating margin. The stock closed +3.06% at its 52-week high. No forward earnings date appears anywhere in this file — the most recent earn_cal entry is 2026-08-04 itself — and on the prior-year cadence (2025-11-04) the next print falls in early November, roughly 90 days away. The verdict is set with a 52-week-high close on the day of the print as an explicit input.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.