ANET: Orders Aren't the Problem. Capacity Is.
Arista's constraint was never demand. RPO and order visibility keep expanding; the real ceiling is external chip and component supply — supply sets the growth ceiling, and valuation only makes sense inside that frame.
Core Conclusion
ANET doesn't make its own chips. Its switches are built on outsourced Broadcom Tomahawk and Jericho merchant silicon, plus third-party memory, optics, and PCBs. So when the CEO said on the earnings call, "this is a two-year problem for the industry, and I don't think we get out of it as an industry till 2028," that sentence isn't about abstract industry sentiment — it's about whether ANET itself can source enough components to turn its order book into shipped product. This isn't something money alone can fix; it depends on how foundries and memory makers allocate capacity.
But the current valuation (roughly 19–20x P/S) is priced off a denominator that assumes "35–40% growth continues next year." That assumption doesn't discount for the fact that supply, not demand, may set the ceiling on growth. If FY2027 growth gets capped near 25% by component constraints, the same multiple compresses to roughly 17.5x — not a crash, but a sign the current price has already bet on the smoothest possible script.
The good news: ANET's $8.4B in RPO is executable contract backlog — not the kind of air-order that comes with a 90-day cancellation clause the way some AI cloud deals do. The demand is real. Whether it can be shipped on time is the actual variable.
01Setting the fundamentals aside for a moment: Q2 in numbers
Q2 FY2026 income statement highlights (US$ millions unless noted)
Item | Q2 FY2026 | Q2 FY2025 | YoY |
|---|---|---|---|
Total revenue | 3,035.7 | ~2,204 | +37.7% |
Non-GAAP gross margin | 63.4% | 65.6% | −2.2pt |
Non-GAAP operating margin | 49.9% | 48.8% | +1.1pt |
Non-GAAP EPS | $1.02 | $0.73 | +40% |
Pre-earnings consensus revenue | $2.8B | — | Beat by 8.5% |
Three numbers form the narrative backbone of this print: a large revenue beat, a third consecutive full-year guidance raise, and a gross-margin decline that's primarily about customer mix, not discounting. The CFO was explicit on the call that pricing effects "haven't shown up yet" and won't until year-end or next year — in other words, the company itself thinks there's still room to expand margin, not that it's cutting prices to win orders.
None of this is in dispute, and it's not what this piece is here to challenge. Fundamental strength is why ANET is worth studying seriously; but strong fundamentals have never been the same thing as a reasonable entry price. These are two separate questions.
02What does today's valuation actually assume?
At the post-earnings high, ANET's market cap sits around $245B (roughly 1.26B shares outstanding × ~$194). Divide that by different revenue bases and you get completely different answers to "is this expensive" — which is exactly where valuation debates lose the plot: the multiple most people cite has usually already picked the denominator that flatters their thesis.
Implied P/S under different revenue bases
Revenue base | Amount | Implied P/S | What this base assumes |
|---|---|---|---|
Q2 annualized | $12.14B | ~20.2× | Assumes the next four quarters hold flat at Q2's absolute level — most conservative |
FY2026 guidance | $12.6B | ~19.4× | The denominator the market is actually using; assumes the full 40% guided growth lands |
If growth slows to 25% | ~$14.0B (FY2027 estimate) | ~17.5× | Supply-chain drag, but still healthy growth |
19–20x P/S is not outrageous for a company with non-GAAP operating margin near 50% and 37% YoY growth — which is exactly why the stock made a new high after earnings and analyst targets moved up across the board. The issue isn't that the multiple itself is unreasonable; it's whether that multiple has left any cushion for the risk management itself has flagged — that next year's growth might not be 40%, but something capped at 25–30% by supply.
The bull case that has to be acknowledged
On EV/EBITDA or PEG (price/earnings-to-growth), ANET's current valuation isn't expensive relative to its growth rate — unlike many higher-multiple software names whose growth has already decelerated. The market is willing to pay a premium because three consecutive guidance raises are a track record, not blind optimism.
03What "a two-year problem until 2028" really means
This line was worth verifying directly. On the earnings call, a Goldman Sachs analyst asked about supply-chain progress. After affirming the strength of Arista's silicon-vendor relationships, CEO Jayshree Ullal continued:
"I don't want you to believe that suddenly we waved a magic wand and all our problems have gone away. The industry is going to have a two-year problem. I don't think we get out of it as an industry till 2028." She immediately followed with: "But Arista is taking individually and specifically steps in the first half of this year that we believe will have results in the back half of this year."
COO Todd Nightingale later clarified that this specifically referred to silicon, memory availability, PCBs, and optical components — a supply-chain bottleneck, not a demand warning or a structural-decline signal.
This line deserves to be taken seriously precisely because ANET is fabless — it doesn't manufacture chips. The core compute in its switches runs entirely on outsourced Broadcom Tomahawk and Jericho merchant silicon, and memory, optical transceivers, and PCBs all depend on third-party suppliers too. Arista has no "build our own fab" option — how much capacity foundries and memory makers allocate, and to whom, is something Arista can compete for a share of, but cannot dictate the total supply of.
How this line should enter the valuation model
If the bottleneck sits on the supply side rather than the demand side, that's a near-term positive (demand > supply means more pricing power and more say in who gets served first). But over the medium term, it means ANET itself is admitting that the ceiling on growth over the next one to two years isn't set by orders — it's set by whether the company can source enough silicon, memory, and optics. The market is currently pricing in "40% growth keeps landing" without discounting at all for "management itself says this bottleneck doesn't ease until 2028." This is the largest unreconciled gap between valuation and fundamentals in this story.
This isn't speculation — management's own FY2026 guidance holds the gross-margin target at 62–64% while explicitly stating it already factors in expected increases in memory and silicon costs. In other words, the official guidance also admits input costs are rising; management is just choosing to absorb it within a maintained range rather than cut the target. That's more direct confirmation of a real supply-side cost pressure than any secondhand commentary.
This quarter's gross-margin number can also be tested directly against the two thresholds this site's May piece set: an upgrade trigger of gross margin above 65% for two consecutive quarters, and a downgrade trigger of below 62% for two consecutive quarters. Q2 actual was 63.4% — sitting between both thresholds, neither tripped. But the direction isn't neutral: down 2.2pt from 65.6% a year ago, it's moving toward the downgrade line, not the upgrade one. No alarm needed yet, but if it keeps sliding next quarter, 63.4% becomes the last bit of cushion before the 62% downgrade line.
04Working backward from RPO: how much of "next year" is already locked in?
ANET disclosed Q2-end RPO (remaining performance obligations) of $8.4B, a clear jump from the prior period. This number's significance needs to be separated from the SPCX-style cloud subscriptions that "can be cancelled unconditionally with 90 days' notice" — ANET's RPO comes from traditional hardware/software contractual obligations, which carry far higher enforceability than cloud leases with cancellation clauses.
RPO coverage: how much "future revenue" is already locked in
Metric | Amount | Conversion |
|---|---|---|
RPO (remaining performance obligations) | $8.4B | — |
FY2026 guided revenue | $12.6B | — |
RPO ÷ annualized revenue | ~67% | Roughly 8 months of guided revenue is already under contract |
A 67% coverage ratio is on the high side for this class of high-growth tech company — most pure-subscription software companies run RPO coverage in the 40–55% range. This is the key difference between ANET and last year's batch of hyped-then-disappointing "AI story stocks": its order visibility is real, not narrative-supported.
Bull case: this isn't an air estimate
Over the same period, the company also expanded its multi-year purchase commitments (supply-lock obligations to vendors) to $9.7B — management is putting real capital behind the bet that "demand persists for the next two years." If ANET genuinely believed growth would halve next year, it wouldn't be doubling down on locking in this much upstream capacity.
These numbers echo, to some degree, one of the upgrade triggers set in the May piece — the stated condition then was "Meta or Microsoft announce expanded AI capex." There's been no press release in that exact form, but RPO jumping to $8.4B and multi-year purchase commitments tripling to $9.7B is essentially the same thing showing up in a different format: both the customers and Arista itself are writing "expanded capex" onto the balance sheet in contract dollars instead of headlines.
But high RPO coverage answers "is the revenue number real" — it doesn't answer "can it be delivered on time, and will supply-chain costs eat into the margin when it is delivered." That's exactly what Section 03's "two-year problem" is about. Orders genuinely existing and orders being efficiently converted into profit are two different things.
05Customer concentration: an old risk, but not a worsening one
Annual customer concentration (10-Q disclosure, not quarterly)
Customer | 2025 | 2024 | 2023 |
|---|---|---|---|
Customer A | 16% | 15% | 21% |
Customer B | 26% | 20% | 18% |
When asked on the call whether any additional 10%-plus customers would emerge, management said only that it expects "one, maybe two," while reaffirming its commitment to long-term partners Microsoft and Meta — but without positively committing that either share would stop rising. The 10-Q itself states: "the pricing discounts typically required for these large-scale orders adversely impact our gross margins" — this maps directly to the gross-margin decline seen in Section 01, and it's a cause-and-effect relationship the company acknowledges itself, not an analyst's overreach.
This risk isn't new — it has existed for three years, and there's no evidence it's worsening (combined top-two concentration: 39% in 2023, 35% in 2024, 42% in 2025 — fluctuating within a range, not trending one way). It belongs in the valuation discount rate, but it isn't grounds for "sell immediately."
A balancing number worth including
Revenue concentrated in two customers doesn't mean the buyer base hasn't widened. ANET's AI fabric line (Etherlink) has grown from 4–5 customers in 2024 to more than 100 cumulative customers today. In other words, the top-line revenue contribution is still concentrated in Microsoft and Meta, but the tier below has clearly broadened — that lowers the tail risk of "one of the two big customers pulls back and the company immediately stalls," even though it hasn't yet shown up in the concentration percentages.
06Where this call could be wrong
The counter-evidence that has to be stated honestly
- Three consecutive guidance raises aren't a coincidence. That track record over the past year says management's visibility model has been credible, and it shouldn't be fully overturned by one quarter's cautious remark.
- RPO coverage is on the high side versus peers, and if the supply bottleneck really is constraining growth, it's also constraining the odds of "orders disappearing" — when supply is short, customers typically don't cancel, they queue for delivery.
- The gross-margin decline has a reasonable, transparent explanation — the company acknowledged the causal link itself in its 10-Q, rather than being caught out by the market.
- Analyst price targets have moved up across the board (Piper Sandler, JPMorgan, and Barclays all raised targets after earnings) — the caution taken here is a minority position.
- No downgrade-side signal has emerged on NVIDIA Spectrum-X either. On the Q2 call, BNP Paribas analyst Karl Ackerman pressed on whether new cloud players adopting custom silicon (AMD MI, Google TPU) constitutes an alternative path to the NVLink ecosystem. Ullal acknowledged ANET's participation is indeed higher in non-NVIDIA accelerator environments, but remained cautious about breaking into the NVIDIA ecosystem itself: "That's going to take time... it feels like three years ago when we were talking about InfiniBand." No deployment contract above $500M was disclosed — one of the downgrade triggers set in the May piece wasn't tripped this quarter.
If the real-world impact of the supply bottleneck turns out milder than "a two-year problem" sounds — for instance, if Arista's scale and cash position let it capture a disproportionate share of scarce capacity versus peers — then a 19–20x multiple may not be expensive at all, and could even be cheap. That scenario can't be ruled out; the judgment here is simply that the current price looks like it has already assumed the most favorable version of events.
07Tracking indicators and falsification points
Verification checklist for the next print (Q3, ~November 2026)
Indicator | Signal supporting the valuation | Signal the valuation should correct for |
|---|---|---|
Q3 revenue vs. $3.3B guide | Beats again | Flat or misses → growth curve starting to be constrained by supply |
RPO coverage | Holds above 60% | Meaningful decline → order authenticity needs re-examination |
Non-GAAP gross margin | Stabilizes or improves | Continues falling → customer-discount pressure worse than expected |
Top-two customer concentration | Flat or declining | Meaningfully rising → single-point risk heating up |
Supply-chain language | "Two-year problem" tone softens | Holds or worsens → FY2027 estimates should come down in parallel |
A few product-line developments are worth watching going forward, though none of them currently move the valuation call: the next-generation 7060X-E7 (1.6T bandwidth, first liquid-cooling option) is expected to enter trials in H2 2026, with real production shipments not until 2027; meanwhile management introduced a set of AI-specific features — SSU (Smart System Upgrade), MRC (Multipath Reliable Connection), SRv6 load balancing — aimed at deepening EOS's technical differentiation within AI clusters. These are medium-to-long-term moat-building moves. They only earn a place back in the valuation model once the 1.6T platform is actually in production and shows up in revenue and margin — for now, this is narrative, not a verifiable financial event.
08Assumptions and known limitations
Items calculated in this piece
- Market cap and share price: estimated using roughly 1.26B shares outstanding and a price near the post-earnings high, not an exact closing price on a specific date.
- FY2027 estimated revenue of $14.0B: assumes growth decelerates from FY2026's 40% to 25%; a scenario illustration only, not company guidance.
- 67% RPO coverage: RPO divided by FY2026 guided revenue; does not account for the portion of RPO whose recognition timeline may extend beyond one year.
Known limitations and a note on evidence tiers
The financial figures in this piece are drawn primarily from the company's earnings press release, 10-Q disclosures, and the official earnings-call transcript; some industry commentary and stock-reaction color is drawn from media reporting that hasn't been cross-checked line-by-line against primary filings — readers are welcome to verify against the sources listed. The 10-Q customer-concentration figures are annual disclosures, not this quarter's numbers, as noted in Section 05.
Disclaimer This report is an analytical record of financial-statement structure and valuation methodology, intended solely for research and educational discussion, and does not constitute investment advice, a buy/sell recommendation, or financial planning guidance. The multiples, scenario assumptions, and inferences presented are illustrations of an analytical framework and do not represent a price target or directional forecast.
The author does not warrant the completeness or accuracy of the information presented. Readers should conduct their own due diligence and make independent judgments based on their own financial circumstances and risk tolerance, or consult a qualified professional. Investing involves risk, including possible loss of principal.
This report takes a cautious stance, but no matter how solid the reasoning, markets can disagree for extended periods. Section 06 lists the counter-evidence to this report's judgment in full — please read it alongside the rest.
Shiba the Disciplined
ProfitVision LAB · 2026.08.08