Earnings Highlights: Arista Networks (ANET) Q3 FY2026: Revenue Up 58.5%, a Post-IPO Record. 2027 Guide of $18.5B, AI Doubling to $7B, Scale-Up Waits Until 2028
1. Three Core Takeaways
First, growth accelerated without diluting profit. Revenue grew 20.5% sequentially while non-GAAP operating margin held at 49.1%. R&D rose $130M in one quarter and gross margin was 1.2pp below last year, yet the margin line held.
Second, two-thirds of 2027's growth comes from AI. Working from management's targets, about $3.5B of the $5.2B increase is AI, or roughly 67%. AI moves from about 26% of revenue to about 38%, tilting both the customer and product mix toward large AI deployments.
Third, the next product cycle isn't in the numbers yet. Scale-up has no material revenue in 2027, and CPO comes after 2028. The $18.5B for 2027 rests on scale-out and scale-across, meaning today's Etherlink and 7800 platforms.
2. Revenue and Financials
The card uses GAAP; management speaks in non-GAAP on the call. The gap between them was meaningful this quarter, so both are shown:
Metric | GAAP | YoY | non-GAAP | YoY |
Revenue | $3,657.6M | +58.5% | $3,657.6M | +58.5% |
Gross margin | 63.6% | −1.0pp | 64.0% | −1.2pp |
Operating income | $1,642.4M | +67.9% | $1,795.8M | +60.1% |
Operating margin | 44.9% | +2.5pp | 49.1% | +0.5pp |
Net income | $1,560.4M | +82.9% | $1,530.8M | +56.3% |
Diluted EPS | $1.22 | +82.1% | $1.20 | +55.8% |
GAAP net income came in above non-GAAP, and that needs explaining first. The main reason is a $134.5M gain on strategic investments, which non-GAAP excludes. So the 82.9% GAAP increase isn't the speed of the core business. For that, use the 56.3% non-GAAP figure.
▲ Arista Networks FY2026 Q3 core financials on a GAAP basis: revenue, gross margin, operating margin, net income, EPS, YoY, segment revenue and next-quarter guidance. (Image source: Simple Tech Trend original; data from Arista's Q3 2026 earnings release and call)
Segments and regions. Product revenue was $3,190.0M, up 66.9%; service revenue was $467.6M, up 17.9%. International revenue was $682.5M, falling from 23% of revenue last quarter to 18.6%. The CFO attributed that to the geographic mix of large global customers, and all three regions still grew double digits.
Gross margin. Non-GAAP 64.0% was below last year's 65.2% because product cost inflation was only partly offset by supply chain efficiency. It was above last quarter's 63.4% on customer mix.
R&D was the biggest surprise. Non-GAAP R&D jumped from $278.1M to $408.5M, or 11.2% of revenue. CTO Ken Duda pointed to liquid cooling, high-radix products and co-development with hyperscale customers, and the CEO added that the prototype bill was very high. Asked whether this is a new run rate, the CFO kept the 9%-11% framework, so for now 11.2% reads as an investment peak.
Cash flow. Operating cash flow was about $1.2B, capex was $100.7M, and the quarter ended with about $14.4B in cash and marketable securities and no debt. Inventory reached $3.2B, up from $2.5B, which the CFO described as an intentional build.
3. Breaking Down the 2027 Guide: What's Inside $18.5B
Last quarter the full-year guide stood at $12.6B, with scale-across as the main story. This quarter added another $700M and went straight to next year's number. Management confirmed the 2026 baselines in the Q&A:
Item | 2026 target | 2027 target | Growth |
AI Fabrics | $3.5B | ~$7B | ~+100% |
Campus | $1.25B | ~$2B | +60% |
Other (STT estimate) | ~$8.55B | ~$9.5B | ~+11% |
Total revenue | ~$13.3B | ~$18.5B | +38%–40% |
Three numbers deserve a closer look.
The implied 11% for everything else needs a discount. This is what the JPMorgan analyst asked about. Jayshree said not to read conservatism into it and that everything will grow double digits. But $3.5B and $1.25B are targets, and management says actuals won't be known until year-end. If 2026 AI beats its target, the base for the rest shrinks and the growth rate moves with it. The 11% is simply targets subtracted from each other, not company guidance.
The guide doesn't demand a steep slope. STT's rough math: starting from the Q4 midpoint of $3.95B, about 6.4% sequential growth per quarter reaches $18.5B. Compare that with 12.1% in Q2 and 20.5% in Q3, quarters that also beat guidance by about 8% and 11%. Jayshree says not to call it conservative, but the numbers leave room, and the real ceiling is supply.
Visibility is only two quarters. The CFO was clear: two quarters of visibility, with the second half resting on conversations with large customers and enterprise momentum. Jayshree's version is two quarters of visibility plus two years of work. Half of this guide comes from conversations that haven't become orders.
4. Technology and Business Highlights
Scale-up: Trials in 2027, Production in 2028
At OCP in October, Arista showed three rack-scale designs: orthogonal, cabled backplane and cross-rack. Partners include Arm, Microsoft Maia, AMD Helios, Qualcomm and Meta MTIA, all based on Ethernet for Scale-Up (ESUN). Management's timeline is trials starting in 2027, with radix reaching 144 within a rack and more than 1,000 across racks, and production in 2028. 2027 has no material revenue.
Jayshree called scale-up a tough market for Arista because it doesn't use much of the company's software advantage; the opening is non-NVIDIA accelerators. She compares UALink, NVLink and ESUN to the old InfiniBand versus Ethernet fight. That's a position, and 2028 shipments will test it. For the buyer's view from the same event, see Meta's Tahan, a scale-up switch built by subtraction and AMD Helios and the open rack.
The business model isn't settled. Asked whether Arista will sell whole racks or only switches, Jayshree said they're still figuring it out, with each of the three designs shipping differently. That echoes the rack manufacturing challenges Arista and Jabil described at OCP: once a switch becomes a rack, gross margin structure and the contract manufacturing split both get recalculated. For now, management says the long-term 62%-64% gross margin doesn't change.
Scale-out and Scale-across: A 70/30 Split, and the 30 Grows
About 70% of 2026 AI revenue is scale-out and 30% scale-across, and Jayshree expects scale-across to take a bigger share in 2027. Revenue concentrates in the 7800 AI spine at large customers, while site count favors the 7280. Competitors are pushing here too: compare Cisco elevating scale-across to a third fabric layer at OCP.
Etherlink has passed 100 customers. The company forecasts more than 10 GW deployed by the end of 2027 across 10 or more types of compute accelerators. Pressed by Goldman Sachs on whether this is an Arista or an industry number, Jayshree's answer pointed to Arista's own deployments, but she also said it includes scale-across over existing gigawatts, so it can't be read as entirely new capacity.
Optics: "Copper When You Can, Optics If You Must"
This is the part that matters most to optical readers. Jayshree's order is copper first, then pluggables beyond one or two meters, then XPO for higher front panel density, which she says over 150 vendors support. On CPO, Arista won't build proprietary designs and instead favors open CPO, using socketed optical engines and pigtail fibers rather than soldering so modules can be pre-tested. The timing is 2028 to 2030. In her words it is "not imminent," followed by "long live pluggable optics."
For the measured cost of XPO, see the system-level evaluation by Microsoft and Arista at OCP. Keep in mind that Arista is a lead backer of XPO, so this is also a product position rather than a neutral forecast.
Security: Turning AI Vulnerability Discovery into a Campus Sales Pitch
The CTO spent more than three minutes on security, saying Arista is working with Anthropic, Google, OpenAI and others to bring frontier-model vulnerability discovery into its development process, and has been invited into Project Glasswing. COO Todd Nightingale went further and tied campus growth to "the Glasswing and Mythos effect": enterprises now want rapid patching without downtime, which fits EOS's single operating system and process isolation.
Arista cites fewer than 100 cumulative CVEs and says it is 10 to 50 times better than alternatives. That's the vendor's own claim, with no third-party comparison. Jayshree also mentioned that its own vulnerability count doubled from 30 to 67, all fixed. AVA is still an alpha feature inside CloudVision with pricing under study, so it shouldn't figure into near-term revenue.
5. Management Outlook: Three Levels of Language
Q4 FY2026 guidance (non-GAAP except revenue):
Revenue $3.9B-$4.0B; midpoint up about 58.8% YoY and about 8% QoQ
Gross margin 63%-64%, operating margin 48%-49%
Diluted EPS $1.26-$1.30, about 1.28B diluted shares, tax rate about 21%
FY2027: Revenue growth of 38%-40%, about $18.5B; gross margin 62%-64%, operating margin about 49%, tax rate about 21%. Total addressable market above $200 billion by 2030.
Highest certainty: the CFO said she is confident twice, citing agreements with key suppliers and rising purchase commitments.
Medium: the number of 10% customers. Beyond Microsoft and Meta, she expects "at least one, maybe two" more, depending on how much ships by year-end.
Discount the most: a "promising 2028" and scale-up's "of course it's going to be ESUN." These are directions with no numbers behind them.
6. Supply Chain and Customer Signals
Purchase commitments of $11.4B, up from $9.7B. The CFO says these are mainly chips for new products and AI deployments, now including a multi-year memory agreement, and they should rise again in Q4. STT's rough math: at $18.5B and a 63% gross margin, 2027 cost of goods is about $6.8B, so $11.4B is about 1.7 times that. Multi-year contracts mean it can't all count toward 2027, but the direction is clear: Arista has placed its money ahead of demand.
Deferred revenue is the gap the market cares about most. The balance was about $6.8B, slightly below last quarter's $6.9B, and product deferred revenue fell $86M while revenue grew 58.5%. Several analysts pressed on whether the guide depends on drawing deferred down. The CFO's answer was that they don't guide deferred, though she expects the year-end balance to exceed last year-end. Jayshree said this turns demand into a physics problem she can't solve. The question wasn't really resolved and is worth revisiting next quarter.
Lead times come in two kinds. Enterprise products have shorter lead times and can carry buffer inventory; AI and hyperscale products run longer. Todd says long-term agreements cover silicon, PCB capacity and memory, and he has very high confidence in delivering to forecast, while conceding that not every 2027 order is placed. That fits the shortages upstream companies have described on their calls, and execution remains Arista's edge over peers.
Liquid cooling is a new cost line. Jayshree described a lot of switches "swimming in water" in rack designs. Ken named liquid cooling as one of the main R&D drivers, which suggests design-in for Taiwanese contract manufacturers and thermal suppliers could come before scale-up production in 2028. That's STT's inference; management named no suppliers.
7. Summary
Three indicators to watch next quarter:
Direction of deferred and product deferred revenue. The CFO committed to a year-end balance above last year-end, and Q4 is the first check.
Whether R&D returns to 9%-11% and gross margin holds 63%-64%. If R&D stays above 11%, a 49% operating margin has to lean on even higher revenue.
Purchase commitments and new 10% customers. The first shows how big Arista is betting, the second shows where the money comes from. Whether a third and fourth 10% customer appear in Q4 will shape customer concentration in 2027.
In one line: Arista delivered a story about 2027 order visibility this quarter, and the hard evidence in hand is $11.4B of purchase commitments. Nearly all of 2027's growth comes from existing scale-out and scale-across. Scale-up and open CPO wait until after 2028, so pluggables remain the main battlefield for the optical supply chain for the next two years.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments