Inve US ▾
ANANET

Arista Networks (ANET) Q1 FY26 earnings call: guidance and Q&A

Earnings call· Mar 2026 (Q1 FY26)· Information Technology· May 5, 2026

Source: earnings call transcript

Executive summary

Arista Networks Q1 FY26 — Record demand and raised AI/revenue outlook amid worsening supply constraints

Demand is the best management has seen in its tenure, but the binding constraint has shifted from demand to a multi-year component-supply squeeze that Arista is meeting by pre-committing cash and accepting margin give-up to keep AI infrastructure from sitting idle. The AI-fabric story is broadening beyond scale-out into scale-across and, later, Ethernet scale-up, while enterprise/campus diversification adds ballast. Forward demand outstrips shippable supply, making supply — not orders — the swing factor for 2026-27.

Highlights

5
  • Total revenue $2.71B, up 35.1% YoY and above the $2.6B guidance, led by AI and specialty-cloud providers

  • Non-GAAP diluted EPS $0.87, up 31.8% YoY; non-GAAP operating margin 47.8% of revenue

  • Now #1 market share in high-speed (>10GbE) switching for 2025, overtaking incumbent vendors per major analysts

  • Record operating cash flow of ~$1.69B, the strongest in Arista's history; >100 cumulative 800GbE customers

  • FY26 outlook raised to 27.7% growth (~$11.5B) and AI-fabric goal raised to $3.5B (more than doubling AI sales)

Concerns

5
  • Non-GAAP gross margin 62.4%, down ~170 bps YoY, on adverse customer mix plus memory/silicon cost inflation

  • Industry-wide component shortages (wafers, silicon, CPUs, optics, memory) with ~52-week lead times, now framed as a 1-2 year phenomenon

  • Management flagged unfavorable supplier de-commits and is deliberately accepting gross-margin pressure to assure supply

  • Customer-acceptance/qualification cycles have stretched from 2-4 quarters to 6-8 quarters, driving deferred-revenue volatility

  • No shares repurchased in Q1 despite $817.9M remaining authorization; high customer concentration (only Microsoft and Meta >10%)

Guidance & targets

CategoryTargetConfidence
Full-year 2026 revenue growth
27.7% growth, approximately $11.5B
high materiality
High
Full-year 2026 AI fabric revenue
$3.5B
high materiality
High
Full-year 2026 campus revenue
$1.25B
medium materiality
High
Full-year 2026 gross margin (non-GAAP)
62% to 64%
high materiality
Medium
Full-year 2026 operating margin (non-GAAP)
approximately 46%
high materiality
High
Full-year 2026 effective tax rate
21.5%
low materiality
High
Q2 2026 revenue
approximately $2.8B
high materiality
High
Q2 2026 gross margin (non-GAAP)
62% to 63%
medium materiality
High
Q2 2026 operating margin (non-GAAP)
46% to 47%
medium materiality
High
Q2 2026 diluted EPS (non-GAAP)
approximately $0.88
high materiality
High
Q2 2026 effective tax rate
approximately 21.5%
low materiality
High
2026 Santa Clara facility CapEx (single-number capex guide)
approximately $180M in 2026
low materiality
Medium
Scale-across contribution to 2026 AI revenue
at least one-third of the AI number
high materiality
Medium
New >10% revenue customers in 2026
at least one, maybe two new 10%+ customers
medium materiality
Low

Segment performance

SegmentRevenueYoYQoQMargin
International (geography)
QoQ decline in international share primarily due to Americas-based sales to large global customers.
Share of total revenue: 15.5% (down from 21.2% prior quarter)
$418.9M———
Americas (geography)
Not stated directly by management; derived from total revenue of $2.71B less international $418.9M. Americas rose in mix on large global-customer sales.
Share of total revenue: ~84.5% (derived as 100% minus 15.5% international)
~$2.29B (derived: 84.5% of total)———

Product announcements

ProductTypeDetails
XPO (Extended Pluggable Optics)launch
Ethernet Scale-Up (ESUN) / AI racksroadmap
1.6 Terabit Ethernetroadmap

Deals & partnerships

VeloCloud acquisition

Acquisition integrating well into branch and campus strategy; brings distributed-enterprise use cases and a new channel motion with managed service providers (MSPs).

Palo Alto Networks technology partnership

Technology partnership cited in the fiber-to-the-home service-provider win, ensuring the routing edge integrated securely with the overarching security architecture.

AMD ecosystem/technology partnership

Arista AI leaf-spine EtherLink deployed at 800G to connect AMD MI-series XPUs in a strategic neocloud win; AMD accelerators cited as part of diverse-accelerator support.

Capital programs

Santa Clara facility expansion underway
Period spend: ~$180M estimated in 2026
Spent to date: ~$40M incurred in Q1 FY26
Start: Under construction (ongoing)

Benefit:Expanded facilities in Santa Clara

Q1 total capex was $54.5M, of which ~$40M related to this facility program; full-year 2026 program spend estimated at ~$180M.

Risks & headwinds

Industry-wide component supply shortages and elevated procurement costs (wafers, silicon chips, CPUs, optics, memory) Now framed as a 1-2 year phenomenon (into 2027)

~52-week lead times reliably, with reservations needed beyond; purchase commitments raised to $8.9B from $6.8B

Mitigation:Multiyear purchase commitments and supply agreements; bringing on secondary providers; qualifying new components; strong spending power as a reliable partner

Gross-margin pressure from adverse customer mix and supply-cost inflation FY26

Gross margin 62.4%, down ~170 bps YoY; full-year held at 62-64%

Mitigation:Sourcing-team execution; manufacturing cost control; only modest price increases; potential H2 mix improvement

Supplier de-commits reducing available supply Ongoing into next year

Not quantified ('the amount of de-commits we're seeing doesn't feel good')

Mitigation:Multi-year forecasting and reservations to secure chip allocation

Deferred-revenue volatility from extended customer-acceptance/qualification cycles Ongoing; can create quarterly/sequential dynamics that do not follow prior-year trends

Total deferred revenue $6.2B (up from $5.37B); product deferred up ~$643M QoQ; acceptance cycles stretched from 2-4 to 6-8 quarters

Mitigation:Recognized every quarter as items flow in and out; Arista supports customer readiness and product qualification

Customer concentration Ongoing

Only 2 customers >10% of revenue (Microsoft, Meta), each for over a decade

Mitigation:Diversification into neocloud/sovereign, enterprise/campus, service provider; expecting 1-2 new 10% customers

Ethernet scale-up not production-ready until 2027-28 while ODM competitors announce rack wins now 2027-28

5-7 scale-up rack opportunities in engineering; scale-up virtually zero in 2026

Mitigation:Active engineering with customers; adherence to ESUN spec and higher production bar than ODMs

Q&A highlights

How much did scale-across contribute last year and how material is it to the $3.5B AI forecast and long term?

Last year scale-across was small (majority was scale-out heritage); scale-up is virtually zero in 2026. This year the AI number is shared between scale-out and scale-across, with scale-across contributing at least one-third.

“scale across will definitely contribute at least 1/3 of our AI number”

asked by Simon Leopold · answered by Jayshree Ullal

4 min read 7 chapters

Detailed narrative

Record demand collides with a multi-year supply squeeze

Management called Q1 demand the best in Jayshree Ullal's Arista tenure, but framed supply as the opposite tail — industry-wide shortages across wafers, silicon chips, CPUs, optics and memory, at elevated procurement cost. Ken Duda cited ~52-week lead times "pretty reliably with reservation needs beyond that." What began as a memory problem is now all wafer-fabrication facilities, and the constraint is reframed as a 1-2 year, not 1-2 quarter, phenomenon.

Arista is signing multiyear purchase commitments (now $8.9B) and paying up to assure continuity so customers' GPUs and AI infrastructure are not left idle for lack of network.

AI fabric strategy: scale-up, scale-out, and scale-across

Arista framed three AI fabric use cases. Scale-out (its heritage) has >100 cumulative 800GbE customers, with 1.6T at production scale expected in 2027. Scale-across — distributing AI across data centers for power/bandwidth using the flagship 7800R3/R4 routing platforms — is now expected to contribute at least one-third of the 2026 AI number and has become a bigger use case than management imagined a year ago.

Scale-up (Ethernet, via the ESUN spec) is virtually zero in 2026 and a 2027-28 story, with 5-7 rack opportunities in active engineering. The fourth of the 2024 Ethernet AI training deployments has officially migrated from InfiniBand to Ethernet at production scale.

Enterprise and campus diversification with four flagship wins

Management detailed four wins unified by the EOS stack: a neocloud AI network (800GbE EtherLink connecting AMD MI-series XPUs, moving off a white-box architecture), a regional fiber-to-the-home service provider (7280 routing with EOS FLX, integrated with Palo Alto Networks security), an insurance-services observability win (R3 series, DMS monitoring fabric), and a manufacturer with 100+ global factory sites (universal leaf-spine Cognitive Campus, 100-gig campus spine, WiFi 7, CloudVision).

The VeloCloud acquisition is integrating into the branch/campus strategy, adding an MSP channel motion. NPS improved from 87 to 89 (94% approval).

XPO optics innovation unveiled at OFC

Arista unveiled its extended pluggable optics (XPO) form factor at the Optical Fiber Conference, now endorsed by >100 vendors. Salient specs: 12.8 terabits per pluggable module, 204.8 terabits per OCP rack unit, integrated cold plate cooling up to 400 watts per module, and flexibility across copper and linear/retimed interfaces.

Management positioned XPO as a partner to OSFP — OSFP remains fine at 400G/800G, with XPO the connector of choice at 1.6T/3.2T where liquid cooling and density are required — and expects a 10-year run, particularly for scale-out and scale-across.

Q1 financial results and margin dynamics

Revenue of $2.71B rose 35.1% YoY and beat the $2.6B guide. Non-GAAP gross margin was 62.4% (within the 62-63% guide, down ~170 bps YoY), primarily on customer mix — larger customers carry lower gross-margin accretion — with memory/silicon cost and tariffs a secondary drag.

Non-GAAP operating margin was 47.8%; net income $1.11B (40.9% of revenue); diluted EPS $0.87 (+31.8%). Opex fell to 14.6% of revenue ($396.8M), with R&D $271.5M (10%), S&M $103.5M (3.8%) and G&A $21.8M (0.8%). Operating cash flow of ~$1.69B was a record.

International revenue was 15.5% of the total, down from 21.2%, on Americas-based sales to large global customers.

Deferred revenue and purchase-commitment build

Total deferred revenue rose to $6.2B from $5.37B, with product deferred up ~$643M QoQ; the majority is product-related and increasingly volatile due to customer-specific acceptance clauses. Ullal explained qualification/acceptance cycles have stretched from 2-4 quarters to 6-8 quarters as customers ready facilities, install cables manually, and qualify brand-new EtherLink chips and software.

Chantelle Breithaupt stressed the balance is not simply aging — revenue is recognized every quarter as items flow in and out. Purchase commitments jumped to $8.9B (from $6.8B), mostly chips for new products and AI; inventory rose to $2.38B with turns improving to 1.7 from 1.5.

Capital allocation and guidance raise

No stock was repurchased in Q1; $817.9M remains of the $1.5B program approved May 2025, with timing dependent on conditions. FY26 guidance was raised to 27.7% growth (~$11.5B) and the AI-fabric goal to $3.5B, while campus ($1.25B), full-year gross margin (62-64%) and operating margin (~46%) were held.

Management noted price increases have been modest — unlike competitors, no double or major hikes — and won't flow through until backlog reduces, keeping the customer partnership intact while Arista absorbs cost.

Summary of the company's earnings call; see how call summaries are prepared. Not investment advice.