On August 4, 2026, Arista Networks reported second-quarter results that landed a milestone the company had been approaching for several quarters: its first-ever $3 billion quarter. Revenue came in at $3.036 billion, beating consensus estimates of $2.83 billion by more than 7%. Non-GAAP EPS of $1.02 topped expectations by 14 cents. But the numbers, impressive as they are, were only the starting point of an earnings call that revealed far more about where AI networking is heading — and why Arista is the company most likely to capture the lion's share of it.

The Headline Numbers

Q2 Revenue
$3.04B
+37.7% YoY
Non-GAAP EPS
$1.02
Beat by $0.14
Operating Margin
49.9%
vs 48.8% last year
Product Revenue
$2.61B
Bulk of results
Metric Q2 2026 Actual Consensus Est. vs Estimate
Revenue $3.036B $2.83B +7.2% beat
Non-GAAP EPS $1.02 $0.89 +$0.14 beat
Non-GAAP Op. Margin 49.9% ~48% Expansion
Q3 2026 Revenue Guide ~$3.30B ~$3.0B +10% above est.
FY 2026 Revenue Guide ~$12.6B ~$11.5B Raised $1.1B

The full-year guidance raise of $1.1 billion above the May outlook — to approximately $12.6 billion representing 40% annual growth — is the number that stopped analysts in their tracks. Arista is not guiding for a soft landing from AI-driven demand. It is guiding for acceleration.

The Three Tiers of AI Networking — And Where Arista Sits

The most instructive part of the Q2 call was management's framework for thinking about AI networking. The market tends to lump all AI networking together, but Arista drew a clear distinction between three fundamentally different layers — each with different competitive dynamics and different TAMs.

Tier 1 — Scale-Up
Within the Rack
Connects GPUs within a single server rack. Currently dominated by Nvidia's NVLink proprietary interconnect. Arista has limited presence here by design — NVLink's performance advantage within-rack is real and Arista does not compete where it cannot win.
Arista stance: Limited / Pass
Tier 2 — Scale-Out
Within the Cluster
Connects GPU racks within a single data centre cluster. This is Arista's core strength — EtherLynx switches provide high-bandwidth, low-latency Ethernet fabric that directly competes with InfiniBand at the cluster level. Strong and growing presence.
Arista stance: ✓ Strong — core market
Tier 3 — Scale-Across
Across Data Centres
Connects AI clusters across geographies and multiple data centres. A rapidly emerging TAM. Arista is well-positioned via its WAN and routing capabilities. Management projects this market alone at $15–20B by 2030; Arista is already capturing ~$1.2B of the current $3–4B opportunity.
Arista stance: ✓✓ Fast-growing — $15-20B by 2030

This three-tier framework matters because it explains why Arista's growth story has a longer runway than most investors assume. The scale-across tier is just beginning. As AI models grow larger and training/inference workloads spread across multiple physical locations, the demand for high-performance inter-datacenter networking will compound. Arista, with its EOS software stack and routing heritage, is structurally positioned to capture the majority of that spend.

EtherLynx: From 4–5 Customers to 100+

The single most striking data point on the call was the EtherLynx AI fabric customer count. In 2024, Arista had 4 to 5 cumulative EtherLynx customers. As of the Q2 2026 call, that number has crossed 100+. This is not incremental growth — it is a category shift. EtherLynx has gone from a pilot program to a standard deployment option for AI infrastructure teams across hyperscalers, cloud providers, and AI labs.

Customer Growth Context

Going from 4–5 to 100+ EtherLynx customers in roughly 18 months implies a roughly 20–25× increase in the installed base. This is the kind of adoption curve that typically precedes a TAM expansion — as the customer base scales, so does Arista's ability to upsell additional switching tiers, software subscriptions, and support contracts.

Management also flagged a potential new 10%+ revenue customer — almost certainly a large AI lab — which would join Microsoft and Meta as hyperscale-tier accounts. Arista expects 1 to 2 additional 10%+ customers in the near future. For a company at Arista's scale, adding another hyperscale-tier account is not a marginal event — it resets the baseline for the next several years of revenue.

EOS Technical Innovations: Three Features That Matter

Arista's Extensible Operating System (EOS) is the real competitive moat — not the hardware. The Q2 call highlighted three specific EOS capabilities that are directly relevant to AI workloads:

SSU
Smart System Upgrade — Zero-Downtime Updates
SSU allows Arista switches to be upgraded with no traffic interruption. In AI training environments where a GPU cluster may be running a 30-day training job, even a 30-second maintenance window is unacceptable. SSU eliminates this constraint entirely — a critical differentiator for security patch compliance in regulated AI deployments.
MRC
Multipath Reliable Connection — GPU Utilisation Improvement
MRC implements intelligent packet spraying across multiple network paths simultaneously. The practical effect is dramatic: GPU utilisation rates improve significantly because network congestion — the primary cause of GPU idle time in distributed training — is reduced. In a cluster where GPUs cost $30,000+ each, even a 5% improvement in utilisation compounds to meaningful cost savings.
SRv6
Segment Routing v6 — Real-Time Congestion-Aware Path Selection
SRv6 enables real-time, dynamic routing decisions based on current network congestion. Rather than committing traffic to a fixed path, SRv6 continuously re-evaluates the optimal route across the fabric. This is particularly valuable for scale-across deployments where traffic must traverse multiple data centre interconnects with variable latency.

Supply Chain: Tripling the Bet to $9.7 Billion

Perhaps the most consequential disclosure on the call — and one that received less attention than the revenue numbers — was the supply chain update. Arista has tripled its purchase commitments to $9.7 billion, up from approximately $3.6 billion a year ago. This is a massive forward bet on demand continuity.

Purchase Commitments
$9.7B
vs ~$3.6B a year ago
Contract Manufacturers
3
US · Asia · Mexico
Memory Supply Secured
Through 2026
Visibility into 2027

The decision to triple purchase commitments is not made lightly. It locks Arista into significant cash outflows regardless of whether demand materialises. But it also signals something important: management has visibility into orders and customer commitments that gives them sufficient confidence to make this bet. Companies do not pre-commit $9.7 billion on a hunch.

The three-geography manufacturing footprint — US, Asia, and Mexico — is also notable from a geopolitical risk management perspective. As tariff risks and supply chain disruptions remain a persistent concern for technology hardware companies, Arista has deliberately distributed its manufacturing to avoid single-region concentration risk.

New Product: 7060X E7 — 100Tb Capacity

Arista announced the 7060X E7, a next-generation switch delivering 100 terabits of capacity with 1.6 terabit-per-second throughput. This positions Arista at the frontier of high-density AI spine switching. Equally significant: the 7060X E7 is the first Arista platform to offer liquid cooling options — a recognition that as switch densities and power consumption climb, air cooling is approaching its physical limits in AI data centres.

Enterprise and Campus: The Overlooked Second Engine

Buried beneath the AI headlines is a strengthening enterprise and campus business. Arista raised its campus revenue target to ≥$1.25 billion for full-year 2026 — a segment that was barely mentioned in analyst models two years ago. As enterprises modernise their campus networks to support AI-enabled applications, video conferencing at scale, and IoT device proliferation, Arista's EOS-based campus switches are winning deals that previously went to Cisco by default.

This matters for the investment thesis because it means Arista's revenue base is not purely dependent on hyperscaler capex cycles. Enterprise and campus provide a more stable, less lumpy revenue stream that smooths out the quarterly volatility inherent in large-project-driven hyperscaler spending.

Q3 and Full-Year Guidance

For Q3 2026, Arista guided revenue of approximately $3.30 billion — comfortably above the $3.0 billion analyst consensus — with non-GAAP EPS in the range of $1.06 to $1.08. Full-year guidance was raised to approximately $12.6 billion, implying 40% growth for the year, with gross margins of 62–64% and operating margins of 48–49%.

Guidance Raise in Context

Raising full-year guidance by $1.1 billion after one quarter is extraordinary at Arista's scale. It implies Q3 and Q4 combined must deliver approximately $9.56 billion — roughly $4.78 billion per quarter on average. The Q3 guide of $3.30B means Q4 would need to come in around $6.26B to hit the midpoint, or more likely, the full-year number proves conservative and gets raised again in November.

Bull vs Bear: ANET at Current Levels

▲ Bull Case
  • EtherLynx customer count 20×+ in 18 months — adoption is in the exponential phase, not the plateau phase
  • Scale-across TAM of $15–20B by 2030 is barely penetrated — Arista holds ~$1.2B of a $3–4B current market
  • $9.7B in purchase commitments signals management's conviction in multi-year demand visibility
  • New 10%+ customer (AI lab) about to materialise — would reset revenue baseline upward for years
  • 49.9% operating margins expanding — revenue growth dropping efficiently to the bottom line
  • Campus business at $1.25B+ provides stability against hyperscaler cycle volatility
  • EOS software moat — SSU, MRC, SRv6 are years ahead of competitive alternatives
▼ Bear Case
  • Valuation is elevated — ANET trades at a significant premium to peers, embedding years of continued execution
  • NVLink dominates scale-up networking — if Nvidia extends NVLink deeper into scale-out, Arista's TAM compresses
  • $9.7B in purchase commitments creates downside if AI capex spending pulls back in 2027
  • Microsoft and Meta concentration risk — losing or slowing with either hyperscaler would materially impact revenue
  • InfiniBand (Nvidia/Mellanox) remains a viable alternative in some AI cluster configurations
  • 40% revenue growth at $12.6B scale is impressive but harder to sustain — any deceleration will compress the multiple sharply

The Bottom Line

Arista's Q2 2026 earnings call was not just a financial update — it was a detailed product and strategy briefing that revealed how clearly management understands the evolving topology of AI networking. The three-tier framework (scale-up / scale-out / scale-across), the EOS software innovations, the 100+ EtherLynx customers, and the $9.7 billion supply chain commitment all tell the same story: Arista is not chasing AI networking demand — it is shaping it.

The scale-across opportunity — connecting AI clusters across geographies — may prove to be the defining growth vector for the rest of the decade. Arista is already capturing $1.2 billion of a $3–4 billion current market that management projects will reach $15–20 billion by 2030. If that projection proves correct, ANET's revenue trajectory has significantly more room to run than current consensus models imply.

Analyst Reaction

Multiple Wall Street analysts raised their price targets on ANET following the Q2 print, citing the guidance raise, EtherLynx momentum, and supply chain investments as evidence that Arista's AI networking thesis is entering a new phase of validation. The stock remains a consensus overweight across major brokerages covering the networking sector.