ANET Arista Networks

Research status: Diligence refresh completed 5 September 2026; primary-source baseline remains Q2 2026, supplemented by Q2 call details and late-August/September AI-networking read-throughs. Current market reference: $193.78 close on 4 September 2026.

Arista remains one of the highest-quality assets in AI infrastructure. Q2 revenue was $3.036B, +37.7% YoY, with 45.4% GAAP / 49.9% non-GAAP operating margin. Q3 guidance is approximately $3.3B revenue, 48–49% non-GAAP operating margin and $1.06–1.08 non-GAAP EPS. On the Q2 call, management raised the FY2026 revenue outlook to roughly $12.6B, including at least $3.5B of AI-fabric revenue and $1.25B of campus revenue, while reporting more than 100 cumulative Etherlink AI-fabric customers.

The customer-concentration diligence remains important. Arista's Q2 10-Q reiterates that two customers represented 26% and 16% of FY2025 revenue. Public product evidence validates deep relationships with Microsoft, Meta and Oracle, but current filings do not permit PowerFund to map the 26%/16% shares to named customers with confidence. FY2025 mix was approximately 48% Cloud/AI Titans, 32% Enterprise and 20% AI/Specialty Providers. Enterprise/campus diversification is real, but the economic dependence on a small number of large cloud buyers remains material.

The 7060XE7 diligence is positive but timing-sensitive. Arista's 1.6T portfolio explicitly targets both scale-out and scale-up, with announced availability beginning Q4 2026 for the air-cooled 64x1.6T platform and Q1 2027 for liquid-cooled / additional configurations. Management described second-half 2026 as a trial period and indicated that material production is still principally a 2027 event. This means rack-scale/scale-up Ethernet is a credible upside option, not yet a fully proven earnings engine.

Late-August/September industry evidence strengthens the demand backdrop: Broadcom reported $16.7B of Q3 AI semiconductor revenue and guided $21.7B for Q4; Marvell's Data Center revenue grew 46% YoY with a $3.15B Q3 company-revenue guide; NVIDIA Data Center revenue reached $89.0B, +117% YoY. These read-throughs support continued networking/interconnect intensity but also validate formidable proprietary/custom-silicon ecosystems that Arista must interoperate with and compete around.

Valuation / opportunity ranking: at $193.78 and roughly 1.279B diluted shares, ANET's equity value is about $248B, or approximately 19.7x management's ~$12.6B FY2026 revenue outlook. Using the existing PowerFund 24-month scenario framework with a ~$215 probability-weighted midpoint implies only about 5% annualized expected return from the current price. On the same unchanged internal frameworks, CLS (~29% annualized) and CRDO (~27%) currently rank well ahead for the next AI-networking dollar; MRVL (~4%) is similarly valuation-constrained. These cross-name figures are directional because each dossier's scenarios were built at different dates and should be refreshed before a trade.

Current conclusion: keep ANET at investigate / high-quality watch, not buy-now. The business quality and AI-fabric evidence are excellent, but PowerFund should not pay nearly 20x forward-year revenue for scale-up upside that is still largely in trial/early-production stages. A material de-rating, stronger 7060XE7 production evidence, or Q3/Q4 estimate revisions that move expected return into the mid-teens would justify re-ranking.

Arista's moat combines EOS, merchant-silicon system engineering, high-performance Ethernet fabrics and long-duration cloud operating relationships. The company is increasingly credible across scale-out, scale-across and emerging rack-scale/scale-up networking while enterprise/campus broadens the revenue base. The updated diligence reinforces that demand quality is strong and customer breadth is improving, but the equity case remains valuation-controlled: at nearly 20x the FY2026 revenue outlook, investors are already capitalizing substantial 2027 scale-up success. PowerFund should own ANET only when expected return compensates for large-customer concentration and the possibility that NVIDIA/custom-accelerator ecosystems capture more scale-up economics than open Ethernet.

Q3 execution versus ~$3.3B revenue and 48–49% non-GAAP operating-margin guidance; continued FY2026 AI-fabric revenue toward at least ~$3.5B; >100-customer Etherlink base converting into larger production deployments; 7060XE7 trials progressing into Q4 2026/Q1 2027 production; evidence of meaningful Ethernet scale-up adoption with non-NVIDIA accelerators; continued enterprise/campus growth toward at least ~$1.25B FY2026; software/services mix and EOS monetization; customer diversification.

Valuation at roughly 19.7x the ~$12.6B FY2026 revenue outlook; two-customer concentration of 42% of FY2025 revenue; hyperscaler deployment timing and bargaining power; $9.7B purchase-commitment/supply-chain exposure noted on the Q2 call; customer-specific acceptance clauses and deferred-revenue volatility; NVIDIA's vertically integrated networking stack; hyperscaler/custom-silicon insourcing; Broadcom merchant-silicon ecosystem power; 7060XE7/1.6T production or qualification delays; lower gross margin from large-cloud mix; factor crowding within PowerFund AI Infrastructure.

Warning — freeze additions and investigate

• Revenue growth drops below ~20% while hyperscaler/AI capex remains healthy.
• GAAP gross margin falls below ~60% for more than one quarter without a strategic mix explanation.
• GAAP operating margin falls below ~40%.
• Largest customer rises above ~30% and forward growth becomes increasingly dependent on one deployment cycle.
• 1.6T product availability or production deployment slips materially.
• Enterprise/campus growth slows enough that diversification stops progressing.

Reduce

• Revenue growth <15% for two quarters while AI networking market growth remains >20%.
• Material share loss at a top cloud customer.
• GAAP operating margin falls below ~37–38% structurally.
• Ethernet loses significant scale-up momentum to proprietary/integrated alternatives.

Invalidate

> Invalidate if Arista loses its combination of high-growth Ethernet share gains and elite operating leverage—specifically, if revenue growth falls toward low teens, a major cloud customer is lost, and operating margin structurally compresses below the high-30s without a credible new growth engine.

Additional hard invalidators: material erosion of EOS differentiation/customer retention or evidence Arista's 1.6T/rack-scale strategy is bypassed in next-generation AI architectures.

Arista's clearest advantage is the combination of EOS + system engineering + open Ethernet rather than proprietary silicon ownership. The updated evidence supports three distinct competitive arenas:

1. Scale-out: Arista is already highly credible and is benefiting from broad AI-cluster growth. More than 100 cumulative Etherlink customers materially reduces the risk that the AI business is only two hyperscaler programs.
2. Scale-across: distributed AI campuses/data centers expand the routing/switching problem and play directly to Arista's EOS and cloud-networking heritage.
3. Scale-up / rack-scale: the 7060XE7 creates a credible Ethernet option, including liquid-cooled systems, but production evidence is still early. NVIDIA remains the strongest vertically integrated alternative; custom accelerators from Microsoft/Google/others create an opening for open Ethernet but do not guarantee Arista wins the economics.

Relative to CRDO, Arista has much higher business quality/diversification but currently far lower modeled expected return. Relative to CLS, Arista has structurally better margins/moat but materially richer valuation. Relative to MRVL, Arista has cleaner GAAP economics and less acquisition/dilution complexity, while Marvell has broader silicon/optical content. For the next AI-networking dollar today, PowerFund ranks CLS/CRDO ahead of ANET, with ANET preferred to chase only after a valuation dislocation or stronger scale-up production proof.

1. Review Q3 actual revenue, gross margin and operating margin versus the ~$3.3B / ~63% gross-margin / 48–49% operating-margin framework.
2. Track FY2026 AI-fabric revenue versus the >=$3.5B goal and campus revenue versus >=$1.25B; separate AI, cloud, enterprise and software/services growth where disclosures permit.
3. Track 7060XE7 customer trials in 2H26 and actual production shipments beginning Q4 2026/Q1 2027; require named/quantified production evidence before underwriting large scale-up revenue.
4. Maintain the customer-exposure map: two FY2025 >10% customers were 26% and 16%; do not assign names without explicit disclosure. Track whether concentration falls as the >100 Etherlink-customer base scales.
5. Monitor purchase commitments, inventory and product deferred revenue; distinguish prudent supply reservation from over-ordering or acceptance delays.
6. Compare Ethernet scale-up economics and attach content versus NVIDIA Spectrum/NVLink-style integrated networking, Broadcom-enabled white-box systems and emerging UALink alternatives.
7. Rebuild ANET's 24/60-month valuation after Q3 results using the then-current price and diluted share count; require mid-teens probability-weighted annualized return before promoting to an entry candidate.
8. Re-rank ANET against CRDO, CLS and MRVL after Q3 or after a >15–20% ANET price dislocation.