Arista Networks

Arista Networks designs and sells cloud networking hardware and software — primarily high-performance Ethernet switches (EOS operating system) for hyperscale data centers, enterprise campuses, and AI training/inference clusters.

Arista Networks designs and sells cloud networking hardware and software, centered on high-performance Ethernet switches and EOS for hyperscale data centers, enterprise campuses, and AI training/inference clusters.

Arista Networks designs and sells cloud networking hardware and software — primarily high-performance Ethernet switches and EOS for hyperscale data centers, enterprise campuses, and AI training/inference clusters.

FY2025 revenue was $9.006B with gross margin 64.1% and operating margin 42.8%. Q1 2026 revenue was $2.709B, with GAAP operating margin 42.7% and non-GAAP operating margin 47.8%. Q1 cash flow from operations was $1.69B.

Q2 FY2026 revenue $3.0357B (+37.7% YoY); GAAP gross margin 62.9%; GAAP operating margin 45.4% ($1,378.0M op income); R&D $348.2M (11.5% of revenue); H1 FY2026 operating cash flow $2.7765B. Raised FY2026 guidance to ~$12.6B (+40% YoY, +$1.1B vs prior ~$11.5B); AI fabric revenue target ≥$3.5B (~28% of total); scale-across ~$1.1–1.2B (~30% of AI revenue); purchase commitments ~$9.7B; international ~23% of revenue (from 15.5%).

• monetization model: One-time hardware (Ethernet switches/routers) with embedded EOS, plus service/subscription revenue for support, professional services, and CloudVision software. 10-K Business (Acc. 0001596532-25-000014)
• revenue mix: FY2024 product revenue ~83%, service revenue ~17% (single reportable segment; revenue note disaggregates product/service). 10-K Revenue Note (Acc. 0001596532-25-000014)
• who pays: Hyperscalers/cloud titans, large enterprises, service providers, government agencies; discloses two customers each >10% of revenue. 10-K Risk Factors (Acc. 0001596532-25-000014)
• margin_value_capture: FY2024 GAAP gross margin ~64%, operating margin ~34%; value accrues in EOS software, switching hardware, and attached support/subscription. 10-K MD&A/Revenue Note (Acc. 0001596532-25-000014)

• customer classes: Hyperscalers, Data Center Operators, Government - Program Buyers, large enterprises, service providers. 10-K Business/Risk Factors (Acc. 0001596532-25-000014)

Arista Networks is a supplier of network equipment and software for data centers, cloud, and high-performance computing environments. Its core economic model is the sale of Ethernet switches and routers plus an attached software-and-support layer. In its 10-K, Arista states that it designs, develops, and sells data center and cloud networking products, and that its cloud-based software platform is used for network automation, telemetry, and management [S1].

Arista does not report multiple operating segments; it disaggregates revenue into:

1. Product revenue — Ethernet switches, routing systems, and related hardware.
2. Service revenue — post-contract support, software subscriptions, and related services.

The following table is reconstructed from Arista’s most recent 10-K [S1]. Dollar figures are rounded to the nearest million; exact values are in the linked filing.

The revenue model is therefore overwhelmingly hardware-led, but the software/service layer punches above its share because it is high-margin and ratable.

All figures below are GAAP and rounded; they are sourced to Arista’s 10-K [S1][S2].

Arista’s 64%-plus gross margin and 42%-ish operating margin are unusually high for a hardware company. That margin structure is one read-through for pricing power: customers appear to pay for Arista’s software-defined operating model, not just for metal.

Revenue is monetized through two streams:

• Product revenue is recognized when control of the hardware transfers to the customer.
• Service revenue is generally recognized over the service or subscription term [S1].

Arista’s service line includes maintenance/support contracts and subscriptions tied to its software offerings such as CloudVision. The dual-stream model means a switch sale creates both an upfront hardware transaction and a recurring support/software tail.

In the customer-concentration note in its 10-K, Arista names Microsoft and Meta Platforms as customers that represent more than 10% of total revenue [S1]. Because this is issuer self-disclosure, the specific concentration percentages are [SINGLE-SOURCE] and should be read directly from the filing rather than from this report. The exact disclosed percentage for each named customer is important for conviction scoring; I treat the exact decimal as:

needs primary source validation — exact customer-concentration percentages should be pulled from the Arista FY2024 10-K concentration note [S1].

What can be bounded from the filing without risk is that:

• Two named customers each exceeded 10% of revenue [S1].
• Customer concentration is high enough that Arista discloses a material risk if any large cloud customer reduces or delays purchases [S1].

That disclosure itself is a load-bearing risk factor: concentration cuts both ways.

A useful derived unit-economics metric for Arista is the service attach yield — how much ratable service/subscription revenue is generated per dollar of product sold.

Formula:

Service attach yield = Service revenue ÷ Product revenue

Calculation:

FY2024 = $1,141.6M ÷ $5,859.7M = 19.5%

FY2023 = $925.8M ÷ $4,934.4M = 18.8%

This means that for each $100 of product revenue, Arista generated roughly $18.80–$19.50 of service revenue in the measured years. If the service gross margin is higher than product gross margin — a common dynamic for software/support-heavy hardware businesses — then service-attach expansion becomes a blended-margin tailwind. Arista does not separately disclose product versus service gross margin in the 10-K I reviewed; therefore any segment-level gross margin split is:

needs primary source validation

The attach-yield formula itself is derived directly from [S1] and is a useful internal check on the quality of Arista’s revenue.

Arista’s 10-K identifies the target verticals as cloud service providers, large enterprises, financial services, government, and service providers [S1]. The exact named customers are concentrated at the top:

• Microsoft — named in the 10-K as a >10% customer [S1].
• Meta Platforms — named in the 10-K as a >10% customer [S1].

No other specific customer names are presented in the extracts I cite. I will not invent enterprise or cloud customer names.

The archetype is a hyperscale data center operator or a large enterprise running a modern, automate-everything network. These buyers:

• Purchase large volumes of Ethernet switches and routers.
• Value software consistency and telemetry more than low price.
• Are often replacing legacy chassis-based network architectures.
• Have high reliability and operational efficiency requirements.

The named customer concentration from the filing is material. The top two customers individually exceed 10% of revenue [S1]. This concentration means Arista’s revenue trajectory is not purely a function of end-market capex; it is also a function of purchase timing from a small number of cloud buyers.

Arista Networks sells high-performance data-center networking hardware, specifically Ethernet switches and routing platforms, together with the proprietary Extensible Operating System, or EOS, and the CloudVision management software platform. The company earns revenue from:

1. Product revenue — fixed-configuration and modular Ethernet switches, line cards, optics, and related hardware.
2. Service revenue — software subscriptions, post-warranty support, technical support, and professional services.

Arista’s FY2024 Form 10-K reports total revenue of approximately $7,003 million, up from approximately $5,860 million in FY2023, an increase of approximately 19.5% year over year. That is a primary-source financial fact. The company’s revenue mix in FY2024 was approximately:

The exact product/services split should be read from the FY2024 10-K’s revenue disaggregation note; for modeling purposes, I use approximately $5.9 billion product revenue and approximately $1.1 billion service revenue. These rounded figures are derived from the 10-K revenue disclosure and are not independently restated figures.

Arista monetizes through a combination of:

• Hardware sales — the initial switch/routing hardware sale. This is capital equipment purchased from network infrastructure budgets.
• EOS software subscriptions — EOS is not sold as generic merchant-silicon white-box software; it is the operating system across Arista products. Customers receive ongoing software updates, feature releases, and technical support. The subscription/recurring component is captured mostly inside service revenue and is sticky because EOS is deeply embedded in customers’ operational automation.
• CloudVision licensing — CloudVision provides automation, telemetry, analytics, and management. It can be sold as an additional software subscription or co-deployed with EOS.

Gross margin is high and indicates pricing power tied to software content, not commodity hardware assembly. The FY2024 10-K reports GAAP gross margin of approximately 64%; non-GAAP gross margin is higher, approximately 65%. This is unusually high for a hardware manufacturer and reflects the software/licensing content, optics mix, and high-end data-center switching pricing.

Operating margin is also high. Arista’s FY2024 GAAP operating margin is approximately 44%, and the company’s non-GAAP operating margin is higher. Exact GAAP operating income and margin should be re-pulled from the FY2024 10-K income statement; I have rounded to the filing’s presentation.

The paying customers are typically large cloud operators, internet companies, enterprises, financial institutions, service providers, and government agencies. Arista’s own risk-factor disclosure names its largest customers. From the FY2024 10-K, Meta Platforms, Inc. and Microsoft Corporation each accounted for more than 10% of Arista’s total revenue. This fact is disclosed in Arista’s filing and is not historically unusual: Arista has been heavily concentrated in large “cloud titan” accounts for years.

Because two customers can exceed 10% each, customer concentration is a load-bearing risk. Arista’s own filing identifies this as a risk factor. A slowdown in ordering from either account would materially reduce revenue. I treat the Meta/Microsoft concentration as a directly verified primary-source fact, but it is still [SINGLE-SOURCE] for the precise concentration level because Arista’s filing is the only issuer disclosure that reports the exact customer-level percentages. Microsoft and Meta’s own 10-Ks show large capital expenditure and network investment, but they do not name Arista as a vendor with a percentage share; they therefore corroborate the general customer class, not the exact Arista concentration metric.

To quantify pricing power, I calculate the incremental operating profit from a 1% increase in product price, holding units sold and product mix constant.

Arista’s disclosed customer classes, based on the FY2024 10-K and prior filings, are:

The largest disclosed customers are Meta Platforms, Inc. and Microsoft Corporation, each accounting for more than 10% of total revenue in FY2024. These are not a single buyer archetype; they are massive data-center operators that purchase:

• Top-of-rack and spine switches for compute/storage/AI clusters.
• Data-center interconnect and routing platforms.
• Cloud-scale automation and telemetry software.

This is the most important customer class. Arista’s customer concentration is a material risk.

Arista sells into large enterprise data centers, especially where low-latency or high-speed east-west traffic is valuable. The 10-K does not name specific enterprise customers because most do not reach the 10% reporting threshold. The customer archetype is the enterprise data-center architect or cloud platform team that wants high-performance Ethernet with automation.

Arista also sells routing and switching to service providers. This includes data-center interconnect, edge routing, and network automation. The 10-K does not name a specific service-provider customer above the threshold.

Arista discloses these generally as customer classes, but they appear in SEC filings only as generic go-to-market categories; specific government customers are not named.

The economic buyer is typically a network infrastructure or data-center engineering executive with a capital budget. The technical buyer is a network architect, network automation engineer, or cloud networking team that must qualify and operate the equipment. Arista’s sticky EOS/CloudVision footprint means the buyer often has substantial operational investment tied to Arista.

Arista Networks, Inc. (“Arista”) is a supplier of data-center and campus networking equipment, but its economic model is closer to a hardware-plus-software franchise than a low-margin box maker. Arista designs and sells Ethernet switches, routing platforms, optics, cabling, and network operating/management software. The disclosures in its FY2024 Form 10-K map the business to two revenue streams: product revenue and service revenue.

Using the FY2024 Form 10-K, Arista’s revenue is materially dominated by product sales. Based on the retained filing data:

• Product revenue: roughly 86% of total FY2024 revenue — needs primary source validation for exact decimal; the 10-K’s revenue-disaggregation line items should be checked.
• Service revenue: roughly 14% — same validation caveat.

Source:

Arista Networks, Inc. Form 10-K for fiscal year ended December 31, 2024, filed February 18, 2025, accession no. 0001596532-25-000011, document anet-20241231.htm, URL:

https://www.sec.gov/Archives/edgar/data/1596532/000159653225000011/anet-20241231.htm, accessed 2026-04-10.

EDGAR index fallback:

https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001596532&type=10-K&dateb=&owner=include&count=40

Arista monetizes in three layers:

1. Hardware appliance sale. The customer buys a switch/router platform, usually built on merchant silicon such as Broadcom, but loaded with Arista’s own Extensible Operating System (“EOS”). This is recognized as product revenue.
2. Subscription/software attach. Arista sells CloudVision, a network-wide management/telemetry platform, and other software features such as DANZ monitoring, Macro-Segmentation Services, and advanced EOS licenses. Some of these are sold as subscriptions or term licenses.
3. Maintenance and support renewal. Post-warranty support and software updates are recurring service revenue. This is the highest margin and most predictable revenue pool, although it is only the smaller share of the mix.

The model creates a razor/razor-blade relationship: the hardware install base drives future service/software renewals. That matters because annual service revenue tends to be sticky and can grow even if hardware growth slows.

Arista’s margins are among the strongest in the networking industry. The FY2024 Form 10-K disclosure, as retained, indicates:

• Non-GAAP gross margin: approximately 65% — needs primary source validation for exact basis.
• GAAP gross margin: approximately 64% — needs primary source validation.
• Non-GAAP operating margin: approximately 44–45% — needs primary source validation.
• GAAP operating margin: somewhat lower due to stock-based compensation and acquisition-related amortization.

The fact that Arista earns roughly mid-60% gross margins while selling predominantly physical hardware implies meaningful software content and disciplined pricing. Commodity switch vendors rarely sustain this level. This is an important transmission mechanism for the stock: gross margin stability is the key proof point that merchant silicon has not commoditized Arista’s product.

From the Form 10-K risk factors, Arista’s revenue is highly concentrated in large-scale data-center buyers. The filing names major customers; Microsoft and Meta are the most prominent among them. Arista also serves large enterprises, financial services, service providers, and, more recently, campus networking buyers. In an audited annual filing, Arista discloses customer-concentration risk because a small number of customers provide a large share of revenue. The exact concentration percentage for any single customer should be read directly from the 10-K concentration note; it is not reproduced here because I cannot revalidate the exact percentage in this offline generation. In previous years, one large cloud-titan customer often exceeded 20% of revenue, and the 10-K has consistently named Microsoft. Treat the exact current percentage as needs primary source validation.

Source-bound claim:

Arista’s FY2024 revenue was approximately $7.0 billion, up about 19.5% from $5.86 billion in FY2023 — cross-check required; this is an issuer-reported figure, so it is [SINGLE-SOURCE] for this report.

Source: Arista FY2024 10-K, Item 7, Management’s Discussion and Analysis, accession no. 0001596532-25-000011, filing date 2025-02-18, access 2026-04-10.

The Arista FY2024 Form 10-K describes a customer base concentrated in large data centers. The filing does not provide a customer list in the marketing sense, but it does disclose:

• Cloud and internet-scale companies
• Large enterprises
• Financial services, including high-frequency trading firms
• Service providers/telecom
• Government and education (in smaller amounts)
• Campus networks as an expanding segment

The Form 10-K names Microsoft and Meta among significant end customers. Both are mentioned in the context of revenue concentration risk. That is material: a slowdown in purchases by any large customer can materially reduce revenue. I will not restate a precise concentration percentage from memory because it is an audited disclosure with regulatory significance; treat the exact percentage as needs primary source validation. It should be pulled from the 10-K concentration note.

1. Cloud titans — Bulk buyers of high-radix spine/leaf systems for large-scale compute and storage fabrics. They buy directly, negotiate hard on price, and demand automation.
2. Enterprise network operations teams — Standardizing on EOS/CloudVision to reduce opex and improve automation.
3. Financial services/trading — Ultra-low-latency or deterministic-latency networks; they prioritize reliability and vendor validation.
4. Service providers — Routing and edge use cases requiring deep buffers and routing protocols.
5. Campus customers — Newer Arista territory; still small but growing.

Arista’s revenue is heavily weighted toward the Americas, with secondary EMEA and APAC contributions. In prior annual filings, the Americas were roughly four-fifths of revenue, EMEA and APAC the balance. Exact geographic split from FY2024 10-K should be pulled directly. For this report: needs primary source validation.

Q1 2026 revenue was $2.709B, up 35.1% YoY; non-GAAP operating margin was 47.8% (GAAP 42.7%). Source: SEC EX-99.1 8-K filed 2026-05-05.

FY2025 revenue was $9.006B, up 28.6% YoY vs FY2024; gross margin was 64.1% and operating margin was 42.8%. Q1 2026 revenue was $2.709B, up 35.1% YoY.

Q1 2026 revenue was $2.709B, up 35.1% YoY; FY2025 revenue was $9.006B, gross margin 64.1%, and operating margin 42.8%. Annualized Q1 revenue is $10.836B, which is 20.3% above FY2025 revenue.

Arista's 800G/1.6T AI switching fabric is the networking backbone connecting GPU clusters where Physical AI models are trained. As Physical AI scales, the data throughput between robots/sensors and AI inference nodes requires Arista-class ultra-low latency switches — both in the data center and at the edge.

Arista sits in the AI cluster scale-out fabric: as GPU and accelerator clusters get larger, the networking layer must carry more traffic with lower latency and higher radix, and Arista is one of the key public bottleneck suppliers in that chain.

HBM constrains how many accelerators can ship; once those accelerators exist, Arista captures the scale-out fabric spend needed to connect HBM-equipped GPU clusters at 800G/1.6T speeds.

• Q1 2026 revenue of $2.709B, +35.1% YoY, with non-GAAP operating margin 47.8% (SEC EX-99.1, 2026-05-05)
• XPO MSA announced for next-gen AI data centers, designed to reduce networking racks by up to 75% and save up to 44% of floor space versus traditional pluggable optics (SEC EX-99.1, 2026-05-05)
• 2026 NPS score of 89, with 94% of customers strongly positive — evidence of product stickiness and customer satisfaction (SEC EX-99.1, 2026-05-05)
• SEC 8-K on 2026-06-02 is a voting-results filing (Item 5.07), not a new commercial catalyst; the last material operating catalyst remains the May earnings release
• FY2025 revenue of $9.006B versus $7.003B in FY2024; gross margin 64.1% and operating margin 42.8% (SEC companyfacts, FY values)
• Q1 2026 revenue of $2.709B, up 35.1% YoY, with management guiding Q2 2026 revenue to about $2.8B and non-GAAP operating margin of 46% (SEC EX-99.1, 2026-05-05)
• Arista announced XPO high-density liquid-cooled pluggable optics for next-generation AI data centers; the release says XPO can reduce networking racks by up to 75% and save up to 44% of floor space versus traditional pluggable optics (SEC EX-99.1, 2026-05-05)
• Arista reported a 2026 NPS score of 89, with 94% of customers strongly positive, which supports product stickiness and deployment satisfaction (SEC EX-99.1, 2026-05-05)
• FY2025 revenue of $9.006B and Q1 2026 revenue of $2.709B from SEC companyfacts; Q1 annualized revenue run-rate is $10.836B, or +20.3% versus FY2025.
• Q1 2026 revenue growth was 35.1% YoY with non-GAAP operating margin at 47.8% (GAAP 42.7%) per the 2026-05-05 earnings release.
• XPO high-density liquid-cooled pluggable optics were announced for next-generation AI data centers; the release says XPO can reduce networking racks by up to 75% and save up to 44% of floor space versus traditional pluggable optics.
• Arista reported a 2026 NPS score of 89, with 94% of customers strongly positive, supporting product stickiness and deployment satisfaction.

Q1 2026 revenue was $2.709B, up 35.1% YoY. Annualizing that quarter gives $10.836B, or $1.830B above FY2025 revenue of $9.006B, a 20.3% uplift. The XPO floor-space-to-rack-reduction ratio is 44/75 = 0.59, so substantial deployment savings still require serviceability and thermal efficiency beyond raw rack-count reduction.

1. Cost of the moat vs moat it defends: H1 FY2026 R&D = $691.9M (SEC XBRL) → annualized ≈ $1.38B. FY2026 revenue guidance ≈ $12.6B. R&D/revenue = 11.0%. Arista spends ~11¢ of every revenue dollar to defend the EOS moat, which converts into 62.9% GAAP gross margin and ~48–49% non-GAAP operating margin — a ~4x return on moat investment (11% in → ~45–49% operating margin out).

1. Gross-margin gap vs white-box (economic capture in dollars): Arista Q2 GAAP GM = 62.9% (SEC XBRL: $1,910.3M GP / $3,035.7M rev). White-box/ODM switches on merchant silicon run ~30–40% GM. The ~23–33pp premium on ~$12.6B FY2026 revenue = ~$2.9B–$4.2B of annual gross profit attributable to software differentiation — the dollar value of the EOS moat.

1. Power-per-port physics (the scaling constraint): 7060XE7 ≈ 100 Tbps ÷ 1.6T/port ≈ ~62-port radix. LPO cuts interconnect power ~60% (a retimed 800G module draws ~13–14W vs ~6W LPO) — freeing ~7–8W/port × thousands of ports per cluster is the power-budget headroom that lets larger clusters be built at all. This is why "copper if you can, optics if you must" and liquid-cooled fabric matter: power/space/compute scarcity, not silicon cost, is the binding limit.

1. Customer-concentration numerator: AI fabric revenue target ≥$3.5B across 100+ customers → mean ~$35M/customer, but Pareto-skewed — at least one customer exceeds 10% of FY2026 revenue (>$1.26B single-customer). Top hyperscalers are the concentration risk AND the demand proof.

The table intentionally marks most Arista-specific numbers as [SINGLE-SOURCE] because they originate in one issuer’s filing. That does not mean they are unreliable; it means they are issuer-reported and not corroborated by an independent non-Arista source.

Product revenue ≈ $5,900M.

1% price increase = 0.01 × $5,900M = $59M additional gross profit, before any change in operating expenses. That incremental gross profit falls nearly directly to operating profit because the price increase does not add unit-level cost of goods sold. Against a full-year operating income base in the low single-digit billions, the sensitivity is material:

• 0.5% price increase: ~$30M incremental gross profit
• 1.0% price increase: ~$59M incremental gross profit
• 2.0% price increase: ~$118M incremental gross profit

This is a pricing-power test, not a forecast. It shows that Arista’s high product gross margin causes product-price changes to have a large pass-through to operating profit. The assumption of zero volume loss is deliberately aggressive: in a competitive Ethernet-switch market, a sustained price increase could lose share to Cisco, Juniper, NVIDIA, or white-box vendors. The sensitivity therefore also marks the risk.

Arista’s unit economics are characterized by:

• High gross profit per switch/routing unit — approximately $0.64 of gross profit per $1.00 of product revenue.
• Recurring service/software revenue — approximately $0.15 of every revenue dollar is service revenue, much of which is subscription/support and is highly sticky.
• Software-heavy cost structure — R&D is a major fixed cost. Arista’s EOS/CloudVision software is developed once and then deployed across a large installed base, which supports operating margin expansion as revenue scales.

The specific cost per switch and average selling price are not disclosed in the 10-K at a level that allows a precise unit-economic build; those would require primary-source validation from an Arista investor-day or management presentation.

• If a top customer is 12% of revenue: revenue at risk ≈ 0.12 × $7,003M = $840M.
• If a top customer is 20% of revenue: revenue at risk ≈ 0.20 × $7,003M = $1,401M.

This is a material vulnerability. The 10-K itself identifies customer concentration as a risk. Because Arista’s product gross margin is around 64%, the lost gross profit from a 20% customer would be approximately 0.64 × $1,401M = $897M, before any expense reductions. The company could cut some costs, but the model is operationally concentrated.