Arista Networks, the U.S. cloud networking and switching giant, delivered a second-quarter earnings beat across all key metrics, with quarterly revenue surpassing $3 billion for the first time. The company sharply raised its full-year revenue guidance to approximately $12.6 billion, reinforcing the idea that the AI infrastructure investment wave is accelerating into the networking layer.
Revenue for the second quarter jumped 38% year-over-year to $3.04 billion, exceeding the $2.8 billion midpoint of the company's own guidance and Wall Street's $2.83 billion estimate. Adjusted earnings per share came in at $1.02, also well above analyst expectations. The headline figure was the upward revision of the fiscal 2026 revenue outlook to roughly $12.6 billion, an increase of $1.1 billion from the previous forecast of $11.5 billion given in May, implying roughly 40% year-over-year growth.
During the earnings call, management delivered several key signals: there is no question about demand, with the real bottleneck to revenue recognition being the supply chain; the $1.1 billion incremental guidance is not solely reliant on AI back-end networking, as front-end data centers, campus networks, and routing are all contributing; and supply chain tightness may persist through 2028, with the company securing delivery through multi-year purchase commitments and expanded manufacturing capacity.
CEO Jayshree Ullal stated in a press release, "Customers view the network as the central nervous system of their infrastructure, from the client to the campus, to the data center, and the AI center." Following the news, Arista shares surged as much as 12% in after-hours trading to $213, bringing the year-to-date gain to over 45% and pushing its market value toward the $250 billion mark.
First Quarterly Revenue Above $3 Billion, Profitability Improves
Arista is a leading provider of high-end network switches for cloud and data center environments and a primary competitor to Cisco. Second-quarter revenue of $3.04 billion, up 37.7% year-over-year, surpassed the company's guidance midpoint of $2.8 billion and the consensus estimate of $2.83 billion. This milestone is significant: the company's full-year 2021 revenue was approximately $2.9 billion, meaning a single quarter now exceeds that entire year's total.
The profitability side was equally robust. Non-GAAP gross margin was 63.4%, down from 65.6% in the same quarter last year but up from 62.4% in the previous quarter. Adjusted operating margin reached 49.9%, higher than the 47.8% in the first quarter and 48.8% a year ago. Net income was $1.3 billion, translating to a net margin of 42.9%. Diluted earnings per share were $1.02, up 39.7% year-over-year and above the analyst forecast of $0.89.
The guidance for the third quarter was also strong, with revenue expected around $3.3 billion, non-GAAP gross margin of approximately 63%, operating margin between 48% and 49%, and diluted EPS of $1.06 to $1.08. All of these figures exceeded analyst expectations, which were for revenue of $2.95 billion and EPS of $0.92.
Full-Year Guidance Raised by $1.1 Billion, Growth Sources Broader Than Expected
On the conference call, Evercore analyst Amit Daryanani directly addressed the central theme: AI networking is becoming more complex across multiple dimensions, involving not just larger clusters but also the connection of campuses and multiple data centers. Management's response confirmed this assessment and outlined a broader growth path.
Arista raised its fiscal 2026 revenue guidance to approximately $12.6 billion, $1.1 billion higher than the $11.5 billion forecast given in May, and $2.1 billion above the $10.5 billion target set during an earlier analyst day.
Management clarified that the $1.1 billion increase was not simply attributed to an explosion of AI orders. The full-year target for AI fabric is at least $3.5 billion, campus networking is targeted at a minimum of $1.25 billion, and front-end data centers, enterprise customers, and the routing adjacency market will also contribute incremental revenue.
The company explicitly stated that the front-end data center business will not "flatten out," the enterprise business is still expanding, and routing—covering edge routing, data center interconnect (DCI), and scale-across scenarios—is opening up new market opportunities. This suggests a model where all product lines can contribute to growth, rather than a single-point catalyst.
AI Networking Split into Three Fronts, Arista Bets on Scale-Out and Scale-Across
During the call, management broke down the AI fabric into three clear categories: scale-up, scale-out, and scale-across.
Scale-up occurs primarily within a rack or in close-proximity interconnects. In Nvidia's complete vertical solution, NVLink dominates this space, leaving limited room for Arista. The company is best positioned to play in the other two categories: scale-out, which connects large AI clusters, and scale-across, which connects AI infrastructure across different campuses, data centers, and regions.
Management noted that the strategic importance of scale-across is rising because clients face constraints on computing power, physical space, and electricity, making it difficult to concentrate all computing resources in a single location permanently. Arista's 7800 platform is positioned as the core spine for distributed scale-across applications, emphasizing traffic isolation, context-based routing, and security capabilities.
The call also provided a market size estimate: the total addressable market (TAM) for scale-across switching and routing is approximately $15 billion to $20 billion by 2030, with a TAM of about $3 billion to $4 billion in 2026. Arista expects to reach about $1.2 billion this year. Management indicated this business is still in its early stages but has a steep growth trajectory.
Technology Differentiation Focuses on XPU Utilization, Not Just Bandwidth
On the conference call, management detailed three key technology capabilities to address external skepticism about the competitiveness of the Ethernet solution.
SSU (Smart System Upgrade) aims to minimize business disruption during switch software upgrades. In AI training and inference clusters, XPU resources are extremely costly, and downtime from network device restarts directly impacts compute utilization. SSU's core value is resolving the conflict between security upgrades and continuous compute operation.
MRC (Multipath Reliable Connection) solves the problem in traditional AI networks where a single flow can only travel on one path. It allows a single flow to be dispersed across multiple paths within the fabric and reassembled at the receiving end, aiming to reduce performance loss from path contention.
SRv6 handles path control by tagging data packets with a segment ID to specify the transmission path and integrating real-time congestion signals to move traffic away from hot links. Management emphasized that Arista's advantage is not in a single switch but in the consistency provided by the EOS unified operating system, extending from scale-out fabric to long-distance scale-across routing.
On the hardware front, the 7060XE7 was introduced as a next-generation AI network product, featuring 100Tb capacity and 1.6Tb throughput, with liquid cooling options. Products related to 1.6T will enter the trial phase in the second half of the year, with a small number of large customers expected, and mass production is still planned for 2027.
Supply Chain is the Core Hard Constraint, Tightness May Last Until 2028
On the call, COO Todd Nightingale clearly identified the supply chain as the primary bottleneck to current growth, stating that the company has been investing consistently over the past six months to improve delivery certainty.
Multi-year purchase commitments increased from $3.6 billion a year ago to approximately $9.7 billion at the end of the second quarter, up from $8.9 billion at the end of the first quarter, mainly for chip procurement related to new products and AI deployments. For memory, supply for 2026 is already locked in, with visibility for DDR4, DDR5, and NAND extending into 2027. Manufacturing capacity for PCBs and optical modules can be built within a 12-month window. The liquid cooling supply chain has been established, covering suppliers of cold plates, quick connectors, and tubing.
Manufacturing and distribution capabilities are expanding in parallel. The company now works with three contract manufacturers and has three distribution facilities located in the U.S., Asia, and Mexico. Nightingale stated, "We have ramped up our capacity in manufacturing and distribution and have negotiated better terms for component delivery."
Management also acknowledged that industry-wide component shortages and cost increases are ongoing, extending their estimate of this cycle's duration to 2028. Inventory at the end of the second quarter was $2.5 billion, up from $2.4 billion in the first quarter, with inventory turnover remaining at 1.7. Deferred revenue stood at approximately $6.9 billion, higher than the $6.2 billion in the prior quarter, with product deferred revenue increasing by about $600 million sequentially, reflecting the time lag from new product ramps and customer acceptance cycles.
Cost Pressures Persist, but Gross Margin Bottom May Be Past
The sequential improvement in gross margin was partly attributed to tariff refunds, which management said contributed about 20 to 30 basis points, with the remainder coming from an improved customer mix. The impact of price increases has not yet been significantly reflected in current results and is expected to appear in financial figures by the end of the year or in 2026.
The full-year non-GAAP gross margin guidance remains in the 62% to 64% range, already incorporating factors such as customer mix changes, rising memory and silicon costs. The operating margin target was raised from the previous range to 48% to 49%, indicating that while the company is increasing spending on AI hardware, liquid cooling, and high-radix switch R&D, its expense control remains resilient.
The second-quarter operating expense structure shows R&D expenses of $278.1 million, or 9.2% of revenue; sales and marketing expenses of $109.8 million, or 3.6% of revenue; and G&A expenses of $23.1 million, or 0.8% of revenue. The overall expense ratio remains low, providing a stable profit base for Arista's business model.
Customer Base Anchored by Microsoft and Meta, Room for New Large Customers
On the call, management confirmed that Microsoft and Meta remain the core large customers, with no changes to commitments to these two clients. At the same time, management left room for new customers contributing more than 10% of revenue, stating that in 2026, there could be one or two such customers.
Management pointed out that customer demand itself is not in question; the shipment mix depends on supply chain, product priorities, and specific project timelines. Among AI back-end, front-end data centers, campus networks, and routing, whichever segment receives more supply will contribute more revenue.
For Arista, the core challenge for 2026 has shifted from "proving there is demand for AI networking" to "stably delivering complex products during a supply-constrained industry cycle." The $3 billion quarterly revenue already demonstrates demand strength, and the full-year guidance of $12.6 billion now shifts the pressure to execution.
Comments