The data center is initiating a "high-speed connection revolution"! Arista (ANET.US) saw a 38% surge in revenue, and the strong performance outlook confirms the AI Ethernet supercycle.

Arista Networks (ANET.US) published its latest quarterly results and future outlook after the U.S. stock market closed on Tuesday (Wednesday morning Beijing time). The company provided third-quarter guidance that exceeded Wall Street analysts' consensus expectations, highlighting that the demand for its high-performance networking equipment will continue to expand strongly amid the unprecedented global wave of enterprises building artificial intelligence computing infrastructure. Whether it is NVIDIA Corporation GPUs, AMD GPUs, Alphabet Inc. Class C (Google) TPUs, or the cloud computing giants' self-developed AI ASIC/XPU solutions, the construction of AI data centers cannot do without the high-performance Ethernet infrastructure and software platforms led by Arista Networks. Thanks to its strong performance and outlook, Arista's stock price surged by as much as 16% in after-hours trading. Arista Networks focuses on producing high-speed computer networking devices that support large AI data centers, with core customers including Microsoft Corporation and Amazon.com, Inc. According to data compiled by LSEG, the company expects third-quarter revenue to be around $3.3 billion, exceeding the analysts' average expectation of the recently raised $2.94 billion. In order to diversify its revenue sources, the company has been expanding its enterprise campus network and branch network infrastructure business outside its core cloud computing customer base. This latest earnings report, along with the management's updated future outlook, is a significant positive for Arista Networks' fundamentals and short-term stock price: it not only substantially surpassed the original revenue and profit market consensus, but more importantly, it demonstrated that the previously constrained supply of chips, memory, PCBs, and optical devices is easing, and strong orders are beginning to convert from deferred revenue to confirmed revenue. A technical statement from Arista's blog pointed out that 30% to 50% of processing time in AI computing infrastructure systems may be spent on network data exchange. Network bottlenecks can lead to XPU idling, wasting expensive computing capital expenditures, as well as power and cooling costs. Thus, Arista provides one of the network infrastructure chassis for AI computing super factories: indicating that regardless of which AI chip technology route prevails, as long as the scale of AI computing infrastructure clusters continues to expand, the open Ethernet networks, automated network operations and inter-data center interconnection products that Arista focuses on will become indispensable value components. Although Arista Networks' stock price has retreated since July, following the global AI computing theme's deleveraging and the clearing of extremely crowded positions, the increase since 2026 is still a remarkable 50%. In comparison, the benchmark U.S. stock indices, specifically the S&P 500, has risen less than 15% this year. From computing chips to high-speed interconnects in data centers, Arista Networks is reaping the benefits of AI infrastructure. The core logic behind the strong expansion of Arista Networks' stock price and fundamentals in recent years lies in the unparalleled increase in internal interconnection and high-speed interconnect needs between data centers driven by the AI data center construction frenzy. The bottlenecks in AI training and the immensely massive AI inference workloads are not only within the AI chips/HBM/DRAM computing itself, but also involve AISC/XPU utilization, data transport, cluster synchronization, storage access, and inter-rack communication. From an investment narrative perspective, Arista can be seen as the most typical representative of the AI super bull market, where the focus has shifted from GPU/HBM dominating AI computing infrastructure to AI data center interconnection layer. Arista Networks' large AI infrastructure clients now include Meta, Microsoft Corporation, Oracle Corporation OCI, Anthropic, and Alphabet Inc. Class C cloud computing centersessentially highlighting that massive AI Capex (AI computing infrastructure) requires not only the purchase of AI chips but also significant procurement of networking fabric that prevents the chips from idling. Currently, Arista's business focus remains on internal interconnection within data centers, with its core advantages centered on the Scale-up and Scale-out networks within AI clusters: utilizing Etherlink, the 7700/7800 series switches, and EOS, CloudVision software to cover rack-level accelerator interconnections, GPU backend training networks, frontend storage, and business access networks, solving high bandwidth, low latency, lossless transmission, and congestion control issues between thousands to hundreds of thousands of XPUs. At the same time, Arista has extended its capabilities to DCI data center interconnection, known as Scale-across: when a single campus faces limitations in power, land, and cooling conditions, massive XPU clusters must be distributed across different buildings, cities, or even regions, necessitating network connections through 800G links, deep buffering, intelligent traffic engineering, MACsec/IPsec encryption, and 400ZR/coherent optical technology connecting multiple AI clusters. Arista Networks' core performance in the second quarter also outperformed Wall Street analysts' consensus expectations, marking its first revenue exceeding $3 billion, achieving $3.04 billion, which indicates a significant year-over-year growth of 38%. The adjusted earnings per share were $1.02, higher than analysts' expectations of $2.82 billion and earnings per share of 88 cents. The company expects the adjusted earnings per share for the third quarter to be in the range of $1.06 to $1.08, also higher than the analysts' average expectation of 91 centsthis earnings per share expectation has been continuously revised upward by analysts since June. Other second-quarter performance and operational data show that Arista's product revenue increased from $1.877 billion to $2.605 billion, representing a year-over-year growth of 38.8%. Service revenue rose from $327.8 million to $430.5 million, a year-over-year increase of 31.3%. Under GAAP standards, gross profit grew by 32.8% to $1.9103 billion, while the GAAP gross margin fell from 65.2% to 62.9%. Non-GAAP gross margin also declined from 65.6% to 63.4%, reflecting the cost pressure from AI Ethernet high-performance switches, high-speed optical modules, memory, and silicon chip costs and the customer mix. Meanwhile, Arista's GAAP operating profit grew by 39.7% to $1.378 billion, and non-GAAP operating profit increased by 40.6% to $1.5137 billion, with the non-GAAP operating profit margin surprisingly rising from 48.8% to 49.9%. GAAP net profit increased by 36.5% to $1.2129 billion, while non-GAAP net profit grew by 39.3% to $1.3017 billion. The adjusted EPS rose from 73 cents to $1.02, a year-over-year increase of 39.7%, indicating that revenue expansion is translating into significant operational leverage. Arista's operational quality in the second quarter is also strong, with operating cash flow for the first half of the year rising from $1.8418 billion to $2.7765 billion, a year-over-year growth of 50.7%. R&D expenditure for the quarter increased by 17.4% to $348.2 million, with a focus on liquid cooling, high-port-density switches, and AI network optimization software. The second quarter's operating cash flow was about $1.1 billion, and end-of-period cash and cash equivalents stood at approximately $13.3 billion. Deferred revenue grew from $6.2 billion at the end of the first quarter to $6.9 billion, while procurement commitments rose from $8.9 billion to $9.7 billion, mainly for new AI high-performance networking infrastructure and the chips needed for deployment. It is reported that the company's management has raised its 2026 outlook for the third time: full-year revenue is expected to be around $12.6 billion, reflecting a year-over-year growth of 40%, with an increase of $1.1 billion compared to the May forecast of $11.5 billion. The AI Fabric revenue target is set at least $3.5 billion, and the campus network revenue target is set at least $1.25 billion, and the full-year non-GAAP operating profit margin target has been raised to 48%-49%. The third quarter is expected to generate $3.3 billion in revenue, with adjusted EPS of $1.06 to $1.08, both significantly above market consensus expectations prior to the earnings report. Arista is positioned at the central nervous system of large global AI data center projects. Essentially, Arista Networks is a high-performance networking equipment and software company focused on the field of cloud computing/AI data centers. Its core products are not GPUs, TPUs, or AI servers, but rather high-speed Ethernet switches, routing systems, network operating systems EOS, CloudVision automation/observability platforms, and Etherlink network infrastructure architectures that efficiently connect these computing nodes. The company's official position is as a "client-to-cloud networking" provider serving large data centers/AI, campus, and routing environments; its AI Networking solutions explicitly emphasize providing IP/Ethernet networks for AI/ML workloads, covering various AI chips and storage systems. The underlying logic for Arista benefiting from the AI boom is that large model training and inference do not operate independently on GPUs, but rather involve thousands to hundreds of thousands of GPUs, TPUs, or other XPUs continuously exchanging parameters, gradients, and intermediate results. As long as network congestion, packet loss, or tail latency occurs, the expensive accelerators will wait for data, and GPU utilization and task completion speed will decrease accordingly. Therefore, AI computing server clusters require high bandwidth, low latency, nearly lossless east-west networks, along with deep buffering, congestion control, and load balancing to handle synchronization burst traffic. Arista upgrades ordinary switching devices into the "central nervous system" coordinating the entire AI computing server cluster system through Etherlink switches, EOS unified operating system, RoCE Ethernet, VOQ, MRC multipath transport, and SRv6 traffic engineering; its AI Fabric clients have expanded from four to five in 2024 to over 100 in total. The electricity construction of AI data centers and Arista's demand is essentially a direct transmission chain: the addition of megawatts and gigawatts of capacity to the power grid is necessary to deploy more high-powered GPU racks; the more GPUs, the greater the required switch ports, optical interconnects, and backbone bandwidth. When a single region lacks enough power, cloud vendors must also distribute AI clusters across multiple campuses and even different areas, further driving the need for "scale-across" inter-data center networks. Aristas 7060XE7 can provide 1.6 Tbps per port and approximately 100 Tbps system bandwidth; LPO technology can reduce interconnect power consumption by about 60%; the liquid cooling XPO solution can save up to 75% of network racks and 44% of footprint space, while the high-port-density 7800 AI Spine reduces power consumption and latency by lowering network hierarchy. Arista does not produce electricity but determines how much effective communication bandwidth each megawatt of electricity can support, how many GPUs can remain fully loaded, and ultimately how many tokens are producedthis is precisely the core reason why the value of its networking equipment rises in line with AI computing density during an electricity-scarce era. The global AI computing infrastructure boom cannot bypass high-performance Ethernet, and Arista is one of the most core and profitable suppliers in this market. Recently, Wall Street financial giant JPMorgan raised its price target for Arista Networks from $200 to $220 while maintaining an "overweight" rating. Based on the regular trading closing price of $190.51 on August 4 and an approximate market cap of $242.7 billion, the $220 target suggests about a 15.5% potential upside, equating to a market value of about $280.2 billion. JPMorgan's core bullish logic can be summarized as follows: Arista is upgrading from a traditional cloud switch leader to a network infrastructure platform for AI computing clusters; cloud giants, Neocloud, and large enterprises are simultaneously increasing AI capital expenditures, and the high-frequency parameter exchanges between tens of thousands to hundreds of thousands of GPUs make Ethernet's backend network, 800G/1.6T switching, optical interconnects, and congestion control critical bottlenecks that determine GPU utilization. With Etherlink, EOS, and the full-stack architectures of scale-up, scale-out, and scale-across, Arista is expanding its replacement space for InfiniBand and traditional switch systems; additionally, its latest 7060XE7 platform offers up to 100 Tbps of system bandwidth and reduces interconnect power consumption by about 60% compared to traditional pluggable optical modules, directly aligning with the construction goals of AI data centers for "higher bandwidth, lower latency, and lower power consumption." If Wall Street subsequently makes significant upward revisions to the 2026-2027 revenue guidance, EPS, and AI Ethernet network product revenue forecasts, the price target may be further elevated. In the next bull market phase, whether the company's stock can continue to break through will hinge not just on proving "AI high-performance network demand," but also on showing that the 40% growth, nearly 50% operating profit margin, and the increase in Ethernet's share in AI backend networks can be sustained beyond 2027.