Data centers are starting a "high-speed connection revolution"! Arista (ANET.US) rides the AI cluster trend, and Evercore sets a bullish target price of $300.

Wall Street's well-known investment firm Evercore has maintained Arista Networks (ANET.US) on its "preferred stock" list, giving it a "outperforming the market" bullish rating, reiterating a target price of $200. The core logic is the unprecedented surge in internal interconnection and high-speed interconnection demand between data centers brought about by the AI data center construction boom. As of last Friday's US stock market close, Arista Networks has risen nearly 20% this year, with a stock price around $154, implying a potential increase of up to 30% in the next 12 months, according to Evercore. In addition, Evercore has also set a $300 bullish target price for Arista, indicating a potential doubling of the stock price in the near future. Whether it is NVIDIA Corporation's GPU, AMD GPU, Alphabet Inc. Class C (Google) TPU, or the self-developed AI ASIC/XPU routes of cloud computing giants, AI data center construction cannot avoid the high-performance Ethernet infrastructure and software platform led by Arista Networks. The technology logic behind this is simple: the bottleneck in AI training and massive AI inference workloads is not only in AI chips/HBM/DRAM computing power itself, but also in AISC/XPU utilization, data transfer, cluster synchronization, storage access, and inter-rack communication. A technical declaration from Arista's blog points out that 30%-50% of the processing time in an AI computing infrastructure system may be spent on network data exchange, and network bottlenecks can cause XPU idle time, wasting expensive computing power capital expenditure and power/cooling costs. Therefore, what Arista provides is one of the "network infrastructure chassis" for an AI computing super factory: meaning that no matter which AI chip technology route wins, as long as the cluster size of AI computing infrastructure continues to expand, Arista's focus on open Ethernet networks, network automation, maintenance, and cross-data center interconnect product lines will become an indispensable value link. From an investment narrative perspective, Arista can be said to be a typical representative target of the AI super bull market, expanding from the "GPU/HBM dominating AI computing infrastructure" to the "AI data center interconnect layer." The logic given by Evercore - a network infrastructure TAM of over $150 billion, backend network TAM exceeding $120 billion, and AI infrastructure customers expanding to Meta, Microsoft Corporation, Oracle Corporation OCI, Anthropic, and Alphabet Inc. Class C cloud computing centers - all essentially highlight the fact that the massive AI Capex (AI computing infrastructure) not only needs to buy AI chips but also requires a large purchase of network fabrics that can prevent chips from idling. Arista's growth risks lie in supply chain constraints, exchange ASIC supply, competition from NVIDIA Corporation-dominated Spectrum-X system, customer concentration, and valuation overcrowding; but in the medium to long term, as AI computing clusters continue to scale from thousands of cards to tens of thousands of cards, companies like Arista, specializing in AI Ethernet network infrastructure and data center interconnect, will become key links that are difficult to bypass in AI data center infrastructure. Who is Arista Networks? Arista Networks is essentially a high-performance networking equipment and software company focusing on cloud computing/AI data center fields. Its core products are not GPUs, TPUs, or AI servers, but rather high-speed Ethernet switches, router systems, network operating system EOS, CloudVision automation/observability platform, and Etherlink network infrastructure architecture for AI clusters. The company's official positioning is as a "client-to-cloud networking" provider serving large data centers/AI, campus, and routing environments; its AI Networking solution specifically emphasizes providing IP/Ethernet networks for AI/ML workloads, covering various AI chips and storage systems. Arista Networks' core focus is on data center internal interconnect, especially the scale-out back-end networks within AI clusters, front-end networks, and future scale-across inter-data center linkage; this also means that the company has strong DCI capabilities. In simple terms, the extremely short-range scale-up communication within AI server cabinets is usually supported by NVLink/NVSwitch, UALink, or proprietary/semi-proprietary interconnects, while the large-scale scale-out Ethernet fabrics between server cabinets, within cloud computing pods, and among AI clusters are where Arista excels in niche markets. Arista officially states that its AI fabric includes scale-out and scale-up network architectures, with scale-out further divided into front-end and back-end; its 7700R4 Distributed Etherlink Switch can support a distributed AI back-end network with over 30,000 400GbE accelerators in a single hop. As for DCI - interconnection between data centers, Arista is not a traditional long-haul optical transport company like Ciena but participates in data center interconnect through high-end data center exchange routing platforms, EVPN/VXLAN, MPLS, Segment Routing, long-distance optical modules, and inter-data center routing capabilities. Arista's 7800R/7800R3/R4 series official documentation clearly mentions support for Data Center Interconnect (DCI) and long-haul optics, catering to large L2/L3, EVPN cloud data centers, service provider edges, and internet routing scenarios. Evercore endorses Arista! Unified Ethernet architecture targeting a $150 billion network TAM Amit Daryanani's team of analysts at Evercore stated, "We believe Arista is a core target for investors in AI computing power, as the company holds a unique leadership position in solving AI infrastructure bottlenecks and improving XPU utilization." The analyst team emphasized five key points and proposed a scenario assumption under the bull market, setting a target price as high as $300. Firstly, the total potential market (TAM) for high-performance data center networks exceeds $150 billion, with a compound annual growth rate (CAGR) of over 25%. The analysts predict that the backend network TAM could exceed $120 billion and maintain a CAGR of over 30%, while the frontend network TAM is approximately $300 billion and is expected to maintain a high single-digit CAGR. Daryanani and his team stated, "Given Arista's ability to provide scalable, unified Ethernet architecture in a vendor-agnostic manner in Scale-Up, Scale-Out, and Scale-Across architectures, Arista holds a unique winning position in the high-performance network domain of AI data centers. The EOS software stack of Arista provides critical differentiation, enabling efficient operation, automation, and resilience of AI workloads in complex AI computing infrastructure clusters and multi-vendor environments." Secondly, revenue growth expectations are expected to exceed 30%. The analysts pointed out that Arista recently raised its performance growth expectations for 2028 to over 20% (a significant increase compared to the mid-teens percent range at the time of the analyst's daily). The analysts added that given the expansion of AI infrastructure demand, new customer wins, and corporate growth factors, the company should maintain over 30% growth driven by different end markets, share improvements, and new customer drivers. Thirdly, diversification of major customer sources through Anthropic, Oracle Corporation OCI (ORCL.US), and Alphabet Inc. Class C (GOOGL.US). The analysts stated, "We expect that by CY26 (2026 calendar year), Arista will have OCI and Anthropic as customers accounting for more than 10% of total revenue, and also have potential with Meta and Microsoft Corporation. In addition, Arista has the potential to expand its scale with Alphabet Inc. Class C cloud computing centers by CY27 and beyond. With the rapid expansion of AI clusters, we expect these two customers to bring significant incremental revenue data by CY28." Fourthly, the analysts unanimously believe that revenue guidance of $11.5 billion in 2026 (a 27.7% increase) is conservative and see a path to growth of at least over 30%. Daryanani and his team stated, "Although supply constraints hinder the CY26 guidance, it is important to emphasize that Arista's demand is deferred rather than canceled, which will bring higher and more sustained long-term growth." Lastly, the analysts stated that the Campus business is expected to expand to $800 million in 2025, with guidance for 2026 set at $1.25 billion (roughly a 55% increase). The analysts believe that this business is expected to exceed $2 billion by 2028. The analysts said, "Investors with a long-term holding period should consider allocating Arista, as temporary supply issues are expected to ease in the second half of the year. Arista is expected to achieve sales and earnings growth of over 30% for multiple years." Evercore sets a $300 bullish target price Furthermore, Amit Daryanani and his leadership team at Evercore proposed a scenario assumption under a bullish market, setting a target price of $300. The analysts stated that Arista is expected to maintain growth rates of over 30% in revenue and profits by 2030. Daryanani-led analysts pointed out that the main drivers of growth will be the surge in AI computing infrastructure construction, new customer wins, frontend network recovery, and campus networks. Daryanani and his team stated, "We assume that over time, profit margins will normalize to around 40% to CY30, aiming to achieve over $10 in earnings per share, driven by core cloud computing business, AI scale-up, scale-out, and scale-across opportunities (capturing about 20% of AI infrastructure-related network share), compound growth in campus-level network business (with Arista's share rising from 2% to 8%), and ongoing buybacks. Key catalysts driving valuation multiple expansion include: 1) as deferred revenue shifts, Arista will raise its CY26 AI-related revenue target to over $3.5 billion; 2) securing silicon supply to meet excess demand in the market; 3) Disclosure of a large number of new orders from Anthropic and Alphabet Inc. Class C." As AI intelligent agents are sweeping the globe, the investment theme of AI computing power is shifting from the "competition based on single-point computing power around AI GPU" to the "full-stack computing power system driven by AI intelligent agents." The next round of excess alpha returns will no longer be limited to the strongest leaders in the AI GPU/AI ASIC field but will systematically spread across data center CPUs, high-performance Ethernet infrastructure, DRAM/NAND/HBM storage, AI PCBs, liquid cooling systems, data center optical interconnect systems, ABF substrates/glass substrates, and extensive silicon foundries in the full-stack AI computing power infrastructure layer. On April 30th, Microsoft Corporation, Alphabet Inc. Class C, and Amazon.com, Inc., the three cloud computing super giants, all delivered outstanding results on the same night, highlighting the unexpectedly rapid growth of their cloud computing business benefiting from the AI tidal wave and prompting Wall Street to reevaluate the commercial returns of AI. The latest research report from Morgan Stanley's analyst team indicates that the five gigantic tech giants (Amazon.com, Inc., Alphabet Inc. Class C, Meta, Microsoft Corporation, Oracle Corporation) are expected to have a combined capital expenditure of around $800 billion in 2026, with a potential to exceed $1.1 trillion in 2027, a slight increase from the previous estimate of $950 billion. As the wave of AI computing infrastructure construction led by these tech giants forms a path similar to the early "capital expenditure followed by application explosion" of railways, power grids, broadband, and cloud computing, the soaring trajectory of the global stock market driven by the AI computing industry chain and AI computing bull market narrative may be far from over. Morgan Stanley's analysts emphasize that the core logic behind this massive capital investment lies in: first, reinvesting and building capacity, and then relying on the scalable commercial revenue and ROIC recovery based on AI computing resources; the surge in backlogged orders from cloud computing and the explosion of AI application tokens are direct evidence that this logic can be realized. These giants' cloud computing businesses' unexpectedly rapid growth rates are prompting Wall Street to reassess the commercial returns of AI.