The prominent Wall Street investment firm Evercore has maintained Arista Networks (ANET.US) on its "Top Picks" list, assigning it an "Outperform" rating—its most optimistic bullish stance—and reaffirming a $200 price target. The core rationale centers on the unprecedented surge in demand for both internal interconnectivity within AI data centers and high-speed connections between data centers, driven by the ongoing construction frenzy. As of last Friday's market close, Arista Networks' stock has risen nearly 20% year-to-date, trading around $154, implying a potential 30% upside over the next 12 months according to Evercore's base target. Furthermore, Evercore has also set a bullish scenario price target of $300 for Arista, suggesting the possibility of the stock doubling in the not-too-distant future.

Regardless of whether Nvidia GPUs, AMD GPUs, Google TPUs, or custom AI ASIC/XPU solutions from major cloud providers dominate the landscape, the construction of AI data centers is fundamentally inseparable from the high-performance Ethernet infrastructure and software platforms championed by Arista Networks. The underlying technical logic is straightforward: bottlenecks in AI training and massive-scale AI inference workloads lie not only in the raw compute power of AI chips, HBM, or DRAM but also in factors like ASIC/XPU utilization, data movement, cluster synchronization, storage access, and cross-rack communication. A technical statement from an Arista blog notes that 30% to 50% of processing time in AI compute infrastructure systems can be spent on network data exchange. Network bottlenecks can lead to XPU idling, wasting expensive compute capital expenditures and power/cooling costs.

Therefore, Arista provides one of the "network infrastructure chassis" for AI compute super-factories. This means that irrespective of which AI chip technology ultimately prevails, as long as AI compute infrastructure clusters continue to scale, Arista's focus on open Ethernet networking, network automation/operations, and data center interconnect (DCI) product lines will become an indispensable value layer. From an investment narrative perspective, Arista represents a prime example of the AI super-cycle broadening from the "GPU/HBM dominance of AI compute infrastructure" to the "AI data center interconnect layer."

Evercore's thesis—highlighting a total addressable market (TAM) for network infrastructure exceeding $150 billion, an AI backend network TAM over $120 billion, and an expanding AI infrastructure customer base including Meta, Microsoft, Oracle OCI, Anthropic, and Google Cloud—essentially underscores that massive AI capital expenditures (Capex) require not just purchasing AI chips but also procuring substantial network fabric to prevent those chips from idling.

Risks to Arista's growth include supply chain constraints, switching ASIC availability, competition from Nvidia's Spectrum-X ecosystem, customer concentration, and valuation concerns. However, from a medium- to long-term view, as AI compute clusters scale from tens of thousands to hundreds of thousands of accelerators and expand from single to multiple data centers, companies like Arista, specializing in AI Ethernet network infrastructure and DCI, will become critical, nearly unavoidable components of AI data center infrastructure.

What is Arista Networks? At its core, Arista Networks is a high-performance networking hardware and software company focused on the cloud/AI data center sector. Its core products are not GPUs, TPUs, or AI servers, but rather the high-speed Ethernet switches, routing systems, EOS network operating system, CloudVision automation/observability platform, and Etherlink network infrastructure architecture for AI clusters that efficiently connect these compute nodes. The company officially positions itself as a "client-to-cloud networking" provider for large-scale data center/AI, campus, and routing environments. Its AI Networking solutions explicitly emphasize providing IP/Ethernet networks for AI/ML workloads, supporting various AI chips and storage systems.

Arista Networks' core bet is on data center interconnectivity, particularly the scale-out backend and front-end networks within AI clusters, and future scale-across connectivity linking multiple AI data centers; this also implies strong DCI capabilities. In simple terms, the ultra-short-distance, scale-up GPU-to-GPU connectivity within an AI server rack is typically handled by technologies like NVLink/NVSwitch, UALink, or proprietary/semi-proprietary interconnects. The large-scale, scale-out Ethernet fabric between server racks, within cloud pods, and between AI clusters is precisely Arista's area of expertise.

Arista states its AI fabric encompasses both scale-out and scale-up network architectures, with scale-out further divided into front-end and back-end. Its 7700R4 Distributed Etherlink Switch, for instance, can support single-hop distributed AI backend networks for over 30,000 400GbE accelerators. Regarding DCI—data center interconnect—Arista is not a traditional long-haul optical transport company like Ciena. However, it participates in the DCI space through its high-end data center switching/routing platforms, EVPN/VXLAN, MPLS, Segment Routing, long-reach optical modules, and internet-scale routing capabilities. Official documentation for the Arista 7800R/7800R3/R4 series explicitly mentions support for Data Center Interconnect (DCI) and long-haul optics, targeting large-scale L2/L3, EVPN cloud data centers, service provider edge, and internet-scale routing scenarios.

Evercore Backs Arista! Unified Ethernet Architecture Targets $150B Network TAM The Evercore analyst team, led by Amit Daryanani, stated: "We view Arista as a core AI compute holding for investors, given its unique leadership position in solving AI infrastructure bottlenecks and improving XPU utilization." The team highlighted five key points and also presented a bullish scenario with a price target as high as $300.

First, the total addressable market (TAM) for high-performance data center networking exceeds $150 billion, with a compound annual growth rate (CAGR) over 25%. Analyst projections also indicate the backend network TAM could surpass $120 billion, sustaining a CAGR above 30%, while the front-end network TAM is approximately $30 billion, expected to maintain a high-single-digit CAGR. Daryanani's team noted: "Given Arista's ability to deliver a scalable, unified Ethernet architecture across Scale-Up, Scale-Out, and Scale-Across frameworks in a vendor-agnostic manner, it holds a unique winning position in high-performance AI data center networking. Arista's EOS software stack provides critical differentiation, enabling operational consistency, automation, and resilient growth for AI workloads in complex, multi-vendor AI compute infrastructure clusters."

Second, sales growth is anticipated to exceed 30%. Analysts pointed out that Arista recently raised its growth outlook through 2028 to over 20% (a significant increase from the mid-teens percentage discussed at its Analyst Day). They added that, driven by expanding AI compute infrastructure demand, new customer wins, and enterprise business growth tailwinds, the company should sustain growth above 30% across different end markets, fueled by market share gains and new clients.

Third, customer diversification through Anthropic, Oracle OCI (ORCL.US), and Google (GOOGL.US). Analysts stated: "We expect that by CY26, Arista will have OCI and Anthropic as customers each accounting for over 10% of total revenue, alongside Meta and Microsoft. Furthermore, Arista has the potential to scale with Google Cloud in CY27 and beyond. As AI clusters expand aggressively, we anticipate these two customers will contribute significant incremental revenue by CY28."

Fourth, analysts generally view the CY26 revenue guidance of $11.5 billion (representing 27.7% growth) as conservative and see a path to growth of at least 30%. Daryanani's team commented: "While supply constraints impacted the CY26 guidance, we emphasize that Arista's demand is deferred, not canceled, which should lead to higher and more sustained long-term growth."

Finally, analysts noted that the Campus business is expected to expand to $800 million in 2025, with CY26 guidance at $1.25 billion (approximately 55% growth). They believe this business could exceed $2 billion by 2028. The analysts stated: "Investors with a long-term holding horizon should consider owning Arista, as temporary supply issues are expected to ease in the second half. Arista is positioned for multi-year sales and EPS growth in the 30%+ range."

Evercore's $300 Bull Case Target Price Additionally, Daryanani and his team at Evercore presented a bullish scenario assumption with a target price of $300. The analysts believe Arista can sustain revenue and profit growth exceeding 30% through 2030. The primary growth drivers, according to Daryanani's team, will be the frenzy in scale-out/scale-across AI compute infrastructure build-out, new customer wins, front-end network recovery, and campus networking.

Daryanani's team elaborated: "We assume margins normalize to the low-to-mid 40% range over time, potentially driving EPS above $10 by CY30. This is driven by the foundational cloud business plus AI scale-up, scale-out, and scale-across opportunities (capturing roughly 20% share of AI infrastructure-related networking), compounded Campus-level networking revenue growth (as Arista's share increases from 2% to 8%), and ongoing share repurchases. Key catalysts for multiple expansion include: 1) Arista raising its CY26 AI-related revenue target above $3.5 billion as deferred revenue converts; 2) Securing silicon supply to meet excess market demand; and 3) Further disclosure of substantial new orders from Anthropic and Google."

As AI agents gain global traction, the AI compute investment theme is shifting from a "single-point compute race centered on AI GPUs" to "full-stack compute systems driven by AI agents." The next wave of excess alpha returns will no longer belong solely to the top players in AI GPU/AI ASIC domains but will systematically diffuse across the full-stack AI compute infrastructure layer, encompassing data center CPUs, high-performance Ethernet infrastructure, DRAM/NAND/HBM memory, AI PCBs, liquid cooling systems, data center optical interconnects, ABF substrates/glass substrates, and broad-based wafer foundries.

On April 30th, the three cloud hyperscalers—Microsoft, Google, and Amazon—delivered strong results on the same night, highlighting the unexpectedly rapid growth of their AI-boosted cloud businesses, prompting Wall Street to reprice the commercial returns of AI. A recent Morgan Stanley analyst report estimates that the combined capital expenditures of the five major hyperscalers (Amazon, Google, Meta, Microsoft, Oracle) will reach approximately $800 billion in 2026 and potentially surpass $1.1 trillion in 2027, an upward revision from a prior forecast of $950 billion.

As the AI compute infrastructure construction wave led by these tech giants forms a path reminiscent of the early days of railroads, power grids, broadband, and cloud computing—characterized by "heavy Capex first, followed by application explosion"—the upward trajectory for the AI compute supply chain and the global equity bull run driven by the AI compute narrative may be far from over. Morgan Stanley analysts emphasize the core logic behind these massive investments: heavy upfront investment to build capacity, followed by recouping costs through scaled commercial revenue and ROIC based on the AI compute resources. The surge in cloud backlogs and AI application tokens serves as the most direct evidence that this logic is working, with the hyperscalers' cloud business growth exceeding expectations and leading Wall Street to reassess the commercial payoff of AI.