Arista Networks: Connecting AI Infrastructure

Why the world's largest AI and cloud companies rely on Arista to keep their networks running.

Artificial intelligence has put the spotlight on chips, but chips alone do not build AI infrastructure. They rely on increasingly sophisticated networks to move vast amounts of data quickly and reliably. That is where Arista Networks comes in. The $220 billion company supplies the high-speed switching hardware and software that enable the world's largest cloud and AI data centres to operate as a single, integrated system, and expand to the scale needed for frontier AI.

Founded in the aftermath of the dotcom bubble, Arista recognised cloud computing would require a fundamentally different networking architecture. At the time, existing solutions were fragmented, siloed and difficult to manage. The box-by-box network management that was commonplace was too slow, too fragile and too complex for the scale that cloud demanded.

Cloud computing changed the nature of the network. As data centres grew, more traffic moved within them, with servers communicating rising in intensity and complexity. A modern cloud data centre is a living grid of compute and storage, with workloads appearing, moving and disappearing constantly. The network to orchestrate this turmoil had to become flatter, faster, programmable and far easier to operate at scale. This is Arista’s market.

Arista builds exceptionally high-performance hardware, but its durable competitive advantage lies the software it builds to coordinate it. Its Extensible Operating System (EOS) provides a single operating system, management layer and automation platform across its entire product portfolio, from cloud data centres to enterprise networks and campuses.

The commercial value is straightforward. Customers can automate, monitor, and manage their networks through a single platform rather than multiple disconnected systems. That saves thousands of engineer hours through simpler deployments, easier upgrades and better diagnostics. It also improves operational reliability, helping customers avoid outages that can cost far more than the network itself, justifying its premium pricing

Arista’s primary customers are the world’s leading cloud companies, which account for around half of group revenue. It also serves a growing base of neoclouds, AI and telecom providers, alongside a broad enterprise customer base. Microsoft and Meta have been particularly important customers.

This customer concentration is a risk, but it also reflects the value Arista provides. The world’s largest and most technically demanding network operators choose Arista Networks because they make incredible network complexity (and its associated cost) manageable.

Arista, whose products  were built for the cloud, has steadily gained share in the most demanding parts of the market. From virtually no share in high-speed data centre switching in 2012, it now holds around 35% of the $25 billion market.

AI Drives Networking Growth

For the last few years, investors have focused overwhelmingly on GPUs. They are scarce, expensive and essential to the first phase of the AI build-out. But chips do not operate in isolation. A large AI system only works if thousands of chips can synchronise seamlessly. The more chips you add together, the more critical the network becomes. If the network slows, expensive chips sit idle, wasting precious capital. In an AI data centre, the network increasingly determines how much useful compute a customer gets from the billions being spent on GPUs.

Arista’s integrated hardware and software are well suited to the demands of AI networking. As AI adoption accelerates, it also significantly expands the company’s total addressable market (TAM).

AI places very different demands on a network than traditional cloud computing. Cloud traffic is highly variable, with workloads of various intensity constantly appearing, moving and disappearing. AI training is more coordinated, but far less forgiving. Thousands of expensive accelerators must move tremendous amounts of data simultaneously, meaning a single congested network link slows an entire training run.

AI inference and agentic AI workloads (setting trained AI models to work) adds another layer of networking complexity. Inference  stresses both traditional cloud networks and dedicated AI networks simultaneously. It makes network reliability, congestion control and observability even more valuable, playing directly to Arista’s core strengths.

AI is a Multi-layered Networking Market

AI networking is not a single market, and Arista is unlikely to compete equally across every segment, but is likely to benefit from the largest segments.

The Scale-Up Network

This “scale-up” domain connects tens of AI accelerators together within a single AI datacentre rack (read: a single server). This is today largely Nvidia’s market as it can tightly bundle networking into its rack-scale systems. This may be a future opportunity for Arista as NVIDIA faces more competition.

The Scale-Out Network

Scale-out networking connects servers to build AI clusters. These clusters can scale to hundreds of thousands of AI accelerators (soon, millions are likely). This is where Arista has a meaningful competitive position. Arista’s raw speed, congestion management, observability, and reliability will see it earn its keep. It is also where an open ecosystem and supply-chain resilience are of paramount importance to customers. And especially useful, customer want to avoid becoming dependent on a single vendor (read: Nvidia). Scale-out already represents around 15% of Arista’s business and is growing very rapidly. This is likely to be the largest AI networking market opportunity in general and for Arista specifically.

The Scale-Across Network

The third leg of the AI-network stool is scale-across networking. This connects multiple data centres into a single AI environment. This unprecedented scale is necessary to serve AI systems that burst the seams of a single facility. Compute infrastructure that is increasingly constrained by power, land, cooling and permitting, will also increase the need for scale-across networking.

It is a tremendously network. But the multi-data-centre-coherence problem looks much like the large, distributed, open environments Arista already understands. Although the nascent scale-across business accounts for around 5% of revenue today, it is growing at triple digit rates. Arista strengths in hyperscale networking, EOS software and large-scale network management should put it in prime position to grow strongly.

A Bigger Opportunity

The combined AI networking market is likely to approach $100 billion, roughly matching the size of Arista’s entire existing addressable market. Management expects AI networking to become a multi-billion-dollar business for Arista in 2026, accounting for around 30% of revenue, with scale-out and scale-across driving much of that growth.

And yet AI is only part of the story. Arista has two additional growth vectors. Traditional Cloud networking continues to become larger, faster and more complex, and Arista remains a leader in the highest-speed parts of that market. AI Inference demand may accelerate investment in traditional cloud infrastructure. This remains a large majority of Arista’s business today.

Meanwhile, as Arista takes its software-led hardware differentiation into corporate networks, its Enterprise business offers a long-term growth path.

The growth algorithm looks something like:

• Cloud: ~65% of revenue, around 35% market share and growing at just over 10% per annum.

• AI: ~20% of revenue today, with significant runway as AI networking expands; Should double in 2026.

• Enterprise: ~10% of revenue, approximately 4% market share and presently growing around 50%.

The common thread is consistent across each of these markets: as networks become larger, faster and more complex, Arista’s value proposition becomes stronger.

Attractive Fundamentals

Arista’s financial metrics are extraordinary. Gross margin exceed 60%, non-GAAP operating margin is in the mid-to-high 40s, cash conversion averages around 115% of net income, and the company holds approximately $12 billion (~6% of its market capitalisation) of cash on its fortress balance sheet. Despite the lazy cash, return on invested capital – the truest measure of profitability – is approximately 30% (and around 180% stripping it out, which gives a sense of the profitability of its operations).. These are not the economics of a commodity hardware business. Rather, they reflect a company whose software and systems engineering command premium pricing while delivering meaningful value for customers.

Arista’s management is exceptional. Its original deeply technical founders remain with the company, helping preserve the engineering-led culture that has underpinned its success. Dotcom scar tissue has made management remarkably disciplined - Arista respects shareholder capital, runs a lean operation, and repurchases stock only when it makes economic sense. The company prefers to grow organically, with only small acquisitions that have been capability-led and funded from cash.  

Competition Remains Intense

That culture and managerial discipline matter because Arista operates in one of the most competitive segments of enterprise technology. The competitive set is formidable: Cisco remains a formidable competitor with an enormous installed base, Nvidia is investing aggressively in AI networking, and hyperscale customers continue to develop some networking capabilities in-house.

Moreover, many of those hyperscalers are also Arista’s customers. None of this should be waved away. But Arista has earned credibility by consistently solving the most technically demanding large-scale networking challenges. AI is simply the latest chapter in that story.

A Premium Worth Paying?

Arista is not an obviously ‘cheap’ stock, trading at approximately 46 times its forward earnings, compared with a decade-long average of around 33 times. For good reason, the stock has never looked a bargain.

But it was. Over the past decade, Arista has compounded revenue and earnings per share at approximately 27% and 40% per annum respectively. We have owned Arista Networks since the inception of our Global Equity Fund, and it has been a phenomenal investment. The share price has compounded at more than 40% across any meaningful investment time horizon. The long-term driver of shareholder returns has been earnings growth. The share price has simply followed the growth in the business.

And that engine has not stalled. Arista grew revenue by approximately 29% in 2025 and is guiding to approximately 28% growth in 2026 – a conservative estimate considering it is developing next-gen networks with some of AI’s biggest spenders, including hyperscalers, OpenAI and Anthropic. Revenue for new products, new customers and AI deployments should grow strongly over a multi-year horizon. There is every chance consensus still understates the company’s growth trajectory.

Arista continues to solve one of the most important bottlenecks in cloud and AI infrastructure, and we believe the opportunity ahead of the company is larger than the one it has already captured.

We therefore take comfort that we are not overpaying for an exceptional business. We believe Arista Networks provides a compelling long-term home to considerably grow investors’ capital.