Arista Networks and Arm Are Not Comparable AI Plays — and Confusing Them Costs You

Arm stock dropped 8% after its July 29 earnings beat. Arista stock jumped the next day after its own blowout quarter. Both reports showed 22% to 38% revenue growth. Both were covered by the same chorus of "AI infrastructure supercycle" headlines. And both companies are now routinely paired in comparisons as if they are playing the same game at different speeds.

They are not. The market is missing the structural distinction between a networking equipment company that is already printing free cash flow at scale and a chip-design company that is 35 years into a pure IP-licensing model and is now attempting its most dangerous transition yet — moving into production silicon.

This is not a revenue-growth comparison. It is a comparison of two companies on opposite sides of the AI infrastructure transition, executing fundamentally different product plays, carrying wildly different risk profiles, and trading at valuations that reflect very different assumptions about execution. Let me walk through what the architecture, supply chain, and earnings data actually say.

Arista: The Ethernet Bottleneck in the AI Scale-Out Fabric

Arista Networks is not selling GPUs. It is selling the Ethernet switches that connect them. When NVIDIA builds a cluster of 10,000 H100 or Blackwell GPUs, those GPUs need to communicate with each other — not just within a rack (where NVIDIA's proprietary NVLink still dominates) but across racks and across data center wings. That is the scale-out and scale-across fabric. That is Arista's territory.

In its Q2 2026 results, reported August 4, Arista hit its first $3 billion quarter — $3.036 billion in revenue, up 37.7% year over year. Adjusted EPS came in at $1.02, up 40% from $0.73 a year earlier. Management raised full-year 2026 revenue guidance for the third time to approximately $12.6 billion, implying 40% growth and a $1.1 billion increase from its own May outlook.

What matters here is not just the top-line number. It is the architecture shift happening inside the AI data center. Ethernet, which Arista runs on, is standardizing for scale-out and scale-across AI networks. NVLink remains dominant for scale-up (intra-rack) networking, but that is a narrowing battlefield as clusters grow from thousands to tens of thousands of GPUs. Arista's EtherLynx AI fabric customer base expanded from 4–5 in 2024 to over 100 cumulative customers as of Q2 2026. The company has set an AI fabric revenue objective of at least $3.5 billion for 2026, roughly 28% of its guided total.

But the broader signal is that growth is no longer confined to back-end AI fabrics. Arista's scale-across routing category — the networking layer that connects AI clusters to each other and to the broader data center — is estimated at $3–$4 billion today, with a $15–$20 billion opportunity by 2030. Campus revenue has an objective of at least $1.25 billion. International revenue jumped to 23% of total, up from 15.5% in the prior quarter. Put plainly: Arista is executing a horizontal strategy where AI/data center wins expand into campus, routing, and front-end data center switching, penetrating budgets historically controlled by Cisco and HPE.

The product pipeline is real, not speculative. The 7060XE7 switch features approximately 100 terabits of switching capacity with 1.6T connectivity and liquid-cooling options. 1.6T networking and liquid-cooled switches are expected to have a production inflection in 2027. Arista favors an open architecture using socketed optical engines and pluggable modules over proprietary co-packaged optics, and management expects the majority of the market to remain pluggable-based through 2028–2029.

Financially, the profile is striking. Arista's operating margin sits at 42.79%, its free cash flow margin at 54.36%, and its return on invested capital at 28%. The company generated $5.16 billion in free cash flow over the trailing twelve months while running capital expenditures of just $151 million. This is a business model where the company ships hardware, takes the order upfront, and converts nearly every dollar of revenue into free cash flow within the operating cycle.

The $12.6 billion revenue run rate would place Arista at approximately a $238 billion market cap, or 22.6 times trailing sales. The stock has gained 43.99% year-to-date. That is rich — but for a company growing revenue at 32.57% with QoQ acceleration of 12.06% and converting more than half of every revenue dollar into free cash flow, the multiple is not unearned.

Arm: The Architecture Play Attempting a Business Model Jump

Now look at Arm through the same architecture lens.

Arm Holdings designs the CPU architecture that powers most of the world's smartphones and is increasingly powering AI data center CPUs. It does not manufacture chips. It does not sell silicon. For 35 years, it has been a pure IP licensing and royalty business with 97.54% gross margins, selling architecture licenses and per-chip royalties to companies like Apple, NVIDIA, Qualcomm, Google, and AWS.

That model is changing. On March 24, 2026, Arm unveiled its first production-ready silicon product in its history — the AGI CPU, a 136-core data center processor built on TSMC's 3nm process, designed specifically for agentic AI infrastructure. Customer demand has exceeded $2 billion across fiscal years 2027 and 2028. Meta is the lead partner and co-developer. Oracle, OpenAI, Cerebras, SAP, Cloudflare, and SK Telecom are among the named customers.

This is the transition that defines Arm's investment case going forward. The AGI CPU delivers more than 2x performance per rack compared to x86 platforms, at a 300-watt TDP, with 6GB/s memory bandwidth per core at sub-100ns latency. Arm estimates up to $10 billion in CAPEX savings per gigawatt of AI data center capacity compared to x86 alternatives.

But here is where the comparison with Arista breaks apart entirely.

Arm's IP licensing business is generating strong growth — Q1 FY27 revenue of $1.29 billion, up 22% year over year, with data center royalties more than doubling. The ARMv9 architecture and Arm Compute Subsystems (CSS) are pulling in higher royalty rates per chip. Neoverse shipments have surpassed 1.5 billion cores, with the most recent 500 million in just the past nine months. IDC reports that spending on Arm-based accelerated server platforms nearly doubled in the past two quarters, surpassing x86.

Yet Arm's QoQ revenue growth is negative — down 13.49% sequentially. Operating margin is 18.29%, roughly one-third of Arista's. Return on invested capital is 8.4%, compared to Arista's 28%. Free cash flow growth was exceptional at 124% year over year, but that is a low-base recovery effect, not a sustainable trend for a company in the middle of a capital-intensive business model shift.

And then there is the valuation. Arm trades at a $301.8 billion market cap — 58.5 times trailing sales and 289 times trailing earnings. The forward P/E is approximately 580x. Put plainly: the market has already priced in years of flawless execution on the AGI CPU transition, the continued erosion of x86 in data centers, and Arm capturing a significant share of an expanding CPU TAM. A stock that has gained 158.5% year-to-date and 102.5% on a rolling annual basis has a very high bar to clear.

The 8% post-earnings selloff on July 29 — despite beating both revenue and EPS estimates, with guidance above consensus — is the market's way of saying: the IP story is strong, but the silicon transition is unproven, and the valuation assumes success.

The Supply Chain Signal Layer

This is where the two companies diverge most sharply — and where the supply chain tells a different story than the earnings headlines.

Arista has committed $9.7 billion to supply chain expansion. That is a massive commitment relative to a company running $12.6 billion in annual revenue. It means Arista is placing non-cancellable procurement orders for switches, components, and optical modules well in advance of confirmed delivery. The upside: AI networking demand remains materially above available supply, and Arista is locking in capacity to capture it. The downside: if hyperscaler ordering patterns shift, that $9.7 billion becomes leverage risk — inventory that's hard to redirect, working capital tied up, and margin pressure if the cycle turns.

This is the dual signal I always watch. Surging supply commitments are both a demand confirmation and a leverage warning. At $9.7 billion on a $12.6 billion revenue run rate, Arista's commitment-to-revenue ratio is approximately 77%. That is high for a company in a cyclical industry.

Arm's supply chain story is fundamentally different. As a licensing company transitioning into silicon, its supply risk is not about procurement commitments — it is about execution risk. The AGI CPU depends on TSMC's 3nm capacity, on Meta's deployment schedule, on the software ecosystem catching up to a new silicon baseline. Arm has secured manufacturing capacity for a $1 billion opportunity across FY27 and FY28. That is real, but it is also small relative to the $300 billion valuation that now depends on this product succeeding.

What separates these two supply chain stories: Arista's risk is concentrated in timing — can it convert $9.7 billion in supply commitments into revenue before customer ordering slows? Arm's risk is structural — can a company that has never manufactured chips build a silicon business that moves the needle on a $300 billion market cap?

So Where Does Capital Go?

The debate is not whether both companies are important to the AI infrastructure buildout. They are. The debate is whether the return profile of each is still compelling at these levels — and whether one offers a better risk-adjusted path than the other.

Arista is growing revenue at 33% with 43% operating margins and 54% free cash flow margins. It has 100+ AI fabric customers, a broadening TAM that includes scale-across routing, campus, and front-end switching, and a product roadmap that inflects to 1.6T in 2027. The $238 billion market cap at 22.6x sales is expensive but reflects a business that is already generating $5 billion in annual free cash flow with visible growth acceleration. The $9.7 billion supply chain commitment is the thing to watch — if that ratio gets much higher without commensurate revenue realization, it becomes a leverage issue. But right now, demand is materially above supply, and Arista is positioned to capture it.

Arm is at a different stage entirely. The IP licensing story remains strong — Arm-based server spending is now outsizing x86, and the AGI CPU transition is the next logical step. But the company is in the early innings of a hardware transition it has never attempted, trading at a valuation that prices in success, and showing negative sequential revenue growth. The 158.5% year-to-date return means much of the thesis is already digested by the market.

I believe Arista is on the right side of the current AI infrastructure transition — the shift from scale-up (intra-rack, NVLink-dominated) to scale-out and scale-across (Ethernet-dominated). That is a transition where Ethernet's open architecture wins as clusters grow beyond the reach of proprietary interconnects. Arm is also positioned for a real structural shift — the erosion of x86 in data centers and the rise of Arm-based CPU capacity for inference and agentic AI workloads. But Arm's valuation is already reflecting a $15–$18 billion AGI CPU run rate by 2031, according to analyst estimates, and the company has not yet shipped meaningful silicon volume.

Where does capital go? Arista, at a smaller market cap with demonstrated cash flow generation and a product roadmap that aligns with the Ethernet standardization trend, offers a more defensible entry point. Arm requires conviction that the silicon transition succeeds flawlessly, that the AGI CPU captures significant share from x86, and that the market will continue to pay 58x sales for a company that is still negative on QoQ revenue growth. That is not impossible — but it is a higher bar.

The break condition for Arista is straightforward: if hyperscaler ordering slows and the $9.7 billion in supply commitments starts turning into inventory risk, the multiple compresses. For Arm, the break condition is whether AGI CPU shipments fail to accelerate beyond the $2 billion demand window or whether the broader x86-to-Arm transition stalls. Either outcome at a $300 billion valuation is painful.

In my view, the opportunity cost calculus favors Arista today. Not because Arm is a bad company — but because the market has already paid for Arm's best-case scenario and is still pricing Arista as if the AI networking buildout has just begun.

The transition from training-dominated to inference-dominated AI compute architecture is the broader frame here. Both companies benefit from it. But one is already delivering cash flow at the bottleneck; the other is asking you to trust that it can build a silicon business from scratch while trading at a valuation that assumes it already has.