Arista vs. Arm: One Is Already Cashing In on AI Infrastructure. The Other Is Betting on It.

Arista Networks reported its first $3 billion quarter on August 4, raising full-year guidance to $12.6 billion for the third time this year. Five days earlier, Arm Holdings beat quarterly estimates and said demand for its new AGI CPU had surpassed $2 billion — then watched its stock slide 7% anyway.

Both companies are riding the same AI infrastructure wave. One is selling the pipes that keep data centers running today. The other is betting its future on a chip that won't ship at volume until the second half of this year. The growth trajectories look similar at a distance — Arista at 33% year-over-year revenue growth, Arm at 25% — but the architecture of their businesses, the timing of their AI exposure, and the capital allocation they represent are fundamentally different.

The market seems to be treating them as interchangeable "AI growth stocks." It shouldn't.

The Product Architecture Question

Let me start where I always do with these comparisons: not valuation, not growth rates, but product architecture. The question is whether each company sits on the right side of the current infrastructure buildout, and whether its competitive position today translates to durable cash flow.

Arista Networks builds Ethernet switching fabric — the physical networking layer that connects GPU clusters within and between data centers. When hyperscalers deploy thousands of Nvidia GPUs, those chips communicate through NVLink inside a server rack, but the traffic between racks and between data centers flows through Ethernet. Arista owns that Ethernet switching layer with platforms like the 7060XE7, which delivers approximately 100 terabits of switching capacity with 1.6T connectivity and liquid-cooling options.

Here's what matters: Arista management told investors on the Q2 2026 call that AI networking demand remains "materially above available supply". Their Etherlink AI fabric has expanded from 4–5 customers in 2024 to more than 100 cumulative customers. AI fabric revenue is targeted at at least $3.5 billion for 2026 — roughly 28% of total guided revenue. And the company sees a $15–20 billion scale-across routing market by 2030, which refers to connecting geographically distributed AI clusters — a structural need driven by power and cooling scarcity, not a discretionary add-on.

Arm is a chip architecture company transitioning into a chip company. For 35 years, Arm licensed its processor designs and collected royalties from companies like Apple, Qualcomm, Samsung, and MediaTek. In March 2026, it unveiled the AGI CPU — its first internally designed silicon product — built on TSMC's 3-nanometer process and designed for AI inference orchestration, specifically what Arm CEO Rene Haas calls "agentic AI" workloads that run with minimal human oversight.

The AGI CPU demand picture looks compelling on paper: over $2 billion in customer demand across fiscal 2027 and 2028, Meta as the co-design lead customer, with OpenAI, Cloudflare, SAP, and Oracle among early buyers. Volume production starts in H2 2026. Arm projects $15 billion in annual revenue from this chip by 2031 — which would represent a six-fold increase from the $4 billion in total revenue the company generated in 2025.

This is what separates the two investment cases. Arista's AI revenue is real, accelerating, and flowing through the income statement now. Arm's $15 billion AGI opportunity is real but back-end loaded — it requires the company to execute on a business model it has never operated, competing with former customers (Amazon, Microsoft, Google, Nvidia) in the data center silicon market. The AGI CPU is a five-year revenue curve that Arm needs to compress into a believable near-term growth narrative.

The Financial Proof Layer

Let me lay the numbers side by side.

Metric	Arista Networks	Arm Holdings
TTM Revenue	~$10.5B	~$5.15B
Revenue Growth YoY	32.6%	25.1%
Revenue Growth QoQ	+12.1%	-13.5%
Gross Margin	63.5%	97.5%
Operating Margin	42.8%	18.3%
FCF Margin	54.4%	19.3%
ROIC	28.0%	8.4%
Free Cash Flow TTM	$5.2B	$1.5B
Market Cap	$238B	$302B
P/S (TTM)	22.6x	58.5x
P/E Forward	70x	580x
Stock YTD Return	+44%	+159%

Two things jump out immediately.

First, Arm's gross margin of 97.5% is the IP licensing business — it costs virtually nothing to hand out chip designs. Once Arm starts manufacturing and selling its own silicon, those margins collapse toward the 50% gross profit that CEO Rene Haas has already disclosed for the AGI CPU. So the 97.5% margin you see today is the margin Arm is about to sacrifice in pursuit of higher absolute revenue. That's not inherently a bad trade — the question is whether the revenue growth justifies the margin compression.

Second, Arista's free cash flow margin of 54.4% means more than half of every revenue dollar becomes cash. That's extraordinary for a hardware-adjacent business. Combined with a 28% return on invested capital, Arista is generating capital at a rate that would make most software companies envious. The company produces $5.2 billion in free cash flow against a $238 billion market cap — roughly a 2.2% FCF yield. That's not cheap by value standards, but it's real cash being produced today, not a forward promise.

Arm generates $1.5 billion in FCF against a $302 billion market cap — a 0.5% FCF yield. At 580x forward earnings, the stock is pricing in exceptional execution across the AGI CPU ramp, continued IP licensing growth, and margin expansion that the transition to silicon manufacturing threatens to undermine. The market has already priced this stock as if the $15 billion AGI revenue by 2031 is a near-certainty.

The Sequential Growth Story

Revenue growth rates tell you where the companies are. Sequential momentum tells you where they're heading.

Arista's QoQ revenue growth is +12.1%. Q2 2026 came in at $3.04 billion versus Q1 at $2.71 billion — and the Q3 guide of ~$3.3 billion implies another ~10% sequential acceleration. Full-year guidance has been raised three times, from $10 billion in late 2025 to $11.5 billion in May, to $12.6 billion in August. The company is consistently missing its own guide on the upside.

Arm's QoQ revenue growth is -13.5%. The company went from $1.46 billion in Q4 to $1.29 billion in Q1 FY2027, then guided Q2 to $1.33–1.43 billion — a modest recovery at best. Management attributed the quarterly weakness to smartphone royalty deceleration, calling it the "only real weakness". Smartphone royalty growth is forecast at 10–15% for Q2, down from the high-teens and low-20s the market has grown accustomed to.

Put plainly: Arista's growth engine is accelerating while Arm's is experiencing a speed bump in its cash cow business. Arm expects the AGI CPU to fill the gap — but volume production doesn't start until H2 2026, and the company has never sold silicon at this scale. The demand pipeline of $2 billion across two fiscal years sounds impressive, but that's roughly one and a half quarters of total revenue for a company currently doing $5 billion annually.

The Market Structure Shift

The broader frame here is that AI infrastructure spending is splitting into distinct layers with different competitive dynamics, different timing, and different margin profiles.

Layer 1: Scale-out and scale-across networking (Arista's domain) is winning because Ethernet is cheaper, more flexible, and more open than proprietary interconnects like InfiniBand. Arista's management has conviction that NVLink will dominate scale-up (inside a server) for the next 2–3 years, but Ethernet controls the scale-out and scale-across markets. These markets are growing fast because power and cooling constraints force clusters to be distributed across multiple buildings and regions. The scale-across market alone — connecting geographically separated data centers — is estimated at $3–4 billion today with a $15–20 billion opportunity by 2030.

Layer 2: CPU orchestration for inference (Arm's AGI play) addresses a real bottleneck. As AI workloads move from training to inference, the CPU layer that orchestrates thousands of API calls, manages agentic workflows, and coordinates multi-model pipelines becomes critical. Nvidia's GPUs process the models, but the CPU handles the orchestration. Arm's architectural advantage — power efficiency per instruction — makes its designs attractive for data center operators managing energy costs. Over the last six years, Arm has shipped 1.5 billion data center cores, with 30% of those shipped in the last nine months alone.

But the inference market is also the one where Arm faces the most structural ambiguity. The AGI CPU is Arm's first attempt to sell silicon directly — and its initial customers include companies (Meta, OpenAI, Google, Amazon) that have their own custom silicon roadmaps. Arm needs these same customers to buy its chips instead of building their own. That's a fundamental tension in the business model that doesn't show up in the growth table.

The Valuation Gap

This is where the comparison gets uncomfortable.

Arista trades at 22.6x trailing sales and 70x forward earnings. That's not cheap — but for a company growing revenue 33% annually, generating 54% free cash flow margins, and returning 28% on invested capital, the multiple is supported by cash flow that exists today. The PEG ratio sits at 2.4x, which is elevated but not absurd for a compounder.

Arm trades at 58.5x trailing sales and 580x forward earnings. The stock is up 159% year-to-date, up 125% over 120 days, and has moved from a $100 low to a $453 high over the past year. At 580x forward earnings, the stock requires five years of flawless execution — the AGI CPU ramp must hit plan, licensing royalties must stay strong despite smartphone softness, and the transition to silicon manufacturing must not crush margins more than the market expects.

The $64 billion gap in market cap between Arm ($302 billion) and Arista ($238 billion) says the market believes Arm's IP moat and the AGI CPU optionality are worth 2.7x more than Arista's cash-generating networking business. That's a view, not a fact. And it's a view that gets tested every quarter as the AGI CPU ramp materializes — or doesn't.

The Risk That Matters

Both companies carry real risk. Let me be specific about what I'm watching.

Arista's risk is customer concentration and supply chain leverage. At least one — possibly two — customers now exceed 10% of annual revenue. If a hyperscaler shifts ordering patterns, delays a deployment, or pivots to in-house networking, revenue volatility follows. The company also carries elevated non-cancellable procurement obligations as it locks in switch silicon and optics supply ahead of demand. These commitments are a dual signal: they reflect demand confidence but also create leverage risk if demand softens.

Arm's risk is execution on a business model it has never operated. The company is asking the market to believe it can design, manufacture, sell, and support data center silicon at a competitive price point while maintaining its IP licensing relationships with customers who are also its competitors. The smartphone royalty slowdown in Q1 — which management dismissed as the "only real weakness" — is worth watching because it could compound with margin pressure from the silicon transition. And at 580x forward earnings, even a modest execution slip gets punished severely.

Where Capital Goes

Here's my judgment.

Arista is the compounder that's already working. The growth is accelerating, the margins are exceptional, the free cash flow is real, and the scale-across market expansion gives the company a second growth curve beyond AI fabrics. The stock trades at a premium, but the premium is backed by $5.2 billion in annual free cash flow and 28% ROIC. If you're looking for an AI infrastructure play where the earnings power is already visible and the growth trajectory is accelerating, Arista fits.

The question for Arista isn't whether the thesis is broken — it isn't. The question is whether the return profile from here justifies the allocation. At 70x forward earnings, much of the near-term upside requires continued guide-raises and flawless sequential growth. The stock has returned 44% year-to-date. I believe the long-term trajectory is strong, but I'd be cautious about chasing a stock that's already had a strong run, especially when the AI fabric market is only beginning to see 1.6T networking and liquid cooling inflection — which Arista management itself expects to ramp in 2027.

Arm is the optionality play. The AGI CPU is a legitimate attempt to monetize Arm's architectural advantage in the inference orchestration layer. If the $15 billion by 2031 projection is even half right, the stock could compound significantly from here. But at 580x forward earnings and 58x sales, the stock is already pricing in most of that success. The market sold off 7% after the last earnings beat because investors are sensing the gap between the $15 billion promise and the $2 billion in near-term demand that's currently visible.

I believe Arm's long-term thesis — that its power-efficient architecture captures a disproportionate share of the inference orchestration market — is compelling. But the return curve is back-half weighted, the business model transition carries execution risk, and the smartphone royalty softness is a real headwind, not a blip. At this valuation, I'd wait for the AGI CPU to start shipping at volume and demonstrate margin accretion before building a large position.

The debate is not whether both companies will benefit from AI infrastructure expansion. It's whether the return profiles — Arista's proven cash generation at a premium price versus Arm's speculative silicon transition at a speculative valuation — still justify equal confidence today.

For me, Arista's cash flow and growth acceleration make it the more defensible allocation in the near term. Arm's AGI CPU story is too valuable to ignore — but it's the kind of opportunity where patience and a smaller position size respect the execution risk. The market has already bid Arm up 159% this year. That kind of move front-loads returns.