Pieter Slegers — Can Arista Networks win the AI race? (Compounding Quality)

One-line take: the most useful issue in the batch for seeing the method, because the whole 15-step checklist is run in public on a single name and then the name is declined anyway. Arista at $145.9 and a $179.9bn market cap passes almost everything: 63.5% gross margin, 41.0% ROIC, a net cash position, CAPEX at 1.1% of revenue, a 38.3% net margin, FCF at 142.9% of net income, revenue compounding 26.9% over five years and free cash flow 46.7% a year over a decade, and a 36.7% CAGR since the 2014 IPO. Two things fail. Stock-based compensation at 12.6% of net income (16.1% on a five-year average) is called "a red flag". And the valuation: 37.7x forward against a 32.1x ten-year average, with the reverse DCF requiring 18.9% annual free-cash-flow growth for a decade just to deliver 10% a year. The Earnings Growth Model still returns 11.0%, so two of the three valuation methods disagree — and the tie is broken by judgement, not by the model: "Arista Networks is a great business but we don't like the valuation level the company is currently trading at. We're happy to stay patient and wait for better opportunities." The strategic argument is a pick-and-shovel one — Arista connects Nvidia's GPUs over Ethernet while Nvidia pushes its own more expensive InfiniBand — and the standing risk is concentration: 35% of revenue from Microsoft and Meta alone. Total Quality Score 7.9/10.

1. Stocks & names mentioned

A single-name deep dive; the other companies appear as customers or competitors within the Arista analysis and carry no independent view. Research legend: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis. Written post with no timestamps — the At link opens the article.

Ticker	Name	Research	View	What he said	At
ANET	Arista Networks	QT · SA · STK · FA	Neutral	The subject of the dive, and a pass on price. "Arista Networks is a great business but we don't like the valuation level the company is currently trading at. We're happy to stay patient and wait for better opportunities in the market." Quality is not in doubt — Total Quality Score 7.9/10, gross margin 63.5%, ROIC 41.0% (5-yr avg 49.1%), ROE 28.8%, net margin 38.3%, FCF/net income 142.9%, net cash, CAPEX/revenue 1.1%, R&D/revenue 13.5%, revenue +26.9% and EPS +34.6% a year over five years, FCF +46.7% a year over ten, 36.7% CAGR since the 2014 IPO (4,089.4% total against SPY's 396.8%). The two failures: SBC at 12.6% of net income, "a red flag"; and 37.7x forward PE against a 32.1x ten-year average, with the reverse DCF demanding 18.9% FCF growth for ten years for a 10% return. Earnings Growth Model: 11.0%. Called an Owner-Operator Stock — insiders hold 17.1%, CEO Jayshree Ullal 2.2% ($4.8bn), co-founder Andreas Bechtolsheim 14.6%. Risks named: 35% of revenue from Microsoft and Meta, technological disruption, and inventory built for long lead times.	read ↗
NVDA	NVIDIA	QT · SA · STK · FA	Neutral	Framed as both partner and rival, and the framing is the thesis: "Although Arista and Nvidia collaborate closely, they're also becoming competitors. Arista connects Nvidia GPUs via Ethernet, while Nvidia promotes its own expensive InfiniBand. As AI grows, Arista's affordable solution becomes more attractive." No view on Nvidia itself.	read ↗
CSCO	Cisco Systems	QT · SA · STK · FA	Neutral	Named as "Arista's main competitor" and the source of the share Arista is taking: "Arista outperforms Cisco with simpler, more reliable software that make updates and automation easier. Once a customer starts using Arista's products, they almost never switch." Also the CEO's former employer. No independent view.	read ↗
MSFT	Microsoft	QT · SA · STK · FA	Neutral	A customer, and half of the concentration risk: "35% of the revenue comes from two companies, Microsoft and Meta, making Arista vulnerable to changes in their spending." Also cited among the hyperscalers "investing billions into AI infrastructure".	read ↗
META	Meta Platforms	QT · SA · STK · FA	Neutral	The other half of the concentration risk — Microsoft and Meta together are 35% of Arista's revenue — and one of the hyperscalers named as investing billions in AI infrastructure. No independent view.	read ↗
GOOGL	Alphabet	QT · SA · STK · FA	Neutral	Listed among Arista's customers ("Microsoft, Meta, Google, Oracle…") and among the hyperscalers driving the end-market tailwind. No view here — Alphabet had been added to the rated Buy list a week earlier, on 7 May.	read ↗
ORCL	Oracle	QT · SA · STK · FA	Neutral	Named once, in the customer list — "Microsoft, Meta, Google, Oracle,…". No view.	read ↗

Two things worth flagging. (1) The three valuation methods disagree and the verdict follows the strictest. Forward PE fails (37.7x vs a 32.1x average), the Earnings Growth Model passes (11.0% against a 10% bar), and the reverse DCF is left as an open question mark — 18.9% required FCF growth against a decade of 46.7% actual. The written conclusion sides with the failing test. (2) The SBC red flag is raised and then not applied. The text says the 12.6% figure "is a red flag" and "we will take this into account in our valuation section later"; the reverse DCF does subtract the $467m of SBC from forward free cash flow, but the headline 37.7x forward PE is left unadjusted — the same asymmetry noted for FICO on 7 May and corrected for FICO a week later on 21 May.

2. Talking points

The pick-and-shovel framing

"Think of Arista as a pick-and-shovel business in the booming AI industry. During the gold rush, the people who got rich weren't the gold diggers... but the ones selling the shovels."
The same argument used for Brookfield a week earlier, applied one layer up the stack: not land and power, but the switches and routers that move the data.
Revenue split: products 84.1% (switches, routers, plus the EOS and CloudVision software) and services 15.9% (maintenance and premium support).

The Ethernet-versus-InfiniBand bet

The clearest competitive claim in the piece: Arista connects Nvidia's GPUs over standard Ethernet, while Nvidia promotes its own, more expensive InfiniBand. "As AI grows, Arista's affordable solution becomes more attractive."
This makes Arista simultaneously Nvidia's partner and its competitor — a structural tension worth tracking, because Arista's biggest customers are also the ones with the scale to standardise on either.
Against Cisco the claim is software: simpler, more reliable, easier to update and automate, and once installed customers "almost never switch".

Quality: fourteen greens

Moat proxies both clear the bar comfortably — gross margin 63.5% against a 40% threshold and ROIC 41.0% against 15%.
Balance sheet: essentially no debt (no meaningful interest-coverage ratio), a net cash position, goodwill just 1.9% of assets.
Capital intensity: CAPEX 1.1% of revenue and 2.0% of operating cash flow. R&D is the real spend at 13.5% of revenue — inside the 10–20% target band — which is the point: "Arista is a capital-light company that grows through innovation."
Capital allocation and profitability: 5-year average ROE 28.8%, ROIC 49.1%, net margin 38.3%, and free cash flow at 142.9% of net income.

The one quality failure: stock-based compensation

SBC at 12.6% of net income, against a stated preference for under 10% — and 16.1% on a five-year average.
Stated plainly: "Arista Networks uses quite some Stock-Based Compensation. This is a red flag. We will take this into account in our valuation section later."
It is taken into account in the reverse DCF — $467.0m subtracted from forward free cash flow — but not in the headline forward multiple.

The growth record, and the growth required

History: revenue +26.9% (5-yr) and +24.0% (10-yr); EPS +34.6% and +34.1%; free cash flow +46.4% and +46.7%.
Forecast: revenue +25.9% and EPS +20.7% over two years, long-term EPS +18.5% — with the caveat attached, "making long-term predictions is very difficult, and analysts are often too optimistic."
The reverse DCF turns the forecast into a hurdle: starting from $5,130.0m of expected free cash flow, less $467.0m of SBC, plus $26.5m of growth CAPEX = $4,689.5m in year one, from which 18.9% annual FCF growth for ten years is needed to earn 10%.
The honest sentence: "Although Arista Networks grew Free Cash Flow at a 46.7% CAGR over the past decade, it's uncertain if this growth rate can be sustained."

Customer concentration as the standing risk

35% of revenue from two customers, Microsoft and Meta. Note the second-order problem: those are also the customers most able to design their own networking or switch standards.
Long lead times force extra inventory — fine while demand grows, expensive if it stops.
The TAM claim on the other side of the ledger: management sees $105bn by 2029, across Core (data centre and cloud), Cognitive Adjacency (campus and routing) and Cognitive Network (software and services).

The verdict, and how it was reached

Total Quality Score 7.9/10 — and a pass. "Arista Networks is a great business but we don't like the valuation level."
What decided it was the multiple, not the business: 37.7x forward against a 32.1x ten-year average, at a moment when the same author was arguing quality is on sale across 49 other names.
The discipline is worth noting in both directions. Declining a wonderful business on price is the stated policy; it also means the AI exposure in this portfolio stays indirect, via infrastructure rather than the networking layer.

3. In plain English

A jargon-free summary of the thesis behind each argued name. (Renders on each name's consolidated page.)

ANET — Arista Networks Neutral

Arista makes the switches and routers that move data between the computers inside a data centre. Every AI system and every cloud service depends on that plumbing, and Arista sells it to Microsoft, Meta, Google and Oracle. It is the gold-rush shovel seller: it gets paid whichever AI company wins.

By almost every measure of business quality it is exceptional. It keeps 63.5 cents of gross profit on every dollar of sales, earns a 41% return on the money invested in the business, has more cash than debt, spends barely 1% of revenue on physical equipment, and turns more cash into free cash flow than it reports as profit. Sales have grown about 27% a year for five years and free cash flow nearly 47% a year for ten. Since listing in 2014 the shares are up more than 4,000%, against under 400% for the S&P 500.

Two things stop it being bought. First, staff are paid heavily in shares — 12.6% of profit, and 16% averaged over five years — which quietly transfers value from existing shareholders and is called a red flag. Second, the price. The shares cost 37.7 times next year's expected profits against a ten-year average of 32.1, and working backwards from today's price the company would need to grow its cash generation by 18.9% every year for a decade just to deliver a 10% annual return.

So the conclusion is a wait, not a rejection: "Arista Networks is a great business but we don't like the valuation level the company is currently trading at. We're happy to stay patient and wait for better opportunities."

The risk to watch if it ever does get cheap: 35% of the revenue comes from just Microsoft and Meta, and those two customers are large enough to build or buy alternatives.

Summary derived from the archived Compounding Quality post (text in transcript.txt) for personal study. Not investment advice. © Compounding Quality / Pieter Slegers for source material.