Arista Networks, explained for a non-technical reader

Arista is growing about 40% a year, keeps almost half of every sales dollar as operating profit, and holds $13 billion of cash with no debt. Wall Street is almost unanimously positive: 31 of 32 analysts rate it a buy, with an average price target about 37% above today's price.

The catch is what you pay for that. The shares trade at roughly 56 times last year's profit and 38 times next year's expected profit, about double the broad market. At that price the stock gets punished whenever results are merely good instead of great. Twice in the past year a beat on earnings was followed by a drop of 9% to 14%, because the outlook only matched expectations.

Three things deserve a non-technical reader's attention. First, two customers, understood to be Microsoft and Meta, were 42% of 2025 revenue. Second, Nvidia, the AI chip giant, now sells competing network gear and can bundle it with its chips. Third, insiders sold about $1.5 billion of stock in the past twelve months and bought none, mostly under pre-set plans, but the pattern is worth knowing.

Every data center is a warehouse full of computers that need to talk to each other at enormous speed. The boxes that route that traffic are called switches. Arista designs them, writes the software that runs them, and sells them mostly to the biggest technology companies on earth.

Think of a switch as a very fast, very reliable traffic interchange. A single large AI cluster might contain tens of thousands of specialized chips that must constantly share work. If the network between them slows down, the expensive chips sit idle. That is why network gear has become a serious line item in AI spending, and why Arista's growth tracks the AI build-out so closely.

A little history

Arista was founded in 2004 by Andy Bechtolsheim, a co-founder of Sun Microsystems and one of Google's first investors, together with Ken Duda and Stanford professor David Cheriton. Jayshree Ullal, a former Cisco executive, became CEO in 2008 and still runs the company. Arista went public in 2014 and survived a four-year patent war with Cisco that ended in 2018 with a settlement widely reported at $400 million. It overtook Cisco as the largest seller of data-center switches around 2024.

Who runs it

Arista sells hardware, but the recurring parts of the business are growing. Software and support subscriptions are about 17% of revenue. More telling is deferred revenue, which is money customers have already been billed for gear that has not yet been formally accepted. It stood at $6.9 billion at the end of June 2026, up 69% in a year. Large AI projects take months to test and accept, so this is a backlog that will turn into reported revenue over the next couple of years.

Geographically, about three quarters of sales are in the Americas. Arista does not build its own hardware. It designs the products and has contract manufacturers assemble them, which keeps its own costs low but exposes it to tariffs and parts shortages.

Arista's filings say two unnamed customers were 26% and 16% of 2025 revenue. Analysts and the trade press identify them as Microsoft and Meta, which is consistent with earlier years when the filings did name them. This is the single biggest structural risk in the story: when Meta trimmed its spending in 2023, Arista's sales to Meta fell 17% the following year.

Management says the mix is broadening. On the August 2026 call the CEO said she expects "one, maybe two" additional customers to pass 10% of revenue, and that newer AI cloud firms are becoming meaningful buyers. Arista claims more than 10,000 customers in total and over 100 using its AI-cluster designs, up from a handful in 2024.

In the overall data-center switch market Arista and Nvidia are now neck and neck at about 21% each, with Cisco close behind. In the narrower, faster-growing market for networks inside AI clusters (the "back-end"), the industry tracker Dell'Oro ranks Arista third, behind contract manufacturer Celestica and Nvidia, with Cisco fourth and gaining fastest. Arista is still beating its own targets, but the pie is growing faster than Arista is and being divided among more players.

Why customers pick Arista anyway

The repeated answer is software. One operating system across every box, automation that works at the scale of a hundred thousand ports, and reliability. Newer AI cloud companies "don't have staff," in the CEO's words, so they pay for something that just works. Arista also backs open Ethernet rather than any one chip vendor's standard, which matters as customers add AMD and Google chips alongside Nvidia's. Dell'Oro reports Ethernet was more than twice the size of Nvidia's InfiniBand in AI networks in 2025, a bet Arista made early and won.

In office networks the picture is different. Arista beat its $800 million 2025 target and aims for at least $1.25 billion in 2026, but Cisco holds roughly 29% of all Ethernet switching and Arista remains a small challenger there.

Balance sheet

Arista holds $13.3 billion in cash and investments and has no debt. It generated about $5 billion of free cash in the past twelve months. It bought back $1.6 billion of stock in 2025 but none in the first half of 2026, with $818 million of authorization unused. The company pays no dividend.

Two balance-sheet items deserve a plain-language note. Arista has placed $9.7 billion of binding orders with suppliers, up from $6.8 billion at the end of 2025, to lock in scarce chips and memory. That is fine while demand runs hot. If demand reversed, Arista would be holding parts it must pay for. And $616 million of its $2.5 billion inventory is gear sitting at customer sites on trial, which is a normal part of winning large deals but worth watching.

The stock set a record close of $216.74 on 9 October 2026, then fell 18% in a week with no company news. The trigger was a broad scare about whether debt-funded AI cloud firms can pay their lenders, plus hot inflation data. Insider sales near the top added to the pressure. It has since steadied around $177.

What you are paying

Arista is roughly twice as expensive as Cisco and the broad market on next year's profits, and even pricier than Nvidia and Broadcom, which are expected to grow faster still. The market is paying for 40% growth to continue. When growth is that fast the multiple can shrink quickly if the pace slows, which is why the stock reacts so violently to guidance.

What analysts say

Of 32 analysts tracked, 24 rate the stock Strong Buy, 7 Buy, 1 Hold and none Sell. The average price target is about $244, with a range of $190 to $290. No downgrades were found in 2026. The crowd is firmly on one side, which means little room for a positive surprise in sentiment and a long way to fall if the story cracks. Very few investors are betting against it: short interest is under 2% of shares.

No credible source assigns probabilities. The ranking below is a judgement about consequence, with the first two being the ones the market is actively worrying about.

Insiders and governance

Insiders sold about $1.48 billion of stock in the twelve months to 10 October 2026 and bought none on the open market. CEO Ullal sold roughly $154 million on the day after the August results and more near the October peak. Co-founder Bechtolsheim sold at least $61 million in August. President Duda sold in October. Most of these sales run under pre-scheduled plans set months earlier, which weakens any "they know something" reading, but the scale and the timing near highs are a fair flag.

Two governance points. In March 2024 Bechtolsheim settled an SEC insider-trading case over personal trades in another company's stock, paying a $924,000 penalty and accepting a five-year ban on serving as an officer or director of any public company. Arista itself was not charged, and he remains Chief Architect and a 14.6% holder. Separately, only 60% of shareholders approved executive pay at the May 2026 annual meeting, a weak result that usually prompts a board to revisit its pay plan.

Stock-based pay runs about $120 million a quarter and is excluded from the "adjusted" profit figures the company headlines. With no buybacks in 2026, that pay is slowly diluting existing holders.

A fair reading: the bull case is about the business and the bear case is mostly about the price and the customer list. Both can be true at once. The business is excellent. The question for a buyer is whether 40% growth lasts long enough to justify paying double the market multiple, and whether you can stomach 10% to 18% swings on days when nothing fundamental has changed.

1. The 3 November report, and especially the Q4 guide. Revenue above $3.3 billion is expected. The tell is the next-quarter outlook: about $3.6 billion or more would say the AI cloud credit scare has not reached Arista's order book.
2. Customer mix. Any new customer crossing 10% of revenue, or either of the big two shrinking, changes the risk picture more than any product launch.
3. Gross margin. The guide is 62% to 64%. A print below 62% would mean the cost squeeze is worse than planned; above 64% would mean price rises are landing early.
4. Nvidia's networking numbers and any named customer losses. Nvidia reports its own networking revenue each quarter. Watch for a hyperscaler publicly standardizing on Nvidia's switches.
5. Supplier commitments and inventory. If the $9.7 billion of binding orders keeps rising while growth slows, that is the setup for a write-down.