TheChek News Flow

The Takeaway

All the data centers hastily going up around the country for AI need networking pipes and software for moving data around. A few companies specialize in selling the networking gear, but one company, Arista Networks, stands out as undervalued on Wall Street.

Arista has a much higher gross margin—64%—than rival Celestica, whose margin is only 11%. The two companies have been growing at the same rate, although Celestica is projected to grow a bit faster next year. That might explain why Arista is trading at 40 times next year’s expected earnings, a key valuation metric, while Celestica is at about 44, according to S&P Global Market Intelligence. But Arista has the potential to grow faster than it has projected, which makes it the better bet for investors.

Arista is one of a few companies that will benefit indirectly from the AI revolution as a result of big tech firms’ massively expanded capital expenditures for new data centers.

Arista’s biggest customers, for instance, are Microsoft and Meta Platforms, which accounted for 20% and 15% of its sales in 2024, respectively. Both companies have been vocal about their plans to increase spending on data centers for AI. Meta, for one, is on track to lift its spending on capex this year to as high as $72 billion, up more than $30 billion on what it spent in 2024. Its executives said in October that its spending in 2026 would grow by a dollar amount “notably larger” than in 2025. A chunk of the increased spending from both Meta and Microsoft should show up in Arista’s top line over the next two years.

And that’s not all. Arista’s chief financial officer, Chantelle Breithaupt, said last week at an investment conference that the company is working on networking systems for at least four large customers—likely including Microsoft and Meta—which are building out their own AI data centers, each containing over 100,000 GPUs. Broadly speaking, the more AI chips in a data center, the more data it processes—and that requires more switches, such as the ones Arista sells, to move that data around.

In addition to those projects, the company also says it has 30 to 40 other customers it is working with in its AI data center business, including “neocloud” customers, which rent out data centers full of graphics processing units to AI firms.

AI-related revenue is a small but fast-growing portion of the cloud business. Last year, Arista executives didn’t provide a breakdown of how much revenue came from its fledgling AI business, but this year the company expects to make $1.5 billion from that segment—about 17% of the total—nearly doubling to a projected $2.75 billion next year, when it will be 26% of the total.

Arista’s problem is that the rest of its revenue is barely growing. The company generates most of it from selling networking hardware to other firms for use in traditional cloud data centers. It also installs networking gear and related software in office buildings and other workplaces. It has forecast rapid growth from office sales, a smaller and newer business than traditional data centers, where growth is sluggish.

As a result, even with the surge in AI and office building revenue, the company only expects revenue to grow a projected 20% to 22%, which is lower than the anticipated 27% growth rate this year.

Arista bulls have a simple explanation for why they’re not worried about the company’s apparent revenue slowdown next year: Arista’s projections are always too low.

In recent years, its revenue growth has exceeded its executives’ initial guidance by anywhere between 9 and 19 percentage points, according to UBS hardware analyst David Vogt.

With that in mind, Vogt thinks Arista has a strong chance of beating the 2026 revenue target its executives outlined. He suggested that the company would likely raise its guidance in February to reflect that possibility, adding that it hasn’t done so yet because AI revenue is so new for Arista that it’s not yet clear exactly how long customers will take to formally sign off on new installments, which is necessary for the company to recognize that revenue.

“Based on every metric that we track, along its supply chain, its balance sheet, inventory, purchase commitments, et cetera…they’re going to kill that [2026 AI revenue] number,” he said.

Another indicator of Arista’s promise is the 87% year-over-year increase in deferred revenue for the nine months to September, which amounts to $4.7 billion the company expects to generate but has not yet recognized. The strength of that pipeline suggests Arista’s projection that it will grow its top line by 20% in 2026 to $10.5 billion could be an underestimate.

Then there’s its fat gross profit margin, largely due to Arista selling systems it designs and builds on its own. Celestica, on the other hand, co-develops more of its hardware and software in tandem with its customers, which means it can’t keep as large a share of the profits.

To some observers, a recent sell-off of Arista stock presented a buying opportunity. Arista shares fell 16% after its third-quarter earnings call last month, when its executives projected its fourth-quarter revenue would be roughly in line with the $2.3 billion it generated in the third quarter.

“The guidance implies nothing beyond the fact that management bears a cautious view as to its ability to supply demand, in our view,” Morningstar analyst William Kerwin wrote in a report last month, noting that he is optimistic about the firm’s ability to “increase supply more quickly and beat its targets as it’s done in the past.

“Investors should be comfortable with the high multiple on the stock given the immense growth opportunity before the firm, as well as its market dominance,” Kerwin wrote.