Cerebras's stock pulls back after a blowout opening day

By Britney Nguyen

The chip maker's shares debuted Thursday with a 68% rally, but pulled back at the start of their first full day of trading.

Cerebras Systems shares started trading on the Nasdaq on Thursday.

Shares of Cerebras Systems fell in early trading Friday, to take some of the shine of the the inference-chip maker's blowout debut in the previous session.

The stock's (CBRS) first trade on the Nasdaq stock exchange COMP on Thursday was at $350 - 89.2% above where its initial public offering was priced - for 4.4 million shares at 12:59 p.m. Eastern time, according to FactSet data.

As investors clamored to buy in, with the ensuing volatility triggering a brief trading halt within seconds of the first trade. The stock climbed as much as 108.8% above its IPO price, to an intraday high of $386.34, before paring gains to close up 68.2% at $311.07.

At the closing price, Cerebras's market capitalization was $66.95 billion. On Wednesday, before the IPO priced, the company's market cap was expected to be as high as $34.44 billion.

"The pricing action at least indicates investors are very excited about its prospects, and the potential for Cerebras to become a key player in chip development," Matt Kennedy, senior strategist at Renaissance Capital, told MarketWatch in emailed comments on Thursday morning before the company began trading.

That excitement wasn't enough to keep the rally going, in the face of a chip-sector and broader-market selloff on Friday. The stock slumped 3.6% to start its first full-day of trading, while the PHLX Semiconductor Index SPX shed 3.9% and the S&P 500 index SPX fell 1.3%.

Cerebras is going directly after Nvidia (NVDA), Kennedy said, "with a product that excels at inference, an emerging frontier in AI competitive superiority." And though it's still "a long way off," some investors are betting that the company could potentially take on the chip giant in the future, he added.

The IPO had priced late Wednesday at $185, topping its previously announced expected range of $150 to $160, which had already been raised from $115 to $125. Cerebras had also increased its offering to 30 million shares, from 28 million shares.

"We're so happy that the market has sort of understood our story and responded in the way it has," Cerebras CEO Andrew Feldman told MarketWatch shortly after shares began trading on Thursday.

The company's chips are built to support inference, or the process of running artificial-intelligence models after training, which is increasingly a major focus for companies deploying applications.

Cerebras makes macro chips that it said are 58 times larger than Nvidia's B200 chips. Feldman boasted in an interview on CNBC that, in the 70 years of the computing industry, Cerebras was the only company to succeed at building a big chip.

"Since we did it, others have tried and also failed," Feldman said on CNBC. "So we're really confident that the technical moat is wide and deep."

Investors have been hot for AI-infrastructure plays such as chips and networking, as hyperscalers and other tech companies plan to pour hundreds of billions of dollars into building out AI data centers. And Cerebras doesn't just offer something new - it has also shown that it can make money.

Cerebras reported revenue of $510 million for 2025, which was up 76% from $290.3 million in 2024. The company saw net income of $237.8 million last year, compared with a net loss of $481.6 million in 2024.

Cerebras's approach to using an entire silicon wafer as one chip is known as wafer-scale integration, which the company said allows it "to bring together quantities of compute and memory never before assembled on a single commercial chip," in turn making AI models faster.

With Cerebras, "everything is already connected in one piece, which reduces communication overhead and makes things faster for certain workloads," Deming Chen, a professor of electrical and computer engineering at the Grainger College of Engineering at the University of Illinois Urbana-Champaign, told MarketWatch in written comments.

One advantage of this technique is that it allows for "extremely fast speed due to less data movement," Chen said, as well as "very high parallelism," which refers to a chip completing multiple tasks simultaneously - the same as Nvidia's graphics processing units do.

If there's a disadvantage, it's that this type of chip is expensive and difficult to manufacture. And a larger chip also raises the risk of lower yields, since the possibility of defects increases. A chip yield refers to the amount of usable chips that come from one wafer.

But Feldman told MarketWatch that even at its size, the company's chip can be clustered to reach several exaflops of compute just as Nvidia's and Advanced Micro Devices's (AMD) chips are.

Cerebras's current chip is built on 5-nanometer process technology, which is slightly behind the 4-nanometer node Nvidia's latest Blackwell uses. Feldman said the company has its sights set on the leading-edge 3-nanometer process.

"We know exactly what we need to do to continue to increase throughput and drive speed up," Feldman said.

Feldman also addressed concerns about Cerebras's customer concentration, which now includes OpenAI and Amazon.com (AMZN) Web Services. A majority of the company's revenue has so far been driven by Group 42 Holding and Mohamed bin Zayed University of Artificial Intelligence.

"The way you learn as an organization, to support three large customers or five large customers, is one at a time," he said, adding that the company wants to eventually support a hyperscaler customer.

Like Nvidia, which has "a small number of very big buyers," Cerebras has one so far, Feldman said, and is looking to add more over time.

-Britney Nguyen

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

05-15-26 0946ET