Competition in the artificial intelligence (AI) chip market is about to ramp even higher. In mid-April, Cerebras Systems filed an S-1 registration statement with the Securities and Exchange Commission (SEC) announcing plans for its long-awaited initial public offering (IPO). At the time, the start-up hadn’t provided all the key details for investors about its public debut.
Cerebras has taken a novel approach to AI chipmaking, which may put the company on a collision course with Nvidia (NVDA +0.04%), whose graphics processing units (GPUs) currently dominate the AI chip market.
In a subsequent regulatory filing after the market close on Monday, Cerebras filed in some of the blanks.

Image source: Getty Images.
Shares and pricing
In an updated prospectus, Cerebras provided key details about its public debut. The company plans to offer 28 million shares of stock priced between $115 and $125 each, depending on investor demand. At the high end of the range, Cerebras could raise as much as $3.5 billion. Moreover, the company would be worth as much as $26.6 billion, nearly 16% higher than the $23 valuation it fetched in February.
The company has also granted the underwriters the option to sell an additional 4.2 million shares in the event of an over-allotment — or higher-than-expected demand. In that case, the company could raise an additional $525 million, bringing total funds raised to $4.025 billion.
While the details have yet to be finalized, Cerebras is expected to price its IPO on May 13, with the stock to begin trading the following day, according to Bloomberg. The company plans to list on the Nasdaq under the ticker “CBRS.”
A new approach to AI chipmaking
Cerebras has taken a different approach to chipmaking. Instead of cutting a silicon wafer into hundreds of smaller chips, the company uses the entire wafer to create the Wafer-Scale Engine (WSE). This giant semiconductor is 58 times larger than Nvidia’s B200 AI chip. The WSE boasts 900,000 compute cores, “19 times more transistors, 250 times more on-chip memory, and 2,625 times more memory bandwidth” than the B200.
One of the biggest issues inherent in AI processing is that of latency, or the time delay that occurs while data is transferred between the various chips that make up AI systems. By keeping all the processing on a single chip, Cerebras say it has solved the latency issue, since “communications is thousands of times faster on-chip than across chips.”
This novel solution has won the company several high-profile customers. In early 2026, Cerebras signed a $20 billion, 750-megawatt deal with OpenAI. The start-up also has a multi-year deal with Amazon Web Services (AWS) to deploy the WSE in its data centers to accelerate AI inference. Last year, Cerebras struck a deal with Meta Platforms to run inference on its Llama 4 model. Taken together, these high-profile customers represent validation for Cerebras’ unique approach.
Growing like a weed, but not yet profitable
In 2025, Cerebras generated revenue of $510 million, which rose 76% year over year, but posted an operating loss of $146 million. At the same time, its remaining performance obligation (RPO) — or contractually obligate revenue that hasn’t yet been recognized — was $25 billion. The company will recognize 15% of its RPO in 2026 and 2027, 43% in 2028 and 2029, and the rest thereafter.
Investors should also be aware of Cerebras’s complex multiclass share structure. The company has three classes of common stock:
- Class A shares carry one vote per share and will be issued to the public.
- Class B shares are entitled to 20 votes per share and will be held by early investors and insiders, who will retain majority voting control.
- Non-voting N Class shares (Warrants issued to OpenAI and Amazon would allow them to buy up to $1.27 billion in these shares).
Lots of potential, plenty of risk
IPOs are inherently risky, particularly for a start-up that isn’t yet profitable. Furthermore, Cerebras faces significant customer concentration risk, with two customers accounting for 86% of its revenue last year.
If Cerebras’s WSE stands the test of time, it could represent a viable competitor to Nvidia, but only time will tell. Personally, I’m content to wait until Cerebras spends a little time under the public spotlight before considering buying shares.

