Cerebras prepares IPO of up to $4 billion, targeting a valuation of about $40 billion

Cerebras is heading to the stock market again, and it is going after one of the most visible tech deals of the year. The AI chipmaker wants to raise up to $4 billion at a valuation of about $40 billion, after a failed IPO attempt in 2024 and with a major OpenAI contract behind it.

Back to the market

In autumn 2024, Cerebras filed for an IPO, but a review by US regulators stalled the listing. The problem was not the technology but the structure of its shareholders and customers: G42 of Abu Dhabi played an important role in the business, and that link raised questions at the Committee on Foreign Investment in the United States. In October 2025 the registration was withdrawn, and the episode looked like an example of how geopolitics can stop even a very prominent AI startup.

Now the company is back with a more aggressive plan. On April 17, 2026, Cerebras refiled its S-1 with the SEC, and on May 2 it emerged that the company expects to raise up to $4 billion at a valuation of roughly $40 billion. That is noticeably above the range analysts had previously expected, and almost five times the private valuation of $8.1 billion at which Cerebras raised money in September 2025. If the offering prices at the top of the range, it could become the first major AI-hardware IPO of 2026.

The company plans to list on Nasdaq under the ticker CBRS. The underwriters are Morgan Stanley, Citigroup, Barclays and UBS. The lineup of banks alone shows that Cerebras no longer looks like a niche experiment: this is a bid for a large public story, in which the company will be compared not with other startups but with the leaders of the AI infrastructure market and with compute suppliers carrying very high expectations.

The bet on inference

Cerebras' main argument to investors is not just an unusual chip architecture but an anchor customer. In its updated prospectus, the company disclosed a multi-year agreement with OpenAI for inference capacity through 2028. For the market this is an important signal: the startup is not selling a promise to find demand someday, but has already plugged into the real compute supply chain of one of the biggest players in AI, and that changes the tone of the whole deal.

Against that backdrop, even the current financials look different. Cerebras' revenue for 2025 was $510 million, up 76% from a year earlier. For a hardware maker this is a notable growth rate, but more important is that the OpenAI contract gives visibility into future utilization and helps explain the jump in valuation. Infrastructure investors currently place a particular premium on companies with a large, clear and long-term buyer of capacity.

• up to $4 billion is what the company wants to raise in the IPO
• about $40 billion is the target valuation
• $510 million is the revenue Cerebras earned in 2025
• up to 750 MW of inference capacity is planned for deployment for OpenAI by 2028
• the company's valuation rose from $8.1 billion in September 2025 to $23 billion in February 2026

It also matters that Cerebras is not launching a direct attack on Nvidia in model training, but is targeting a narrower and fast-growing segment: inference. That is where finished models run, user requests are served, and cost pressure is felt most acutely. Cerebras' logic is simple: if its wafer-scale processors really deliver advantages in response speed and compute density, it can claim its own niche alongside Nvidia rather than in place of it.

Where the risks remain

Despite the new momentum, the deal has several weak points. First, the G42 issue has not fully gone away. In March 2025 the company received regulatory approval after its ownership structure was changed and G42's stake was converted into non-voting shares. Formally the obstacle has been removed, but for some public investors the very fact of past dependence on a single foreign customer and shareholder will still remain a risk factor.

Second, Cerebras' technological difference is also its vulnerability. The company is known for its wafer-scale chips, processors so large that they are usually compared not with ordinary GPUs but with an entire silicon wafer. This sets Cerebras apart from Nvidia, but the approach is harder to scale and to manufacture in volume. The public market usually pays readily for uniqueness only when it sees clear production economics and margins over the long run.

Finally, there is a broader market question. In late 2025, several large AI offerings debuted weaker than expected, and investors grew more cautious toward loss-making infrastructure companies with very high private multiples. So the Cerebras IPO will be a test not only for the company itself but for the whole thesis that the market is ready to automatically translate AI hype into the value of hardware.

What it means

If Cerebras can really price on terms close to those announced, it will be an important signal for the entire AI infrastructure segment. The market will show it is ready to value generously not only model builders but also compute providers with a strong contract and a clear specialization. If demand turns out weaker, it will be a reminder that even in AI, public money now demands not just a loud technology but hard proof of the economics.