Cerebras Is Going Public at Exactly the Moment the AI Chip Market Starts Looking Less Like Nvidia’s Private Kingdom
Cerebras Is Going Public at Exactly the Moment the AI Chip Market Starts Looking Less Like Nvidia’s Private Kingdom
Cerebras Systems filed publicly for a U.S. IPO on April 17, and the timing says more about the AI market than the filing itself. For years, Nvidia looked untouchable in AI hardware. Then demand for inference exploded, cloud providers started shopping more aggressively, and OpenAI reportedly agreed to spend more than $20 billion on Cerebras-powered servers over the next three years. That is not a minor supplier contract. That is a shot across the bow.
The Sunnyvale company has been pitching the same big idea for years: bigger chips, fewer bottlenecks, faster AI training and inference. This time the pitch landed in a market that suddenly wants alternatives.
According to Reuters, Cerebras disclosed its IPO filing on Friday, April 17. The company did not say how much it plans to raise. A spokesperson told TechCrunch that the offering is planned for mid-May. That alone would make the story notable. The bigger detail is what sits behind the filing: a sudden burst of revenue, an OpenAI deal measured in gigawatts and billions, and a real effort to position Cerebras as something more dangerous to Nvidia than a niche chip startup.
The numbers got big in a hurry
Cerebras reported $510 million in revenue for 2025, according to its filing, with net income of $237.8 million. Strip out one-time items and the picture changes. TechCrunch said the company posted a non-GAAP net loss of $75.7 million. That split matters. It suggests a business that has found real commercial demand, but not yet a clean, boring, public-company earnings profile.
Investors may not care much about the messiness if the backlog keeps growing. Reuters reported that much of the recent growth is tied to OpenAI, including a multi-year arrangement under which the ChatGPT maker will deploy 750 megawatts of Cerebras chips. CNBC added a useful detail: the contract calls for 250 megawatts a year from 2026 through 2028, and the whole deal is valued at more than $20 billion. In AI infrastructure terms, that is enormous. A lot of chip startups spend years begging for pilot projects. Cerebras is talking about utility-scale compute.
There is another clue in the financing trail. TechCrunch reported that Cerebras raised a $1.1 billion Series G in 2025, then a $1 billion Series H in February 2026 at a $23 billion valuation, citing the Wall Street Journal. CNBC separately described that February financing as a $1 billion round at a $23 billion valuation. The market does not hand out that number on a whim, especially after the mood swings of the past two years.
This filing is really about inference
The obvious headline is that Cerebras wants public-market money. The real story is where it thinks AI demand is moving next.
Training still grabs attention, but inference is where the revenue pressure lives. Every chatbot query, every image generation request, every enterprise agent call turns into inference load. Reuters noted that Cerebras is focused on inference, which helps explain why Andrew Feldman has been talking less like a chip designer and more like someone trying to steal a cash machine. In a recent Wall Street Journal interview quoted by TechCrunch, Feldman said, “Obviously, Nvidia didn’t want to lose the fast inference business at OpenAI, and we took that from them.” It is a brash quote. It also sounds like a chief executive who knows this IPO depends on a simple story investors can repeat in one sentence.
That story goes like this: Nvidia won the first AI boom, but the second boom may belong to whoever can serve inference faster and at lower cost. Cerebras wants to be that company.
The timing is not subtle. Reuters described the filing as part of broader optimism around new listings tied to the AI boom. A few years ago, an IPO filing from an expensive semiconductor startup would have raised questions about burn rate, customer concentration, and whether incumbents could crush it on pricing. Those questions still exist. But April 2026 is a different market. Compute is scarce, demand is stubborn, and customers with giant AI bills are now motivated to avoid depending on one vendor forever.
AWS makes this look less like a one-customer bet
The OpenAI contract is the attention magnet, but Amazon may be just as important. TechCrunch reported that Cerebras recently struck an agreement with Amazon Web Services to use Cerebras chips in Amazon data centers. That matters because public investors will worry, rightly, about concentration risk. A chip company that relies too heavily on a single customer can look amazing right up until purchasing plans change.
The AWS relationship gives Cerebras a cleaner argument: this is not just an OpenAI sidecar. It is an infrastructure supplier trying to wedge its hardware into mainstream cloud distribution. SiliconANGLE reported that AWS agreed to deploy the company’s WSE-3 in its data centers as part of a new disaggregated architecture. Even if that rollout starts modestly, the signal is plain enough. Amazon does not make room in its data centers for charity.
There is also a strategic irony here. Hyperscalers have spent years building custom silicon to reduce dependence on outsiders. Yet the AI surge has been so violent that even the largest cloud companies are willing to pull in more external hardware if it helps them meet demand. That is good news for Cerebras. It is also a quiet admission that the market still cannot get enough useful compute.
The old IPO problem has not disappeared
Cerebras did try this before. TechCrunch noted that the company filed for an IPO in 2024, but the plan was delayed by a federal review tied to an investment from Abu Dhabi-based G42 and was eventually withdrawn. That history should keep the excitement in check. This is not a startup floating effortlessly into public markets. It is a company returning after a failed attempt, now armed with bigger customers and better timing.
There is risk everywhere in this setup. The filing lands while investors are still debating how durable today’s AI spending really is. If enterprise buyers slow down, if OpenAI reshuffles suppliers, or if Nvidia cuts hard on price, the story can change fast. Cerebras also has to prove that headline deals turn into stable, repeatable economics. Public investors are far less forgiving than private ones.
Still, one opinion feels hard to avoid: this may be the first AI hardware IPO in a while that actually changes the conversation. Not because Cerebras is certain to beat Nvidia. That claim would be absurd. But because the filing turns a quiet industry suspicion into something public and testable, namely that the AI chip market is finally large enough for a real second power center to form.
That is why this filing matters. Cerebras is not asking investors to fund an idea. It is asking them to back a market shift already underway, one measured in 750 megawatts, $510 million in annual revenue, a planned mid-May offering, and a chief executive bold enough to say Nvidia already lost one important piece of the fight. Public markets will now decide whether that sounds like vision or overreach.
For tool-by-tool comparisons, see our AI coding listings and the comparisons section.