Cerebras beat its first public earnings test on revenue, but investors sold the stock because the harder question is margin. Fast AI chips are valuable. Proving they can scale profitably is the next exam.
Cerebras Systems gave Wall Street the kind of first earnings report a newly public AI chipmaker is supposed to deliver: revenue up sharply, OpenAI on the customer list, AWS in the story, and guidance above analyst expectations. Then the stock fell anyway.
That is the useful part of the story. Not the selloff by itself, and not the headline growth number on its own. Cerebras reported $193.4 million in first-quarter revenue on June 23, its first quarterly update since going public in May. Investor's Business Daily reported that analysts polled by FactSet had expected $181.2 million. MarketWatch put core revenue at $191.3 million, up 92% from a year earlier, with a core net loss of just $2.5 million. For most companies, that would be a clean win.
Cerebras is not being valued like most companies.
The Sunnyvale company came public in May at $185 a share, raised billions in the IPO, and quickly became one of the market's cleanest public bets on AI inference outside Nvidia. Investopedia noted before the earnings report that the stock had already fallen more than a third from its first-day highs, while still trading above its IPO price. After the report, shares slipped again in extended trading, with MarketWatch showing a drop of more than 9% and The Wall Street Journal reporting an after-hours slide to $202.73.
The market was not rejecting the demand story. OpenAI agreed in January to deploy 750 megawatts of Cerebras inference capacity in a multiyear agreement that the company values at more than $20 billion, according to MarketWatch. AWS is also a named customer, and Cerebras chief executive Andrew Feldman told MarketWatch that the company is serving OpenAI's GPT-5.4, released in early March. If you are looking for proof that large buyers will at least test non-Nvidia AI compute, this is it.
The problem is the cost of proving it.
According to The Wall Street Journal, Cerebras projected full-year core operating margins of negative 28% to negative 32%, even as it guided full-year core revenue to $855 million to $865 million. That revenue outlook is above analyst expectations, but it also tells you the company is spending heavily to bring capacity online. Barron's reported that Cerebras is ramping its $20 billion service contract with OpenAI and expects reduced profitability over the rest of the year as it does so.
Frankly, that is the right thing for investors to focus on. Anyone can admire a fast chip. You still need to know whether the company can turn that speed into durable economics once the data center bills, customer warrants, capacity commitments and operating losses all show up in the same model.
Cerebras has a real technical argument. Its wafer-scale design uses a whole silicon wafer rather than cutting the wafer into many smaller chips, and MarketWatch noted that its engine is 58 times larger than Nvidia's B200 chip. The company also uses SRAM on chip, which Wedbush analyst Matt Bryson said makes memory constraints less of an issue than they are for chipmakers tied to high-bandwidth DRAM. That is not marketing fluff. It is a specific hardware bet, and it explains why customers chasing faster inference would pay attention.
But speed alone does not settle the business case. The company can win benchmark attention and still have to rent, build or reserve enough data center capacity to serve customers who want AI inference at serious scale. The Wall Street Journal said hardware revenue rose 59% to $110.6 million, while cloud and other services revenue nearly tripled to $82.8 million. That mix matters because cloud growth brings operational burden with it. You are no longer just selling a box. You are delivering compute as an ongoing service.
The real test is capacity, not hype
If you buy enterprise AI infrastructure, this is the part to watch. The case for Cerebras rests on performance, availability and a promise that the economics improve as more capacity comes online. Morgan Stanley analyst Joseph Moore, cited by MarketWatch, said the key driver of revenue and gross-margin upside would be how quickly Cerebras can deploy compute capacity, with expectations for 250 megawatts online by the middle of next year.
That is a very concrete target. It is also a hard one. If Cerebras can bring capacity online quickly, keep OpenAI and AWS engaged, and show that losses narrow as utilization rises, the first earnings selloff will look like the market doing what it often does after an IPO pop: forcing the story back down to numbers. If margins stay under pressure while revenue grows, the company will still be important, but the stock will be a much more difficult argument.
The OpenAI deal cuts both ways. It is the strongest customer endorsement Cerebras could ask for, and it anchors demand in a market where everyone wants proof that alternatives to Nvidia are real. It also concentrates attention on execution. A $20 billion agreement sounds enormous, but investors still need to see deployment schedules, revenue recognition and margin behavior over more than one quarter. You don't get credit forever for signing the right customer. You have to deliver the capacity.
The first earnings report confirmed Cerebras is not a paper story. It has revenue, customers and a technical design that serious buyers are testing. It also confirmed that Wall Street is no longer pricing the company on enthusiasm alone. That is healthier than the IPO pop, even if it is less fun for shareholders.
Also read: The EU has joined Pax Silica and the global AI chip map just changed permanently • Trump's quantum executive orders gave the sector a jolt but the hard question is whether Washington can sustain it • Meta ditches the Ray-Ban name and bets its own brand can own smart glasses
