Cerebras Falls 10% After Admitting Profitability Is Shrinking Fast

Cerebras posted its first earnings report as a public company on Tuesday and immediately reminded investors why revenue growth means nothing without margins. The stock fell roughly 10% despite first-quarter revenue hitting $193.4 million, up 92% year-over-year, and a full-year forecast that beat consensus expectations. This is what happens when you promise Wall Street growth and deliver instead a cautionary tale about the cost of scale.

The numbers looked plausible on the surface. Net losses narrowed to $14 million from $23.9 million in the prior year. Full-year revenue guidance came in at $855 million to $865 million, comfortably ahead of the $824.8 million analysts were modeling. For a company that just raised $5.55 billion in one of the largest U.S. tech IPOs in years—shares opened at $350, up from the $185 IPO price—this should have been a victory lap. Instead, the market did what it does best: it priced in reality.

That reality is written in margins, and the numbers are brutal. Cerebras guided gross margin to collapse from 47% in Q1 to 36-38% in Q2. That is roughly 1,000 basis points of compression in a single quarter. For context, Nvidia's gross margins sit at 75%. The gap is not a rounding error; it is the difference between a moat and a commodity business.

The excuse is almost quaint in its honesty. CEO Andrew Feldman confirmed on the earnings call that demand is not the constraint. Cerebras is not struggling to sell chips to customers desperate for AI compute. The problem, he explained, is buildings. Specifically, the lack of them. CFO Bob Komin elaborated on the call that renting outside capacity will "depress core cloud and other services margin temporarily." This is corporate speak for: we are so supply-constrained on our own infrastructure that we are outsourcing compute at lower margins just to fulfill contracts.

The crown jewel is a multi-year deal with OpenAI valued at over $20 billion for 750 megawatts of compute. It is the kind of customer that should silence every skeptic in the room. Instead, it underscores the trap: landing massive contracts is worthless if you cannot fulfill them profitably. Cerebras will operate at negative 28% to 32% operating margin through 2026, according to guidance. That means despite $20 billion in OpenAI revenue on the books, the company will lose money on a full-year basis for at least the next two years.

This is where the broader sector narrative breaks down. The IPO market has spent the better part of eighteen months treating AI infrastructure as a license to print cash. Cerebras was supposed to be the proof point: a company with genuine proprietary technology, a real revenue base, and absolute demand tailwinds. And it still cannot pencil profitability because the capital intensity of the business has no off switch.

What makes Tuesday's selloff genuinely important is not Cerebras itself—it is what it tells us about every other AI chipmaker in pre-IPO or newly public territory. If Cerebras, with a $20 billion OpenAI contract locked in, cannot map a path to acceptable margins, what does that mean for companies with smaller customer bases and less differentiated technology? The market is clearly asking that question, and the answer it is pricing in is not optimistic.

The irony is that Cerebras is executing the strategy it promised: shipping more chips, landing bigger customers, growing revenue at a pace that would make most semiconductor companies blush. What it did not promise, because no one wants to hear it, is that growth at this scale requires infrastructure investment that swallows profitability whole. The margin compression forecast is not a surprise to anyone who understands data center economics. It is a surprise only to investors who bought the stock believing that AI demand was sufficient to overcome everything else.