Cerebras Falls 10% as AI Chip Reality Meets Margin Fantasy

Cerebras Holdings fell more than 10% in premarket trading Wednesday following its first earnings report as a public company, a neat reminder that the market's appetite for artificial intelligence semiconductor stories has a hard floor: the moment guidance suggests profitability might actually take a step backward.

The chipmaker reported first-quarter revenue of $193.4 million, nearly double the $99.5 million from the same quarter last year. On any other day, that would be presented as evidence of something working. But Cerebras, which raised $5.55 billion in its IPO last month on the back of a $20 billion multi-year deal to supply OpenAI with inference chips—750 megawatts of them, the kind of number that makes investors' pupils dilate—made the critical error of providing margin guidance.

The company forecasted adjusted gross margins of 38% to 41% for full-year 2026. That's down from the 47% it posted in Q1. For context, Nvidia's gross margins sit in the mid-70s. Advanced Micro Devices manages the mid-50s. Cerebras just told the market it expects to operate like a contract manufacturer with premium complexity and zero premium pricing.

The explanation, at least, has the virtue of honesty. CEO Ben Bajarin noted that Cerebras' manufacturing approach—making some of the world's largest chips—is difficult. Large dies have lower yields. Yields destroy margins. This is not secret knowledge. It has been true since the Pentium 4. CFO Bob Komin added that the company is temporarily renting back its own systems from an existing client to meet demand while building data center capacity. That rental expense, he said, will depress margins temporarily. The word "temporarily" doing a tremendous amount of lifting in that sentence.

Cerebrash stock had reached an all-time high of $386.34 on May 14, 2026. It hit an all-time low of $161.26 on June 25, 2026. Wednesday's premarket drop of 14% fits neatly into that volatility band—the price of a story that came in too hot and a set of numbers that couldn't support the temperature.

Here is what happened: Cerebras told the market it had secured extraordinary demand from the most important customer in tech. Investors extrapolated that into extraordinary profitability. Cerebras then disclosed that serving that customer at scale requires renting server capacity at rates that compress margins below industry peers. This is what's known in trading floors as "the gotcha."

The company's long-term target is 60% gross margins. That's a reasonable ambition. It requires that manufacturing yields improve, that scale drives down per-unit costs, that rented capacity becomes in-house capacity, and that none of the variables move the wrong direction in the meantime. None of those things are guaranteed. Some of them are impossible until something changes fundamentally about how large chips get made.

What's more predictable is the pattern: growth-stage semiconductor companies enter the public market on a wave of application-layer demand. Their first earnings beat expectations. Their second or third quarters reveal that the economics of scaling don't match the narrative of disruption. Margins compress. Stock falls. Investors who bought the story rather than the numbers take the loss. The company spends two years proving it can sustain 50% margins instead of 60%, the stock rebounds, and a new cohort of buyers learns the same lesson again.

Cerebrash has a real customer, real demand, and real revenue. It also has real manufacturing constraints and a margin profile that suggests those constraints are going to cost money for longer than the IPO pricing suggested. That's not a death sentence. It's just the moment when the conversation changes from "isn't this exciting" to "what will this actually earn." The market has spoken: it finds the second conversation considerably less compelling.