Cerebras's 'Miss' Was Actually A Beat — The Market Sold The Wrong Number

Cerebras just handed the market a beat and got punished for a miss. The wafer-scale AI chipmaker and chief Nvidia challenger reported core revenue of $209.9 million for the quarter ended June 30, up 103 percent year over year and comfortably above the roughly $194 million the Street expected — yet the headline GAAP number of $180.1 million missed that same consensus, and shares slid as much as 16 percent after hours before paring losses. The roughly $30 million gap between the two revenue lines is the technicality — data-center pass-through revenue that GAAP strips out. The market sold the wrong number.

Here's the technicality. The difference between core and GAAP revenue is data-center pass-through revenue — money that flows through Cerebras's books for infrastructure it doesn't ultimately own. GAAP excludes it; the “core” metric management reports and analysts model includes it. Strip that out and the quarter was a clean beat on the number that actually matters. Adjusted EPS of roughly negative $0.04 to $0.05 also crushed the roughly negative $0.17 consensus — still a loss, but a far smaller burn than anyone modeled.

The durable part of this story isn't the headline at all — it's the mix. Core cloud and other services revenue hit $127.7 million, up 287 percent year over year — nearly quadrupled — and now accounts for about 61 percent of core revenue, versus $82.1 million of core hardware. That's the whole thesis in one line: Cerebras is increasingly an inference cloud company wearing a chip company's clothes, and cloud revenue is where the compounding happens.

Then there's the anchor customer. The OpenAI partnership announced in January for 750 megawatts of wafer-scale inference systems — the largest high-speed AI inference deployment ever built — was later upgraded to a definitive multi-year agreement worth more than $20 billion, with an option to scale toward $30 billion. OpenAI put up a $1 billion upfront deposit, and Cerebras granted warrants for up to a 10 percent equity stake. GPT-5.6 Sol already runs on Cerebras hardware at roughly 750 tokens per second. The proof points stack up: remaining performance obligations of $25.4 billion, 600 megawatts of data-center capacity under contract, and a planned 10x expansion in manufacturing capacity this year.

Management used the quarter to raise the bar. Full-year core revenue guidance moves up to $880–890 million from $855–865 million, with core gross margin of 41–43 percent and core operating margin of negative 19 to negative 17 percent. Q3 core revenue is guided to $214–216 million at a 38–40 percent gross margin. And CFO Bob Komin put a marker down for next year, saying Cerebras plans to “more than triple revenue in 2027.” For a company guiding to roughly $885 million this year, that implies a 2027 run rate measured in the billions.

So why did the stock tumble anyway? Start with the GAAP income statement, which is genuinely ugly. Net loss of $450.5 million, of which roughly $386.6 million was stock-based compensation, against $320.2 million of R&D spend. Core gross margin came in at 41 percent, up 940 basis points year over year — but GAAP gross margin sits at 14 percent and GAAP operating margin at negative 265 percent versus core's negative 16 percent. In a tape where every AI name trades on optics, a headline loss that big reads worse than it is.

There's also a genuine yellow flag in the guide: Q3 gross margin of 38–40 percent is a step down from Q2's 41 percent as Cerebras absorbs the cost of its manufacturing expansion. Layer that on a valuation with no room for error — a market cap in the $50 billion range against roughly $880 million of guided core revenue, a forward multiple north of 200 times — and you get a stock that punishes small stumbles. Extreme multiples demand perfect prints, and this wasn't one on the surface.

The bottom line: the selloff is the market pricing a beat as a miss and ignoring what actually improved — triple-digit revenue growth, a cloud business that nearly quadrupled, a more-than-$20-billion OpenAI anchor contract, and raised guidance. The Street still rates CBRS a Strong Buy with a mean target around $292, implying meaningful upside if the cloud-and-inference flywheel keeps compounding. But make no mistake: with margins guided down a notch and cash burn still real, this is a show-me story. The numbers are moving in the right direction — the market just needs to start believing the right ones.

Disclosure: The Signal holds no position in CBRS. Positions may change. This is not financial advice.