Cerebras Reports $193 Million in Quarterly Revenue and a $20 Billion OpenAI Deal in Its First Earnings as a Public Company

What Cerebras reported

Cerebras Systems told investors on June 23 that revenue for the quarter ended March 31 reached $193.4 million, up 94% from a year earlier and 13% from the prior three months. Hardware sales, the computers customers install in their own data centers, came to $110.6 million. The faster-growing piece was cloud and other services, where Cerebras rents access to its machines over the internet, and that line jumped 178% to $82.8 million. Gross margin, the share of revenue left after the direct cost of delivering a product, was 45%. The company still lost money, with a $14.0 million net loss, though that was small against its sales. It closed the quarter with $3.3 billion in cash and short-term investments, a cushion built largely from its May initial public offering.

A $20 billion deal with OpenAI

The bigger news sat next to the numbers. Cerebras said it had signed a multi-year agreement with OpenAI valued at more than $20 billion, under which OpenAI will deploy 750 megawatts of Cerebras inference compute over the next several years. Inference is the work a finished AI model does when it answers a question, as opposed to the training that builds the model in the first place. The two also launched Codex-Spark, a coding model the company says generates more than 1,000 tokens a second, a token being a chunk of text close to a word. The relationship already ran deep, because OpenAI had extended a $1 billion working-capital loan in January. Cerebras separately began a partnership with Amazon Web Services, splitting each request so Amazon’s Trainium chips handle one stage and Cerebras hardware races through the rest.

Speed now, profits later

Cerebras builds what it calls wafer-scale chips, processors etched onto a single slice of silicon about the size of a dinner plate rather than cut into many small pieces. The design moves data faster than racks of standard chips wired together, which is why the company sells speed as its main advantage. That approach is costly to scale. For the current quarter Cerebras guided to roughly $194 million in core revenue and a core operating margin of negative 30% to negative 32%, meaning it still expects to spend more than it earns. For the full year it sees revenue of $855 million to $865 million, up about 69% at the midpoint. The May IPO that raised about $6.4 billion, a $1 billion pre-IPO round in February, and an $850 million credit line in April give it room to fund the data centers the OpenAI contract will need. Shares trade on Nasdaq as CBRS.

What to watch

1. The OpenAI build-out. The agreement covers 750 megawatts of compute delivered over several years, so the test is how quickly Cerebras can stand up that capacity and how much cash it must spend to do it.

2. The path to profit. Core operating margin is still deeply negative, and full-year guidance implies the losses continue. Watch whether rising revenue and the Amazon partnership begin to close the gap.

3. Customer concentration. OpenAI now anchors a large share of future demand. Watch whether Cerebras can add enough enterprise and cloud customers to avoid leaning on a single buyer.