Cerebras' AI speed game

For a company obsessed with speed, Cerebras’ stock sure knows how to move fast.

Shares of the AI chipmaker jumped 9.08% today after OpenAI CEO Sam Altman called Cerebras a “close partner,” helping reverse some of last week’s 15% slide. Investors had gotten spooked by reports that Nvidia, rather than Cerebras, is powering the Ultrafast mode for OpenAI’s new GPT-6.1 Sol model.

The whiplash points to the bigger trend Cerebras is tapping into: As the AI trade shifts from building models to putting them to work, Cerebras is betting speed can give it an edge.

Faster AI needs faster chips

Cerebras’ answer to faster AI was to build the biggest AI chip ever made. Its wafer-sized processor packs 4 trillion transistors onto one chip, reducing the time data spends moving between separate processors. Its big bet is on inference, the computing that happens when you actually ask an AI model to do something. As AI gets put to work writing code, running agents and tackling other tasks, Cerebras is betting users will pay for answers that arrive really, really fast.

And investors are starting to put a big number on that opportunity. Mizuho estimates that the fast-inference market could reach $550 billion by 2030 and account for roughly 20% of AI workloads.

Since Brew Markets last checked in before Cerebras’ May IPO, that bet has gotten some real-world validation: Its cloud business revenue nearly quadrupled to $126 million last quarter, while core revenue more than doubled to $210 million. Cerebras now has $25.4 billion of contracted business left to deliver, and expects revenue to more than triple next year.

OpenAI is a big reason why: The AI startup has committed to buying 750 megawatts of Cerebras inference capacity from the beginning of this year through 2028, and Cerebras already powers GPT-5.6 Sol at up to 750 tokens per second, as much as 14 times faster than OpenAI’s standard processing.

Room for two?

That’s the good news for Cerebras, but last week’s sudden selloff showed investors the downside.

OpenAI can use Cerebras for workloads where low latency matters and use Nvidia for other tasks—meaning booming demand for fast inference doesn’t necessarily mean Cerebras gets all the business. Nvidia has the scale and a much broader ecosystem spanning training and inference, while Cerebras is making a more specialized bet on workloads where milliseconds matter.

That said, Cerebras is trying to broaden its customer base to companies like Cognition, Lovable, Block, and Figma, while partnering with AMD and Amazon Web Services (AWS). It’s also racing to increase manufacturing capacity more than tenfold this year to meet all those commitments.

Today’s announcement is great for Cerebras, but what matters more is whether the company can diversify beyond OpenAI and turn its speed advantage into durable growth. Cerebras has made speed its selling point—now it has to prove it can scale just as fast.—HC