OpenAI CEO Sam Altman clarifies that Cerebras remains a close partner with deep cooperation on frontier AI speed challenges.

OpenAI CEO Sam Altman has clarified that despite recent speculation, Cerebras remains a close partner of OpenAI, with both companies maintaining deep cooperation at the speed frontier.

The speculation emerged after Cerebras' stock price fell approximately 17%, following reports that OpenAI had moved part of the ultra-fast version of its new model to NVIDIA graphics processors. This raised concerns among investors that the wafer-scale chip company would lose a significant portion of its most important client's inference workload. Analyst Paul Meeks of Freedom Capital subsequently upgraded his rating from hold to buy, with a target price of $209, characterizing the sell-off as an overreaction.

The partnership, however, is grounded in concrete commitments. In January, OpenAI agreed to access 750 megawatts of Cerebras computing power, to be delivered in phases through 2028, specifically for low-latency inference. According to Reuters, the contract is valued at over $10 billion over its term. Cerebras operates as a cloud services provider, selling access to its own chips through company-built and leased data centers, with OpenAI paying based on usage.

In August, OpenAI connected the Ultrafast tier of GPT-5.6 Sol to Cerebras' infrastructure, achieving approximately 750 tokens per second—up to 14 times faster than the standard tier. OpenAI's head of computing, Sachin Katti, stated that combining computing power to match the workload was essential, with Cerebras providing dedicated low-latency inference rather than replacing graphics processors across the board.

Cerebras CEO Andrew Feldman noted that negotiations began following a demonstration last August, when OpenAI's open-source model proved more efficient on Cerebras chips than on traditional graphics processors, particularly for inference and models requiring extended reasoning. The wafer-scale engine integrates computing, storage, and bandwidth on a single large chip, eliminating memory transfer inefficiencies inherent in conventional hardware architectures.

Market dynamics reflect differing interpretations of the shift. Shareholders worried about customer concentration see a threat; other buyers view the tight inference supply market as evidence that one client diversifying to NVIDIA does not undermine the partnership. The 750-megawatt commitment and the already-launched Ultrafast tier sustain Cerebras' revenue expectations, while NVIDIA captures an additional workload tier.

After the rating upgrade, Cerebras' stock price rebounded roughly 2% on Friday. The company also announced a deployment of approximately 100 megawatts of inference hardware with cloud startup Gimlet Labs, signaling that low-latency inference demand extends beyond OpenAI.