OpenAI has signed a multiyear agreement worth about $10 billion with chip start-up Cerebras Systems to secure vast amounts of computing infrastructure, deepening its push to reduce reliance on Nvidia and lock in the resources needed to power its fast-growing artificial intelligence models. The deal, announced on Wednesday, will run through 2028 and provide OpenAI with 750 megawatts of computing capacity, an amount of energy comparable to that used by a major U.S. city.
The agreement marks OpenAI’s latest effort to guarantee access to scarce and expensive computing power at a time when demand for advanced AI chips is surging globally. As the developer of ChatGPT and other widely used AI systems, OpenAI has been racing to secure long-term infrastructure commitments while broadening its roster of suppliers beyond the dominant graphics processing units produced by Nvidia.
Cerebras, which was valued at $8.1 billion late last year, is positioning itself as a challenger to established chipmakers by offering specialised processors roughly the size of a dinner plate. The company says its chips are designed to excel at AI inference, the stage at which trained models generate responses to user queries, delivering dramatically faster performance than conventional GPUs from Nvidia and AMD. France’s AI company Mistral already uses Cerebras chips to power its Le Chat chatbot.
Inference is widely expected to become one of the biggest drivers of AI-related computing demand in the coming years, as consumer and enterprise use of AI tools accelerates. Nvidia has signalled how strategic this area has become, recently hiring talent from rival Groq, another company focused on ultra-fast inference chips.
For OpenAI, the Cerebras deal fits into a broader strategy of diversification and scale. Over the past year, the company has struck a major partnership with AMD and is preparing to launch its own AI chips, designed in collaboration with Broadcom. In total, OpenAI has made infrastructure commitments estimated at around $1.5 trillion over the next decade to support the training and operation of its models, according to people familiar with the matter.
Those commitments far exceed OpenAI’s current financial position. While the company’s annualised revenue is estimated at about $20 billion, it remains loss-making, and the scale of its spending highlights the enormous capital requirements of leading-edge AI development. Chief executive Sam Altman has repeatedly argued that privileged access to computing power, advanced chips and related infrastructure will give OpenAI a decisive edge over rivals such as Google and Meta.
OpenAI’s head of infrastructure, Sachin Katti, said the company’s approach is to build a resilient portfolio of computing systems tailored to different workloads. He said Cerebras’ technology would enable faster responses and more natural interactions from AI models. Katti joined OpenAI in November after leaving Intel, underscoring the company’s effort to recruit senior talent with deep semiconductor expertise.
Cerebras, meanwhile, is preparing for an initial public offering after raising $1.1 billion in September from investors including Fidelity and 1789 Capital, a firm backed by Donald Trump Jr. The start-up’s growing list of high-profile customers and partners is bolstering its challenge to incumbent chipmakers at a moment when alternatives to Nvidia are in high demand.
OpenAI itself has raised about $60 billion to finance its expansion and has entered into complex arrangements with chip partners, including AMD and Nvidia, as it seeks to secure supply while shaping the future of AI hardware. It is also in talks with investors about a new funding round that could raise as much as $80 billion and value the company at more than $800 billion, according to multiple people with knowledge of the discussions.
