OpenAI partners with Broadcom on $10 billion AI chip deal to secure training capacity

OpenAI is reportedly working with Broadcom on custom AI chips to ease the GPU bottlenecks the company is facing. On a recent earnings call, Broadcom disclosed a new "fourth major AI developer" placing a one-time $10 billion order for AI server racks. People familiar with the matter say it's OpenAI.

Broadcom expects the order to begin contributing in the summer quarter of 2025. The move is about securing training capacity, not replacing Nvidia.

Hock Tan, Broadcom's chief executive, says the new customer meaningfully changes the company's 2026 outlook. Broadcom calls the custom parts "XPUs," chips designed for specific applications like AI training. In August, the company launched its Jericho chip, which links data centers up to 60 miles apart to speed up AI workloads.

OpenAI CEO Sam Altman said GPU shortages slowed the release of their ChatGPT-4.5 model and that he has already outlined plans to add tens to hundreds of thousands of GPUs. Large GPU allocations require months of lead time, according to industry sources, a timeline that challenges AI companies.

For OpenAI, the deal promises more predictable training power and less exposure to GPU shortages, though scaling still depends on Nvidia. For Broadcom, it provides a deeper push into AI infrastructure.