OpenAI is reportedly planning to purchase AI chips from Google, specifically its Tensor Processing Units (TPU). While the final terms of any potential agreement remain undisclosed, both companies have declined to comment on the matter.
This move aligns with OpenAI's strategy to diversify its computing infrastructure and reduce reliance on a single supplier. Currently, the company relies heavily on Nvidia GPU clusters for training and running its core models. Nvidia CEO Jensen Huang previously stated that OpenAI committed to deploying Nvidia infrastructure through 2030, with plans reaching up to 16 gigawatts.
OpenAI is also developing its own internal chip, codenamed Jalapeno, in collaboration with Broadcom. The team behind this project is led by Richard Ho, a former Google employee who previously worked on the development of Google's TPU technology.
Google began developing its TPU over a decade ago, initially to optimize its search engine before expanding its use to AI workloads. The company recently started selling these chips directly to select customers. Oracle also announced it will offer Google TPU-based computing power on its cloud platform.
