OpenAI, the creator of ChatGPT, is reportedly in talks with chip giant Broadcom to develop a new AI chip, according to The Information.
CEO Sam Altman is leading this initiative as part of a larger plan to increase OpenAI's computing power for AI development. The company has been hiring chip design experts, many from Google's tensor processing unit team, to explore creating an AI server chip.
This comes as tech companies continue to follow scaling laws in order to build more capable models using even larger computing clusters. Altman has also spoken with Taiwan Semiconductor Manufacturing Co. about possibly funding new chip factories to boost production capacity.
OpenAI spokesperson Liz Bourgeois confirmed to The Information that the company is "having ongoing conversations with industry and government stakeholders about increasing access to the infrastructure needed to ensure AI's benefits are widely accessible." She did not comment specifically on the Broadcom discussions or chip development plans.
The collaboration with Broadcom is notable given the company's experience with Google's TPUs. OpenAI's chip team, led by former Google employee Richard Ho, is likely to partner with an American company for this project.
Altman's plan extends beyond chip design. He has discussed creating new companies with outside investors to finance real estate, power infrastructure, data centers, and specialized AI chip servers. OpenAI would then rent these servers.
This strategy could give OpenAI more leverage in future negotiations with NVIDIA, which has been profiting significantly from its AI-focused GPUs. However, it also risks complicating OpenAI's relationship with this key supplier.
The proposed AI chip development and associated infrastructure projects are still in the early stages. The chip design process has not yet begun, and production would likely not start until at least 2026. However, the discussions with Broadcom and other industry players indicate a clear commitment from OpenAI to pursue this ambitious vision.