According to the report, OpenAI exclusively uses Nvidia’s GPUs to fire up its AI models. However, Sam Altman is looking for a different way. Negotiations are underway with Broadcom and other AI chip designers. Altman wants OpenAI to be capable of making its own proprietary AI chip design. Doing so will weaken the close ties between OpenAI and Nvidia.
OpenAI hires many former Google employees with relevant experience. Those hires were part of the team that created Google’s tensor processing unit. Earlier reports indicated Altman’s goal to raise billions for semiconductor factories. Thus, OpenAI is making huge bets on infrastructure.
A partnership with Broadcom makes sense for OpenAI. Broadcom has expertise in designing custom accelerators for AI models. They had the guts to collaborate with Google on TPUs, now in their sixth generation. Hence, Broadcom could deliver high-performance AI accelerators for OpenAI.
Broadcom also provides customized ASIC solutions, which in this direction becomes very useful as OpenAI has different requirements for its AI accelerator. Broadcom supplies main silicon solutions for data centers as a fabless chip designer. OpenAI can use the full product stack that Broadcom has at its disposal.
Innovation of this kind will not bring OpenAI up to par with Nvidia right away. Optimistically, a new chip could come out as early as 2026. That said, OpenAI continues to find ways to be more independent in AI.
OpenAI opened its first office in Japan earlier this year. This was done to be able to open other new streams of revenue. It also sought collaborations with local businesses, governments, and research institutions. OpenAI’s partners include Khan Academy and Carnegie Mellon.
