OpenAI started using Google TPUs, reducing dependence on Nvidia

OpenAI began using Google’s Tensor Processing Units TPU to support ChatGPT and related services. This is one serious step by the company toward reducing dependence on Microsoft’s infrastructure and Nvidia’s monopoly.

What’s behind this decision? OpenAI is renting Google Cloud capacity with an eye toward cutting inference costs. That very process when the model generates responses to our queries. Until this moment, the company relied exclusively on Nvidia graphics processors through partnerships with Microsoft and Oracle.

Interesting nuance. Google doesn’t offer OpenAI its most powerful TPU versions! The tech giant saved the top accelerators for internal use, including developing its own language models for the Gemini project. But even access to earlier TPU versions is a strategic move for OpenAI. Under conditions of growing shortage of AI chips.

The question remains open so far: will OpenAI use TPU only for inference? Or will it venture into training its models on this platform? In any case, hybrid computing infrastructure gives the company more maneuverability for scaling under conditions of fierce competition.

It seems that Nvidia’s monopoly on the artificial intelligence market is beginning to gradually erode!