OpenAI and Google are battling it out in creating the best possible AI models, but they also appear to be collaborating behind the scenes.
OpenAI has begun using Google’s TPUs (Tensor Processing Units) in its operations, a report from The Information says. This is the first time that OpenAI has tried out an alternative to NVIDIA’s GPUs, which previously powered all of its operations. OpenAI is accessing Google’s TPUs through Google Cloud.
Thus far, OpenAI had been using NVIDIA’s GPUs. These GPUs were invented by NVIDIA as a way to generate better graphics for video games, but the AI community then realized that these devices were also great to train and run AI models. GPUs had then become the standard for AI processing. In parallel, Google had developed its own hardware device named TPU, which was optimized for AI and machine learning operations. Google had been using these TPUs internally to power AI and ML efforts at the company, and began relying on them even more once the AI revolution kicked off. Today, Google mainly uses its own TPUs to power its AI efforts like Gemini and Veo3, while OpenAI uses NVIDIA’s GPUs, largely through Microsoft’s Azure cloud.
It now appears that OpenAI has begun using Google’s TPU for parts of its own operations. Google doesn’t make it top-of-the-line TPUs available for use for anyone outside the company, but lets users and companies use its TPUs through Google Cloud. This presents and interesting dynamic between Google and OpenAI. The two companies are competing fiercely in the AI model race, and are usually on the first or second place in most AI benchmarks. But OpenAI is now using Google’s hardware to power a part of its operations. Using Google hardware could be a clever strategic move for OpenAI. Not only does it give it more bargaining power with NVIDIA — it now has an alternative that can power a part of its operations — but it also signals less reliance on its largest investor in Microsoft, with which it has had a relationship that is increasingly seeming more fraught. It remains to be seen how this relationship plays out, but there are many interesting dynamics — with their own plays and counterplays — developing between the top players in the AI race.