OpenAI Rents Google TPUs as Cheaper Alternative to Nvidia GPUs

In a notable industry shift, OpenAI has rented Google TPUs to help power ChatGPT and its other AI products, marking the first significant time the firm has used chips outside Nvidia’s ecosystem.

OpenAI, already one of the largest buyers of Nvidia’s GPUs for training AI models and running inference workloads, has expanded its hardware strategy by turning to Google’s tensor processing units (TPUs). This move highlights how OpenAI rents Google TPUs through Google Cloud to meet soaring demand for computing power, Reuters first reported earlier this month, underscoring an unexpected collaboration between two fierce AI rivals.

For Google, landing OpenAI as a customer strengthens its push to open up its proprietary TPUs to external clients — chips that were previously reserved mostly for internal use. This strategy has helped Google secure deals with major players like Apple, as well as AI startups Anthropic and Safe Superintelligence, both founded by former OpenAI executives.

OpenAI Looks to Cut Costs with Google TPUs

According to The Information, this marks the first meaningful instance where OpenAI rents Google TPUs instead of relying solely on Nvidia GPUs or Microsoft’s extensive data centers, reflecting a shift in how the Sam Altman-led company manages its infrastructure. OpenAI reportedly hopes that using TPUs rented via Google Cloud will significantly reduce the cost of inference — the process where an AI model applies its learned knowledge to new data.

However, the report also noted that Google isn’t making its most advanced TPUs available to OpenAI, indicating that competitive dynamics still shape these partnerships. Google declined to comment on the matter, and OpenAI did not immediately respond to Reuters.

Adding OpenAI to its growing list of high-profile customers illustrates how Google is leveraging its in-house AI capabilities — from specialized chips to advanced software — to accelerate the growth of its cloud business, even as it competes directly in the race to dominate AI. By branching out, OpenAI’s move to rent Google TPUs could also elevate TPUs as a viable, potentially more affordable alternative to Nvidia’s industry-leading GPUs.