OpenAI reportedly plans Google TPU purchases as it diversifies beyond Nvidia GPUs

The Information reported that OpenAI intends to buy Google's tensor processing units, though the size of the purchase and detailed terms were not disclosed. The report points to a shift for OpenAI, which has relied heavily on Nvidia graphics processors. OpenAI earlier unveiled its first in-house accelerator, developed with Broadcom, with initial deployment planned before 2026 ends. The move also sits in a wider market for Google's chips. Anthropic reached an agreement with Google and Broadcom in April for multiple gigawatts of next-generation TPU capacity, which it expects to begin operating in 2027. Oracle Cloud Infrastructure has said it will become the first hyperscaler outside Google to deploy TPUs. For builders, more buyers for TPUs could mean more capacity and different price points than GPU-only clouds offer. Teams planning long training runs should track which accelerators their providers can supply, since model code and kernels often need changes between chip families.