On October 28, 2026, The Information reported that OpenAI plans to buy Google's TPU chips, the same day Oracle said it would offer TPUs to its cloud customers. Reporting since then has OpenAI discussing a one-gigawatt TPU commitment with neoclouds, alongside an eight-gigawatt Ohio lease that NVIDIA backstops at 4.25 gigawatts. The open argument is whether TPU cost advantages in inference outweigh NVIDIA's CUDA ecosystem, and when supply will outpace demand.