Google TPU expansion to Oracle and OpenAI

Oracle agreed to deploy Google TPU capacity beginning in 2027, becoming the first hyperscale cloud provider outside Google to do so. Oracle's Acceleron network connects the TPU clusters to Oracle databases and storage, enabling enterprises to train and serve models alongside their data without cross-cloud data transfers. Google claims its 8th-generation TPUs deliver up to 12.6 PFLOPS of FP4 compute per chip, keep clusters doing productive work over 97% of the time on large training runs, and give up to 80% better performance per dollar for low-latency serving. Google's open-source TorchTPU stack allows existing PyTorch code and frameworks including vLLM and SGLang to run on TPUs without model rewrites. Bloomberg reported that OpenAI plans to purchase AI chips from Google. Richard Ho, who runs OpenAI's chip team, previously worked on Google's TPU program.