Oracle to Deploy Google TPUs Outside Google Cloud for the First Time From 2027
rohanpaul_ai · x · 2026-10-28
Google TPUs will run on a hyperscale cloud outside Google for the first time: Oracle will deploy capacity on OCI from calendar 2027. Oracle provides power, cooling and data center networking, with its Acceleron network linking TPU clusters to Oracle databases so enterprises can fine-tune and serve models next to their data.
- Google claims 8th-gen TPUs deliver up to 12.6 PFLOPS FP4 per chip, 97%+ cluster utilization on large training runs, and up to 80% better performance per dollar for low-latency serving.
- The open-source TorchTPU stack runs existing PyTorch code plus vLLM and SGLang on TPUs without rewrites, lowering the cost of moving off NVIDIA GPUs.
- The deal extends Google's TPU external sales push, which began earlier this year with smaller providers, to a direct hyperscale competitor.
Related event: Google TPU to Run Outside Google Cloud for the First Time via Oracle(4 posts)→
More from Infra
- Thermo-backed CTO says AI is in its dial-up phase, touts disaggregated compute for an ultrafast 'broadband moment' — beffjezos · 2026-10-28
- DeepSeek-V4.1-Flash Q2 runs at 41 tok/s on one RTX 3090, 90% token match vs API — musicurgy · 2026-10-28
- USearch v3 RC ships after 1,000 days, its libraries now embedded in 550K-star projects — viglovikov · 2026-10-28
- Will local-model progress outrun GPU depreciation? Reddit debates the buy-now window — Sensualities · 2026-10-28
- Dev cuts local AI inference power from 100W to 25W with only a 1.9s→2.8s slowdown — AryanEmbered · 2026-10-28
- SK Hynix posts $62.8B quarterly operating profit at 80% margin as AI memory demand booms — mhdfaran · 2026-10-28