Google TPUs to Run on Oracle Cloud, Making OCI the First Outside Hyperscaler With TPU
rseroter · x · 2026-10-28
Google Cloud and Oracle have partnered to bring Google TPUs to Oracle Cloud Infrastructure (OCI), making OCI the first hyperscaler outside Google to offer TPUs, with capacity rolling out starting calendar year 2027.
Google's blog stresses that the TPU's edge comes from system-level co-design: custom silicon paired with optical networking, power delivery, liquid cooling, and compiler software. The original TPU v1 delivered 30x the throughput per watt of contemporary processors; today's eighth-generation TPUs deliver up to 12.6 PFLOPS of native FP4 compute per chip—over 100x TPU v1's 92 TFLOPS—at up to 2x the performance per watt of the prior generation. TPUs currently train and serve Gemini across Google's billion-user products.
The post also notes that training economics hinge on goodput (share of time a cluster does productive computation) while inference hinges on latency and cost per token; by pairing on-die Inter-Chip Interconnect (ICI) with Optical Circuit Switching (OCS), the control plane detects hardware or link degradations and dynamically reroutes optical paths around faults in real time without job restarts.
More from Infra
- Palmer Luckey Unveils BlueChip, Billed as First American-Made Computer in Decades — whurley · 2026-10-28
- OpenRouter's entire DeepSeek traffic could run on a single GB300 rack, worth just $6-7M — teortaxesTex · 2026-10-28
- Open-source ai-phone: local LLM answers real calls on your Mac, nothing leaves the machine — Agent007_MI9 · 2026-10-28
- Agents built a custom inference engine: AMD MI350X serves GLM 5.3 FP8 37% faster than NVIDIA B200 — yisongyue · 2026-10-28
- Why RTX 5090s cost $10k: leaked factory footage shows cards converted into AI accelerators — Different_Fix_2217 · 2026-10-28
- Anthropic's S-1 admits its data centers rely on natural gas, exposing it to fuel costs — TobyWalsh · 2026-10-28