{"version":1,"type":"story","url":"https://digestai.news/story/oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027","json":"https://digestai.news/story/oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027.json","markdown":"https://digestai.news/story/oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027.md","slug":"oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027","headline":"Oracle Cloud to add Google TPUs for AI training and inference in 2027","summary":"Oracle and Google announced on October 27, 2026, that Oracle Cloud Infrastructure (OCI) will offer Google’s Tensor Processing Units (TPUs) starting in 2027. This makes OCI the first non-Google hyperscale cloud to provide Google TPUs, which will support large-model training, inference, and agentic AI workloads. Oracle’s infrastructure will host the TPUs, which include two eighth-generation models: TPU 8t for training and TPU 8i for inference. TPU 8t delivers **3x** the compute performance per pod over the previous generation and cuts model development cycles from months to weeks, while TPU 8i is optimized for long-context reasoning and high-concurrency tasks like agentic AI. Google claims TPU 8t offers **2.7x** performance-per-dollar for training and TPU 8i **80% better** performance-per-dollar for low-latency inference, with both chips improving efficiency by **2x per watt** compared to prior models. The TPUs integrate Arm-based Axion CPUs to reduce host bottlenecks and feature advanced networking (Virgo Network) and storage (TPUDirect RDMA) for faster data access. Availability will ramp over multiple years, with Google’s own TPUs already powering its Gemini models and products like Search and Maps.","keyPoints":["Oracle Cloud will offer Google’s TPU 8t (training) and TPU 8i (inference) starting in 2027","TPU 8t delivers **3x** compute performance per pod and cuts training cycles from months to weeks","Google claims TPU 8i offers **80% better** performance-per-dollar for low-latency inference workloads"],"whyItMatters":"This partnership expands access to Google’s high-performance TPUs beyond its own cloud, giving AI teams more flexibility to choose hardware for training and inference. The TPU 8t’s speed and efficiency could accelerate frontier model development, while TPU 8i’s improvements for agentic AI may lower costs for real-time reasoning workloads.","category":{"slug":"hardware","name":"Hardware & Compute","url":"https://digestai.news/category/hardware"},"entities":{"companies":["Oracle","Google"],"models":[],"people":["Mahesh Thiagarajan","Amin Vahdat"]},"firstPublishedAt":"2026-10-27T16:46:05Z","updatedAt":"2026-10-27T16:46:05Z","sourceCount":2,"hasPrimarySource":false,"sources":[{"outlet":"Unite.AI","title":"Oracle Cloud Infrastructure to Offer Google’s TPU Accelerators","url":"https://unite.ai/oracle-cloud-infrastructure-to-offer-googles-tpu-accelerators","publishedAt":"2026-10-27T16:46:05Z","type":"press","primary":false,"lead":true},{"outlet":"finance.yahoo.com","title":"Oracle Brings Google TPUs, Snowflake, AI Security and Media Tools to Cloud Infrastructure","url":"https://finance.yahoo.com/technology/ai/articles/oracle-brings-google-tpus-snowflake-165707329.html","publishedAt":"2026-10-27T09:57:00Z","type":"press","primary":false,"lead":false}],"sourceNotes":null,"discussions":[],"thread":{"title":"Oracle Nvidia OpenAI Build AI Infrastructure Empire","url":"https://digestai.news/thread/nvidia-ceo-huang-forecasts-3t-4t-global-ai-infrastructure-spend-by-2030","storyCount":4},"cite":{"text":"Digest AI, \"Oracle Cloud to add Google TPUs for AI training and inference in 2027\", 27 October 2026, https://digestai.news/story/oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027","publisher":"Digest AI","title":"Oracle Cloud to add Google TPUs for AI training and inference in 2027","datePublished":"2026-10-27T16:46:05Z","url":"https://digestai.news/story/oracle-cloud-to-add-google-tpus-for-ai-training-and-inference-in-2027"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}