{"data":{"slug":"google-tpu-8","etype":"chip","name":"Google TPU 8t / TPU 8i","cat":"AI Accelerator","subtitle":"Google's eighth TPU generation, split for the first time into a training chip and an inference chip.","summary":"Announced at Google Cloud Next on April 22, 2026, the eighth TPU generation comes as two chips. TPU 8t targets training: 12.6 PFLOPS of FP4 and 216 GB of HBM per chip, in 9,600-chip superpods with about 2 PB of shared memory. TPU 8i targets serving and reinforcement learning: 10.1 PFLOPS of FP4, 288 GB of HBM and 384 MB of on-chip SRAM, linked by a new Boardfly topology. Google claims up to 2x performance per watt over Ironwood. In November 2026 it added TPU 8tx (confidential computing) and TPU 8i Max (two 8i dies per package); a general availability date is not confirmed.","attrs":[{"g":"Overview","l":"Designer","v":"Google"},{"b":"Official","g":"Overview","l":"Announced","n":"Google Cloud deep dive, 2026-04-22","v":"April 22, 2026 (Google Cloud Next)"},{"g":"Overview","l":"Status","n":"blog.google, Apr 2026; Cloud TPU release notes list no TPU 8 entry","v":"Announced: GA 'later this year' per Apr 2026 launch; no GA date confirmed as of Nov 2026"},{"b":"Vendor spec","g":"Compute","l":"TPU 8t FP4 (peak)","n":"Per chip; Google Cloud deep dive, Apr 2026","u":"PFLOPS","v":"12.6"},{"b":"Vendor spec","g":"Compute","l":"TPU 8i FP4 (peak)","n":"Per chip; Google Cloud deep dive, Apr 2026","u":"PFLOPS","v":"10.1"},{"b":"Vendor spec","g":"Memory","l":"TPU 8t HBM","n":"Per chip; Google Cloud deep dive","v":"216 GB, 6,528 GB/s"},{"b":"Vendor spec","g":"Memory","l":"TPU 8i HBM","n":"Per chip; Google Cloud deep dive","v":"288 GB, 8,601 GB/s"},{"b":"Vendor spec","g":"Memory","l":"On-chip SRAM","n":"Google Cloud deep dive","v":"128 MB (8t); 384 MB (8i)"},{"b":"Vendor claim","g":"Physical","l":"TPU 8t superpod","n":"blog.google, Apr 2026","v":"9,600 chips; ~2 PB shared HBM; 121 ExaFLOPS"},{"b":"Vendor claim","g":"Physical","l":"TPU 8i pod","n":"Google Cloud Next '26 post and blog.google, Apr 2026","v":"Up to 1,152 chips in Boardfly topology; ICI 19.2 Tb/s"},{"b":"Vendor claim","g":"Physical","l":"Scale-out","n":"Google Cloud, Apr 2026","v":"Virgo fabric: 134,000+ TPU 8t chips per data center; 1M+ across sites"},{"b":"Vendor spec","g":"Physical","l":"Host / cooling","n":"blog.google, Apr 2026","v":"Axion Arm CPU hosts; fourth-generation liquid cooling"},{"b":"Vendor claim","g":"Power","l":"Efficiency","n":"Google, Apr 2026","v":"Up to 2x performance per watt vs Ironwood"},{"b":"Official","g":"Overview","l":"Variants","n":"blog.google, 2026-11-09","v":"TPU 8tx (8t silicon, AMD EPYC hosts, confidential compute); TPU 8i Max (two 8i chips per package)"},{"g":"Manufacturing","l":"Process","n":"Not disclosed by Google; press reports of TSMC 2 nm unconfirmed","v":"Not publicly disclosed"}],"kf":{"year":2026,"class":"AI Accelerator","mem_bw":6.53,"memory":"216 GB (8t) / 288 GB (8i) HBM","designer":"Google"},"aliases":["TPU 8t","TPU 8i","TPU v8","TPU 8tx","TPU 8i Max","eighth-generation TPU"],"ticker":null,"exchange":null,"hq":null,"founded":null,"website":null,"status":"Announced","region":null,"asof":"2026-11-16","sources":[{"url":"https://cloud.google.com/blog/products/compute/tpu-8t-and-tpu-8i-technical-deep-dive","label":"Google Cloud: TPU 8t and TPU 8i technical deep dive (Apr 2026)"},{"url":"https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/","label":"Google: Our eighth generation TPUs (Apr 2026)"},{"url":"https://cloud.google.com/blog/products/compute/ai-infrastructure-at-next26","label":"Google Cloud: AI infrastructure at Next '26 (Apr 2026)"},{"url":"https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/","label":"Google: Introducing TPU 8tx and TPU 8i Max (Nov 2026)"}]},"meta":{"found":true,"count":1,"served_from":"site cache (no live upstream calls)","as_of":"2026-11-16T22:03:32+00:00","disclaimer":"Research and information only, not investment advice. Structured records carry basis labels (vendor spec / reported / estimate); market quotes are delayed/indicative.","attribution":"Records compiled from vendor documentation, official filings and reputable public reporting; market quotes via financial data providers. See /data-sources."}}