--- title: "TPU: what it means · The AI Compute Race · DRYBURGH" source: https://dryburgh.com/vol1/glossary/tpu description: "TPU: Tensor processing unit, Google's own accelerator, which comes off TSMC's lines like NVIDIA's. Chapter 5." --- # TPU What does “TPU” mean? TPUTensor processing unit, Google’s own accelerator, which comes off TSMC’s lines like NVIDIA’s. *Chapter 5.* From *The AI Compute Race*. [Read the book](https://dryburgh.com/vol1#buy) ## Where the book uses it - [Chapter 5 — The Island in Every Machine](https://dryburgh.com/vol1/notes/chapter-5)Why every accelerator runs through Taiwan: the foundry, the packaging and the memory that are booked years ahead, and what that does to anyone arriving late with cash. ## Sources that use it - Amin Vahdat, “Our eighth generation TPUs: two chips for the agentic era,” Google, The Keyword, 22 April 2026 (the eighth-generation TPU announced as two parts, TPU 8t for training and TPU 8i for inference, designed with Google DeepMind; TPU 8t nearly 3x compute performance per pod, 9,600 chips and 121 ExaFlops a superpod; TPU 8i 288 GB of HBM with 384 MB of on-chip SRAM; both up to twice the performance per watt of Ironwood and “generally available later this year”; both parts announced as “coming soon”). [In the Sources](https://dryburgh.com/vol1/sources/company#google-tpu8) - Anthropic, “Anthropic expands partnership with Google and Broadcom for multiple gigawatts of next-generation compute,” 6 April 2026 (“We have signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity that we expect to come online starting in 2027”; “Our run-rate revenue has now surpassed $30 billion—up from approximately $9 billion at the end of 2025”; of customers spending over $1 million on an annualized basis, over 500 at the Series G in February and “Today that number exceeds 1,000, doubling in less than two months”; the Claude Code run-rate and the GitHub estimate this row carried until 19 September 2026 are the Series G post’s, not this page’s). [In the Sources](https://dryburgh.com/vol1/sources/company#anthropic-google-broadcom) - Anthropic, “Expanding our use of Google Cloud TPUs and services,” 23 October 2025 (access to up to one million TPUs; “well over a gigawatt” of capacity coming online in 2026; Broadcom’s separate Supply Assurance Agreement with Google for networking components through 2031 is cited under its own key). [In the Sources](https://dryburgh.com/vol1/sources/company#anthropic-tpu-expansion) - Anthropic, “Series H,” 28 May 2026 ($65 billion at $965 billion post-money, MGX a significant investor; “We signed agreements with Amazon for up to five gigawatts of new capacity, with Google and Broadcom for five gigawatts of next-generation TPU capacity, and with SpaceX for access to GPU capacity in Colossus 1 and Colossus 2”). [In the Sources](https://dryburgh.com/vol1/sources/company#anthropic-series-h) - Broadcom, third-quarter fiscal 2026 results call, 2 September 2026, Hock Tan, chief executive, and Amie Thuener, finance chief, read in two third-party transcripts whose quoted words agree, Benzinga as served by Webull and Bullfincher, Broadcom publishing the webcast replay and no transcript (“we expect Anthropic to deploy another 5 gigawatts of TPU version 8i in 2027” after “the 1 gigawatt of Ironwood we are deploying in 2026”, and in 2028 a “clear line of sight” to a further 10 gigawatts; Anthropic “on track to become our largest XPU customer in 2027”; “we’re talking about fiscal years”; of the 30 gigawatts across six customers, “It doesn’t necessarily mean all 30 gigawatts need to have been deployed within that period”; Broadcom’s own content per gigawatt “less than $30 billion”, after an analyst’s “your competitor suggested something like 40”; the finance chief on the financing platform, “to enable more than 20 gigawatts of compute infrastructure for OpenAI and Anthropic by the end of 2028”, “We closed the first $35 billion tranche in June for Anthropic’s 1 gigawatt deployment”, and “Where necessary, we may provide modest residual value guarantees, which are contingent liabilities we view as low risk”; Tan, asked by Vivek Arya why Broadcom finances Anthropic and OpenAI, of the two laboratories: for every gigawatt of compute they deploy they could achieve $30 billion of annual recurring revenue, with no basis given; the call’s audio is not held, so the words are the transcripts’ and the book paraphrases them; Tan on Google, in Bullfincher’s transcript: “This has been recently strengthened by a long-term agreement to develop and supply future generations of TPUs and AI networking. Under this agreement, we are planning to deliver multi-tens of billions of dollars of TPUs annually over the next several years”) [In the Sources](https://dryburgh.com/vol1/sources/company#broadcom-q3fy26-call) And 19 more: [search the Sources](https://dryburgh.com/vol1/sources). ## Related terms - [Accelerator](https://dryburgh.com/vol1/glossary/accelerator) - [TSMC](https://dryburgh.com/vol1/glossary/tsmc) - [Advanced packaging](https://dryburgh.com/vol1/glossary/advanced-packaging) - [AI factory](https://dryburgh.com/vol1/glossary/ai-factory) - [Booked, allocated, rationed](https://dryburgh.com/vol1/glossary/booked-allocated-rationed) - [CHIPS Act](https://dryburgh.com/vol1/glossary/chips-act) - [Convertible](https://dryburgh.com/vol1/glossary/convertible) - [CoWoS](https://dryburgh.com/vol1/glossary/cowos) - [Custom silicon](https://dryburgh.com/vol1/glossary/custom-silicon) - [Cut-out](https://dryburgh.com/vol1/glossary/cut-out) - [DRAM](https://dryburgh.com/vol1/glossary/dram) - [EUV](https://dryburgh.com/vol1/glossary/euv) From the glossary of The AI Compute Race. - [Campuses](https://dryburgh.com/vol1/ledger) - [Tests](https://dryburgh.com/vol1/tests) - [Scorecard](https://dryburgh.com/vol1/scorecard) - [Data](https://dryburgh.com/vol1/data/releases) - [Chapter by chapter](https://dryburgh.com/vol1/companion) - [The record at a glance](https://dryburgh.com/vol1/observatory) - [The machine](https://dryburgh.com/vol1/machine) - [One campus, four futures](https://dryburgh.com/vol1/hyperion) - [Glossary](https://dryburgh.com/vol1/glossary) - [People](https://dryburgh.com/vol1/people) - [Sources](https://dryburgh.com/vol1/sources) - [Notes](https://dryburgh.com/vol1/notes) - [Where the book was wrong](https://dryburgh.com/vol1/corrections) - [Where its sources were wrong](https://dryburgh.com/corrections?book=vol1) - [The Paper](https://dryburgh.com/vol1/paper) - [Consent](https://dryburgh.com/vol1/consent) - [Who builds and pays](https://dryburgh.com/vol1/parties) *The AI Compute Race: Five Races, One Machine* is a book by Lee S. Dryburgh, the first volume of THE COMPUTE ECONOMY, published by Dryburgh, an imprint of Contestra OÜ, Tallinn, Estonia; its record closes on 8 November 2026. [About the book](https://dryburgh.com/vol1)[Tell me when it is out](https://dryburgh.com/vol1#notify)