Chips · accelerators · Google

Google TPU7x (Ironwood) · Ironwood

2,307 teraflops of dense BF16 compute a chip; Google publishes no power figure

Shipped 31 Mar 2026; shipped after the first date promised, up to 3 months. Epoch AI names it at 10 AI data centers and no GPU cluster. Google across the site.

Embed

The frame credits Google, Epoch AI, AI data centers and Epoch AI, Data on GPU clusters and links back to its page. Preview it.

Cite
CSV Post on X

Each figure in Google’s own words

one chip, dense compute; memory is high-bandwidth memory
BF16, dense
2,307 TFLOPS Peak compute per chip (BF16) (TFLOPs) 459 918 2307 TPU7x (Ironwood) (read from a table with a column for each product, whose head reads “Specification v5p v6e (Trillium) TPU7x (Ironwood)”)
FP8, dense
4,614 TFLOPS Peak compute per chip (FP8) (TFLOPs) 459 918 4614 TPU7x (Ironwood) (read from a table with a column for each product, whose head reads “Specification v5p v6e (Trillium) TPU7x (Ironwood)”)
Memory
192 GiB Each chip is equipped with 192 GiB of HBM TPU7x (Ironwood)
Memory bandwidth
7.38 TB/s HBM bandwidth per chip (GBps) 2765 1638 7380 TPU7x (Ironwood) (7380 GB/s on the page; read from a table with a column for each product, whose head reads “Specification v5p v6e (Trillium) TPU7x (Ironwood)”)
List price
$12.00 a chip-hourIronwood us-central1 Iowa $12.00 Cloud TPU pricing (on-demand, us-central1, per chip-hour); 192 TFLOP-hours of dense 16-bit compute a dollar

The 2025-11-06 post is titled as a general availability announcement but says GA will come in the coming weeks; the release notes give Preview on 2025-11-24 and GA on 2026-03-31. The TPU7x page also states HBM bandwidth as approximately 7.37 TB/s.

2 dated promises and the shipment

each date links to the page it is quoted from
  1. 9 Apr 2025 Promised Ironwood available to Google Cloud customers in 2025. when it becomes available later this year Ironwood: The first Google TPU for the age of inference
  2. 6 Nov 2025 Promised general availability "in the coming weeks", read as by the end of 2025 in the fourth quarter of 2025. Ironwood, our seventh generation TPU, will be generally available in the coming weeks. Announcing Ironwood TPUs General Availability and new Axion VMs to power the age of inference
  3. 31 Mar 2026 Shipped: TPU7x generally available (GA). March 31, 2026 Feature Generally available: TPU7x is generally available (GA). Cloud TPU release notes

Named at 10 AI data centers

H100-equivalents and power are the whole site’s today, with every chip it holds
SiteOwnerCountryH100eMWChips
Google Columbus Google United States 331,986 364 TPU v5e, TPU v5p, TPU v6e, TPU v7
Google Council Bluffs (East) Google United States 255,685 284 TPU v5e, TPU v6e, TPU v5p, TPU v7
Google Bristow Google United States 209,861 286 TPU v5e, TPU v5p, TPU v6e, TPU v7
Google Omaha Google United States 198,585 284 TPU v5e, TPU v5p, TPU v6e, TPU v7
Google Lincoln Google United States 197,575 170 TPU v6e, TPU v7
Google The Dalles Google United States 175,341 185 TPU v6e, TPU v5p, TPU v5e, TPU v7
Google Midlothian Google United States 143,581 123 TPU v7, TPU v6e
Google Arcola Google United States 108,135 93 TPU v6e, TPU v7
Anthropic Lake Mariner AI XPV Platform United States 58,565 49 TPU v7
Anthropic Barber Lake Cipher Mining United States 0 0 TPU v7

In no GPU cluster of Epoch AI’s list

Epoch AI’s list of AI supercomputers, which stops in July 2025

Epoch AI’s names for this chip: TPU v7.

Other Google chips

  • TPU v4 275 teraflops of dense BF16 compute a chip; Google publishes no power figure
  • TPU v5e 197 teraflops of dense BF16 compute a chip; Google publishes no power figure
  • TPU v5p 459 teraflops of dense BF16 compute a chip; Google publishes no power figure
  • TPU v6e 918 teraflops of dense BF16 compute a chip; Google publishes no power figure

Sources

accelerators.csv Every dataset → Status → Glossary →