TPU 8t and TPU 8i (eighth-generation TPUs)
Eighth-generation TPUs split into TPU 8t for training and 8i for inference; an 8t superpod scales to 9,600 chips and 121 exaflops.
Two specialized chips for agentic workloads. TPU 8t targets over 97% goodput, near-linear scaling to a million chips with the Virgo network, and nearly 3x compute per pod versus Ironwood. General availability promised for later in 2026.
- Date
- Wednesday, 22 April 2026
- Lab
- Kind
- infra
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| TPU 8t superpod | 9,600 chips, 121 exaflops 2 PB shared HBM; company figures | company |
Peak figures at Google's chosen precision.
Sources
- blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-
- blog.google/innovation-and-ai/technology/ai/google-ai-updates-april-2026/
This record was checked against its sources on 6 October 2026. How we check