Latest / Chips, compute and energy

LaunchComputeFrontierUSConfirmed

Google announces TPU v5p with 8,960-chip pods and says Gemini was trained on TPUs

See how this story connects

Google Cloud announced TPU v5p and its AI Hypercomputer architecture on 6 December 2023, the day Gemini launched. A v5p pod links 8,960 chips, each with more than twice the FLOPS and three times the high-bandwidth memory of TPU v4, and Google said it trains large language models 2.8 times faster than v4. Google said Gemini was trained on and is served using TPUs.

Why it matters

Google showed a frontier model trained without NVIDIA hardware and offered the same chips to cloud customers, adding a large source of training compute outside NVIDIA.

When
6 Dec 2023
ID
WTA-N-2023-0168
Who
Google, Google Cloud

Summary by wheresthe.ai, from Google Cloud Blog. Checked against the source on 9 Oct 2026.

Source copy: download the page as saved on 18 Oct 2026 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: Led toGoogle open-sources TorchTPU, a native PyTorch backend for its TPUs, with Meta

Cite this story

WTA-N-2023-0168: “Google announces TPU v5p with 8,960-chip pods and says Gemini was trained on TPUs.” wheresthe.ai, 6 Dec 2023. https://wheresthe.ai/d/google-tpu-v5p-ai-hypercomputer-2023-12/ (source: Google Cloud Blog).

Spotted an error? Email [email protected].