Glossary term

TPU

Tensor processing unit: a custom AI accelerator chip designed by Google.

How it works

A tensor processing unit is a chip Google designed specifically for the matrix math of machine learning. Many TPUs are joined into 'pods' with fast links to train very large models. They are available mainly through Google's cloud.

A TPU pod

Step 1Many TPU chips
Step 2Fast custom interconnect
Step 3Act as one large machine
Step 4Train a huge model

Why it matters

TPUs show the custom-chip path some large companies take instead of relying only on general-purpose GPUs.

Related terms

GPU ASIC Interconnect

Recent signals mentioning it

Price changeMajor7 h ago · Z.ai

GLM 5.3 output price cut 49%: $2.77 → $1.4 per 1M tokens

Input price$0.07 per 1M tokens-33%Output price$1.4 per 1M tokens-49%

Price change7 h ago · MiniMax

MiniMax M2.5 output price cut 21%: $1.2 → $0.95 per 1M tokens

Output price$0.95 per 1M tokens-21%

Price changeMajor12 h ago · Google

Gemma 4 26B A4B output price cut 25%: $0.3 → $0.225 per 1M tokens

Input price$0.068 per 1M tokens-25%Output price$0.225 per 1M tokens-25%

Price change18 h ago · MoonshotAI

Kimi K2.7 Code output price raised 13%: $3.1 → $3.5 per 1M tokens

Output price$3.5 per 1M tokens+13%