What is a TPU?

A TPU (Tensor Processing Unit) is Google’s custom-designed AI accelerator chip, optimized for machine learning workloads. The supplied material tracks its scale of investment, technical design trade-offs, and evolving business strategy.

• Sep 2025 - Jensen Huang said that when TPUs become a large business, customers will own their own tooling, implying a shift from Google-only use.
• May 2026 - Krishna Rao described TPU investment as “an over hundred billion dollar commitment.”
• May 2026 - Reiner Pope noted that TPUs have deterministic latency in their core, but achieving both deterministic latency and high speed is challenging.
• May 2026 - Reiner Pope explained that GPUs have higher data bandwidth between vector and matrix units than TPUs, due to more wiring lines.
• May 2026 - Andrew Feldman said TPU users are already stepping outside of Google’s own data centers.
• Jun 2026 - Nathaniel Whittemore reported that Google placed an order for 3 million TPUs to be manufactured in 2028.