Google is adding two new variants of its in-house AI chips, the TPU 8tx and the TPU 8i Max, Bloomberg reported on Monday.
TPU stands for tensor processing unit, Google's own processor for AI work, built as an alternative to buying general-purpose graphics chips. Amin Vahdat, Google's chief technologist for AI infrastructure, told Bloomberg that the two are reworked versions of existing chips, built for particular kinds of workloads. The reported targets are secure AI and long-context AI, meaning models that handle very long inputs such as large documents or extended conversations. Specifications, pricing and release timing have not been reported.
The word "reworked" is the useful detail. Vahdat's description points to adapting existing designs rather than starting over, which is typically a shorter path than a full new generation. It also fits a wider pattern: Amazon and Microsoft build their own AI chips too, and cloud providers increasingly compete on silicon they control instead of relying only on Nvidia.
How far the changes go is unclear. A reworked chip can mean modest tweaks to memory or software support, or deeper changes to the design, and the reported comments do not say which. Until Google publishes numbers, the 8tx and 8i Max are best read as a statement of intent: Google wants named products for named needs, and security and long inputs are the two it chose to name.