Google adds two AI chip systems, one built to keep sensitive data protected even while it is being used
TPU 8tx is for governments and regulated businesses; TPU 8i Max doubles up chips for very large AI jobs. Google gave no prices or release dates.
Google, based in Mountain View, announced two more systems built on its eighth-generation AI chips Monday, according to a company blog post. TPU 8tx is for running AI on sensitive data, and TPU 8i Max is for very large AI jobs. Google says it is the first time it has introduced four TPU systems in one year, after TPU 8t and TPU 8i in April.
What a TPU is
A Tensor Processing Unit, or TPU, is a chip Google designs just for AI. Google’s documentation says it is built to do the huge grids of multiplication, called matrix math, that AI models run on. Graphics processing units, or GPUs, are more general-purpose; Google’s guide says GPUs fit models with many custom operations, while TPUs fit models dominated by matrix math. Google trains and runs its own Gemini AI models on TPUs.
The two new systems
Data is usually encrypted when stored or sent, but must be unlocked in memory while a computer works on it, Google says. TPU 8tx, paired with AMD EPYC processors, keeps data and the AI model locked and walled off even while in use, an approach called confidential computing. Google aims it at governments and regulated businesses, and claims up to 2 times less delay than TPU 8t.
TPU 8i Max puts two TPU 8i chips in one package, doubling computing power and memory speed, Google says. It is meant for jobs like many AI agents working through a whole codebase or hours of video, which need a large context window, the amount of information an AI can consider at once.
Google says nearly 100 customers use TPUs, and some can now put them in their own data centers. It gave no prices or release dates, and its TPU pricing page does not list the new systems. All performance figures are Google’s.
Source: Google blog, Nov. 9, 2026; Google Cloud: Introduction to Cloud TPU; Google Cloud TPU pricing