Google announced two new TPU systems on 9 November โ TPU 8tx and TPU 8i Max โ expanding its eighth-generation TPU portfolio to four systems in a single year, which the company says is a first. They join the TPU 8t for large-scale training and TPU 8i for real-time inference, both introduced at Google Cloud Next in April.
TPU 8tx is built for secure and confidential inference and small-scale training. It pairs the 8t silicon with fifth-generation AMD EPYC host processors and Google's low-latency Boardfly network, and Google claims up to two times lower latency than TPU 8t with a near-zero performance penalty from its hardware-backed confidential computing, which protects data and model weights while they are in use. TPU 8i Max tackles the memory wall: two TPU 8i chips in one package, doubling compute, on-chip SRAM and high-bandwidth memory bandwidth against a single 8i, aimed at massive context windows and memory-heavy inference.
Google stresses these are new systems, not new silicon โ the change is in packaging, networking, memory configuration and security rather than a fresh chip architecture. The company has published no public pricing, general-availability date or broad rollout details for either system, and the figures above are Google-reported, not independently benchmarked.
The two systems target the constraints Google sees blocking enterprise AI agents in production: keeping sensitive data protected while models actually use it, and feeding enormous working contexts to agents without starving the processors. TPU 8tx is aimed at regulated enterprises and government workloads running proprietary models on sensitive data; TPU 8i Max is aimed at agents working across large codebases or lengthy video inputs.