Google unveils TPU 8tx and TPU 8i Max, shipping four TPU systems in one generation

Google expanded its custom-silicon lineup with two new variants. TPU 8tx targets government agencies and regulated industries that need confidential computing, keeping data encrypted and isolated during processing. TPU 8i Max pairs two TPU 8i processors to support longer context windows and larger data workloads, aimed at the inference profile of agentic and long-document applications.

The bigger shift is cadence. Instead of a new generation roughly every two years, Google now ships multiple chip variants at once. Sundar Pichai wrote that "four distinct TPU systems in a single generation in one year" was possible because hardware engineers co-design with Google DeepMind, letting the company tune systems for specific workloads quickly.

Bloomberg framed the launch as an escalation of the rivalry with NVIDIA, and the timing is pointed: NVIDIA's board opened an investigation into its CEO the same day, and its shares are down about 20% for November. Google's leverage comes from captive demand: Anthropic holds an option on up to $150B of TPUs for 2028-2031, which gives Google a frontier-lab anchor customer outside its own Gemini work.

Caveats: the sources offer no performance-per-watt or price figures against NVIDIA's latest parts, and TPUs remain available mainly through Google Cloud rather than as merchant silicon. Watch for third-party benchmarks and whether more labs sign Anthropic-style capacity commitments.