Google added TPU 8tx for confidential inference and the dual-chip TPU 8i Max

On November 9, 2026, Google announced two more systems in its eighth-generation TPU lineup, TPU 8tx and TPU 8i Max, joining the TPU 8t and TPU 8i announced in April. Google framed them as hardware “built for the agentic era.”

TPU 8tx uses the same silicon as TPU 8t, paired with 5th Gen AMD EPYC CPU hosts, and is built for confidential inference: it isolates sensitive data and model weights in hardware while they are in use. It also uses TPU 8i’s low-latency Boardfly network, which Google says gives up to 2x lower latency than TPU 8t with a near-zero performance penalty. Google pitches it for fine-tuning, distillation and reinforcement learning on private data without moving that data between clusters.

TPU 8i Max puts two TPU 8i chips on one package, doubling compute, SRAM and HBM bandwidth over TPU 8i and fitting twice the silicon into the existing rack footprint. Google targets it at very large context workloads, such as many agents working across one codebase or hours of video. For reference, the April figures were up to 2.7x better training performance for TPU 8t and 80% better inference price-performance for TPU 8i over the previous generation.

Why it matters: confidential computing on accelerators addresses a real blocker for enterprises and governments that will not send private data or proprietary weights to shared cloud hardware, and Google is now selling it as a TPU product rather than a GPU-only option.

What it does not show: the post gives no pod sizes, availability dates or benchmarks for either new system, and the April generational gains apply to 8t and 8i, not to 8tx or 8i Max.