Google announced TPU 8tx and TPU 8i Max on November 9, extending its eighth-generation accelerator portfolio. The company explicitly distinguishes these new system configurations from designing new silicon.

Different configurations for different constraints

TPU 8tx uses TPU 8t silicon with AMD EPYC hosts and a low-latency network, targeting confidential inference and private model customization. Google describes hardware isolation for sensitive data and model weights while in use.

TPU 8i Max packages two TPU 8i chips together. Google reports twice the compute, SRAM and HBM bandwidth of TPU 8i, targeting large context windows and greater rack density.

These are vendor specifications, not independently reproduced throughput or cost results. The announcement does not establish a new Gemini checkpoint, a higher context limit for a public model, or a tariff for either system.