Google unveils TPU 8tx and 8i Max for secure, long-context AI

Google's TPU team has introduced four systems in 2026. 'This is the first time we have introduced four TPU systems in a single year,' Amin Vahdat, SVP and chief technologist for AI and infrastructure, wrote in a Nov. 9 post on the company blog. The newest two are TPU 8tx and TPU 8i Max.

The 8tx is the confidential-compute play. It runs on the same silicon as TPU 8t and is paired with 5th Generation AMD EPYC CPU hosts to deliver what Vahdat calls 'hardware-level confidential compute for private fine-tuning and secure agent execution.' He framed the buyer as 'regulated enterprises and governments running frontier or proprietary models and autonomous agents alongside their most sensitive data.' Google's benchmark claim: 'up to 2x lower latency compared to TPU 8t with a near-zero performance penalty.'

The 8i Max is the long-context play. It fuses two low-latency TPU 8i chips on a single package, 'delivering double the compute, double the SRAM and double the HBM bandwidth of TPU 8i,' Vahdat wrote, while fitting 'twice the silicon within the existing rack footprint.' His framing: 'With computing moving from human to agent scale, businesses are asking for accelerators that can reason across AI models' massive context windows.'

Google also open-sourced TorchTPU for native PyTorch execution, shipped a TPU Agent Suite, launched the XLA Serving Engine, and made Cloud Pathways generally available on Google Kubernetes Engine. The combined eighth-generation portfolio, Vahdat wrote, offers up to 2.7 times better training performance and 80% better inference price performance over the prior generation; a tool called TPU Sync is pitched as cutting latency by more than 90%.

Neither chip ships now. Vahdat told Bloomberg the systems will be available in 2027. On growth, he put the TPU business at 'nearly 100 customers,' spanning frontier labs, trading firms, retailers and universities.