Google Just Dropped Two New AI Chips in One Day — TPU 8tx and TPU 8i Max

Google's AI chip machine isn't slowing down. On Monday, November 9, the company unveiled two brand-new chips — TPU 8tx and TPU 8i Max — bringing its total to four TPU launches in a single calendar year, a first for the company. One is built for privacy. The other is built for enormous context windows. And in a twist few saw coming, one of them runs on AMD hardware.

What Google Just Announced

Google's AI infrastructure chief Amin Vahdat announced both chips in a November 9 post on the company blog.  They round out Google's eighth-generation TPU lineup, and Vahdat was blunt about the pace: this is the first time Google has introduced four TPU systems in a single year. Alongside the hardware, Google 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.  A tool called TPU Sync is pitched as cutting latency by more than 90%. Neither chip ships right away — both are slated for availability in 2027.

TPU 8tx: AI That Keeps Secrets

The 8tx is the confidential-compute play. It runs on the same silicon as the TPU 8t, but pairs it with 5th Generation AMD EPYC CPU hosts to deliver what Vahdat calls hardware-level confidential compute for private fine-tuning and secure agent execution. The target buyer: regulated enterprises and governments running frontier or proprietary models and autonomous agents alongside their most sensitive data. Google's benchmark claim is up to 2x lower latency compared to TPU 8t, with what it describes as a near-zero performance penalty. In plain English: AI agents that can work on your most private data without it ever leaving a locked-down enclave.

TPU 8i Max: Built for Giant Context Windows

The 8i Max is the long-context play. It fuses two low-latency TPU 8i chips onto a single package, delivering double the compute, double the SRAM and double the HBM bandwidth of a single TPU 8i — while fitting twice the silicon within the existing rack footprint. Vahdat's framing: with computing moving from human to agent scale, businesses want accelerators that can reason across AI models' massive context windows. As agents juggle longer memories and bigger documents, the chip underneath has to keep up.

Why This Matters

Taken together, Google says the eighth-generation portfolio offers up to 2.7 times better training performance and 80% better inference price performance than the prior generation. Vahdat put the TPU business at nearly 100 customers, spanning frontier labs, trading firms, retailers and universities. The bigger signal is the pace: four TPU launches in one year, plus open tooling around them, is Google turning its custom silicon from a quiet advantage into a full product line. And the AMD pairing inside the 8tx is a reminder that the AI chip race is no longer a one-company story — even Google is happy to mix and match hardware to win it.