Google expands eighth-generation TPU lineup with two new chip systems

Google announced two additions to its eighth-generation TPU chip family: TPU 8i Max, built for handling very large context windows with the highest compute density per server rack, and TPU 8tx, designed for secure and confidential inference with hardware-level protection for sensitive data. These join TPU 8t for training and TPU 8i for reasoning, both introduced earlier in the year, which already deliver notably better training performance and inference price efficiency compared to the prior generation. This marks the first time Google has released four distinct TPU systems within a single year, reflecting how specialized hardware is now being tailored to different AI workloads rather than one-size-fits-all chips. Google said its TPU hardware now serves close to 100 customers, and the company is pairing these chips with broader PyTorch support and open software tools to make them easier to adopt outside Google's own infrastructure.

Why it matters

Shows cloud providers splitting AI hardware into specialized chips for context length, security, and cost rather than general-purpose designs.

Source: Google