Google unveils two new TPU 8 chip variants
Original: Google broadens its AI hardware portfolio with two new specialized chips
Why This Matters
Specialized AI chips signal hyperscalers are moving beyond general-purpose GPU reliance.
Google announced TPU 8tx and TPU 8i Max on Monday, expanding its eighth-gen TPU lineup to cover confidential AI inference and large-workload memory needs, cutting cloud reliance on third-party chips.
Google unveiled two new tensor processing unit variants aimed at plugging gaps in its in-house AI chip lineup. The TPU 8tx, co-designed with AMD, is built for confidential AI training and inference — it embeds security at the hardware level, keeping model weights and sensitive data isolated even during active use. It also includes a custom inferencing network that lets training and reasoning run simultaneously, which Google says could allow companies to train models on proprietary data without moving that data between devices. The second chip, TPU 8i Max, is effectively two TPU 8i inference chips combined into a single package, doubling memory capacity, processing power, and bandwidth. That lets models handle larger context windows during queries. Both chips extend the eighth-gen TPU family Google introduced in April. The announcement fits a broader pattern: Microsoft, Amazon, OpenAI, and Anthropic are all building proprietary AI silicon to reduce Nvidia dependence. Google's moves suggest it is filling increasingly specific niches rather than just scaling raw compute.