Google Takes Aim at Nvidia With New AI Chips Built for Privacy and Massive Memory
Summary
Google introduced two eighth-generation TPU configurations: TPU 8tx, which uses hardware-based confidential computing to protect data during processing, and TPU 8i Max, designed to provide more computing power and memory for AI agents handling large contexts. Google claims 8tx can reduce latency by up to half versus its predecessor with little performance cost, while open-source TPU Sync can cut some data-transfer delays by more than 90%. The systems expand Google’s alternatives to Nvidia GPUs, but performance comparisons are against earlier Google hardware. Google gave no pricing or availability dates; reported OpenAI interest in its TPUs remains unconfirmed.
