Overview
- Google announced two new categories of Tensor Processing Units, increasing its family to four chips that are tuned to different AI workloads.
- One new chip class adds hardware support for confidential computing so models can access and process sensitive data without exposing it.
- The company said the expanded lineup targets AI agents and cyber‑security workloads and marks a faster product cadence after releasing two processors this year.
- Nvidia remains the dominant supplier with demand outstripping supply, but that gap and recent reputation pressure create openings for cloud providers and other challengers.
- The market for AI accelerators is worth hundreds of billions of dollars and this push could shift where companies run AI workloads and how they protect user data.