Google TPU v8, Rubin GPUs, and Agentic AI Push
Key Questions
What are the key specs of Google's TPU v8?
Google's TPU v8 delivers 121 EFLOPS, offering 3x training performance and 8x inference speed improvements. It challenges Nvidia in efficiency for AI workloads.
What is the scale of Google's Rubin GPUs?
Google plans 960K Rubin GPUs, providing 4.8x the scale of Colossus. This supports advanced agentic AI applications.
What is Google's strategy for agentic AI at Cloud Next?
At Cloud Next, Google unveiled a full-stack 'OS' for agentic AI, including chips, models, data, and agent-to-agent interactions. It features Chrome AI co-pilot as an efficiency play.
TPU v8: 121 EFLOPS 3x train/8x inf; 960K Rubin GPUs 4.8x Colossus; new chips challenge Nvidia, Cloud Next full-stack agentic 'OS' (chips/models/data, agent-to-agent), Chrome AI co-pilot. Efficiency play for agents.
Sources (2)
Updated Apr 26, 2026