Google Cloud has announced its eighth generation of custom AI chips, split into two models: TPU 8t for training and TPU 8i for inference.
Performance Claims
- 3x faster AI model training
- 80% better performance per dollar
- 1 million+ TPUs in a single cluster
- Reduced energy and cost compared to previous versions
Competition with Nvidia
Google’s chips don’t replace Nvidia—they supplement it. Google Cloud will offer both TPUs and Nvidia’s upcoming Vera Rubin chip later in 2026.
Partnership
Google and Nvidia are collaborating on networking technology called Falcon to improve efficiency in Google Cloud.
Enterprise Impact
The chips offer enterprises more affordable AI compute, potentially lowering entry barriers for AI development.
Written by Massin BSN