Google Expands AI Chip Portfolio with Two New Specialized TPUs
Google showed off two new tensor processing units (TPUs). The chips grow what its AI hardware can do. One targets confidential model training. The other targets workloads that need more memory. The new chips are Google's latest move to compete with Nvidia in AI infrastructure. Google's pitch: specialized silicon for specific jobs. That is meant to pull in enterprises with heavy AI training and inference needs.
New TPUs support confidential model training for enhanced data privacy
Increased memory means you can deploy larger models.
Google keeps pushing TPUs as a cheaper option than Nvidia GPUs.