Google Cloud Unveils Next-Gen AI Chips to Challenge Nvidia

Google Cloud's Next-Gen AI Chip Strategy

Google Cloud has unveiled its eighth generation of custom-built AI chips, or tensor processing units (TPUs), which will be split into two distinct chips: the TPU 8t for model training and the TPU 8i for inference.

The Performance Boost

The new TPUs promise significant performance upgrades, including up to 3x faster AI model training, 80% better performance per dollar, and the ability to cluster over 1 million TPUs together. This should result in more compute power at a lower energy consumption and cost for customers.

Supplementing, Not Replacing Nvidia

While Google's new chips are a strategic move, they are not a direct challenge to Nvidia's future. Instead, Google will continue to offer Nvidia-based systems in its infrastructure, with plans to make Nvidia's latest chip, Vera Rubin, available later this year. The company is also collaborating with Nvidia on software-based networking tech called Falcon.

The Future of AI Chip Development

The hyperscalers, including Amazon, Microsoft, and Google, are investing heavily in their own AI chips. While this may reduce their reliance on Nvidia in the long term, the current market dynamics suggest that Nvidia will continue to thrive. Google's growth as an AI cloud provider could, in fact, lead to more business for Nvidia.

Collaboration and Innovation

Google and Nvidia are working together to engineer computer networking that allows Nvidia-based systems to perform more efficiently in Google's cloud. This partnership highlights the complex and collaborative nature of the AI chip ecosystem.