Google Cloud Unveils Two New AI Chips to Challenge Nvidia

Google Cloud has unveiled its eighth generation of custom AI chips, dividing them into two categories: the TPU 8t designed for model training and the TPU 8i intended for inference. These new TPUs promise training speeds up to three times faster, an 80% increase in performance per dollar, and the capability to cluster over 1 million units. However, Google is not phasing out Nvidia; it plans to introduce Nvidia's Vera Rubin chip later this year. Additionally, Google is working with Nvidia on Falcon networking software to enhance the efficiency of Nvidia-based systems in its cloud. Analyst Patrick Moorhead remarked that his 2016 forecast about Google's TPUs negatively impacting Nvidia was incorrect, as Nvidia currently boasts a market cap nearing $5 trillion. This development highlights the fierce competition among hyperscalers like Amazon, Microsoft, and Google in the AI chip arena while still depending on Nvidia.