Back to all articles
•Technology

Google Unveils Next-Gen AI Chips: Collaboration with Nvidia Continues

View original source

Google Cloud has introduced its eighth generation of custom-built AI chips, termed Tensor Processing Units (TPUs), which will be divided into two categories: TPU 8t for model training and TPU 8i for inference tasks.

  • Performance: The new TPUs boast up to 3x faster AI model training and offer 80% better performance per dollar. They enable over 1 million TPUs to function concurrently in one cluster, aiming to provide enhanced computational power with reduced energy consumption and lower costs.
  • Market Positioning: Despite these advancements, Google's chips are not directly challenging Nvidia's dominance. The tech giant continues to integrate and utilize Nvidia's latest chips, like the Vera Rubin, while concurrently working on its TPUs.
  • Strategic Collaboration: Google is partnering with Nvidia to enhance the Falcon software-based networking tech, aiming to improve the efficiency of Nvidia-based systems within its cloud infrastructure.
  • Technological Evolution: As hyperscalers like Google, Amazon, and Microsoft develop proprietary AI chips, there's potential for them to be less reliant on Nvidia in the future. However, Nvidia's current market cap indicates its strong market position.
  • Context: The article hints at the larger trend of tech giants building their own chips to service their cloud environments more efficiently, signaling gradual shifts in the dynamics of cloud AI infrastructure.