Archive — US — techcrunch
Google Cloud launches two new AI chips to compete with Nvidia
Google Cloud has launched its eighth generation of AI chips, TPU 8t for model training and TPU 8i for inference, achieving up to 3x faster performance and 80% better cost efficiency. While positioning itself as a competitor to Nvidia, Google will still offer Nvidia's chips in its infrastructure, collaborating on networking to enhance efficiency for mutual benefit.
Key points
- Google Cloud announced the TPU 8t for model training and TPU 8i for inference.
- These chips can deliver up to 3x faster training speeds and 80% improved performance per dollar.
- Google aims to reduce energy consumption and costs for customers with these new TPUs.
- Google is enhancing its cloud service but is not directly replacing Nvidia's GPUs, maintaining a partnership with Nvidia.
- Analysts note that while this signals Google's growth in AI, Nvidia remains a dominant player in the market.
Why it matters
- The launch reflects the growing trend of companies developing custom AI chips, such as Amazon and Microsoft.
- The competition indicates an ongoing arms race in AI capabilities among major cloud providers.
- Historically, Nvidia's GPUs have dominated AI training and inference markets, posing significant challenges for emerging competitors like Google.
Google Cloud's new AI chips mark a strategic step in enhancing cloud-based AI services, yet the ongoing reliance on Nvidia indicates a complex relationship that benefits both companies at this stage.