Google Cloud launches two new AI chips to compete with Nvidia

Google Cloud has introduced its eighth-generation custom AI chips, known as Tensor Processing Units (TPUs), splitting this release into two specialized chips. The TPU 8t is designed for AI model training, while the TPU 8i targets inference tasks—the ongoing process of receiving and interpreting user prompts with trained models.

Performance and Cost Benefits

Google claims these new TPUs deliver significant advancements compared to their predecessors, including:

• Up to 3x faster model training
• 80% improvement in performance per dollar
• Ability to connect over one million TPUs in a single cluster

These enhancements promise customers more computing power and energy efficiency at a lower cost.

Relationship with Nvidia

Despite Google’s advances with its custom chips, the company is not turning away from Nvidia hardware. Instead, these TPUs are intended to supplement the Nvidia-based systems already critical to Google’s cloud services. Google also announced upcoming support for Nvidia’s latest chip, Vera Rubin, scheduled to be added later this year.

Looking Ahead

While tech giants like Google, Amazon, and Microsoft are developing their own AI chips, they continue to rely on Nvidia for essential workloads. Industry analysts note that Nvidia’s dominance remains strong, even as cloud providers grow their proprietary chip portfolios.

Collaborative Innovation

Google and Nvidia are actively working together to optimize data center efficiency, particularly by enhancing Falcon, a software-based networking technology developed and open-sourced by Google. This effort aims to further boost networking for Nvidia chips inside Google Cloud environments.

For the original article by Julie Bort, visit TechCrunch.