ai
Google Cloud Announces Eighth-Generation TPUs Split into Training and Inference Chips
Google Cloud unveiled its latest tensor processing units, divided into models for AI training and inference, with significant performance gains. The company also committed to offering Nvidia's Vera Rubin chip later in 2026 and is collaborating on networking technology. TechCrunch Disrupt is scheduled for October 13-15, 2026, in San Francisco.
Apr 22, 2:46 PM·1m read1 source
TE
Substrate placeholder — needs reviewAudio version
Tap play to generate a narrated version.
Developing·Limited corroboration so far. This page will refresh as more sources emerge.
Google Cloud announced its eighth generation of tensor processing units on Wednesday, splitting the lineup into two chips: the TPU 8t geared for model training and the TPU 8i aimed at inference. Inference refers to the ongoing usage of models after users submit prompts, according to details from the announcement.
The new TPUs provide up to three times faster AI model training compared to previous generations, along with 80% better performance per dollar.
These chips enable more than 1 million TPUs to work together in a single cluster, as stated in the reveal. Google Cloud promises its cloud will have Nvidia’s Vera Rubin chip available later in 2026. Google created and open sourced the software-based networking technology called Falcon in 2023 under the Open Compute Project.
The company has agreed to work with Nvidia to engineer computer networking for Nvidia-based systems in its cloud. The two companies are collaborating to improve the Falcon networking technology. Nvidia stands as a nearly $5 trillion market cap company.
Separately, TechCrunch Disrupt 2026 will take place in San Francisco from October 13-15, 2026. @techcrunch reported these developments, including the TPU specifications and the partnership details.
