While Nvidia is busy with a leadership crisis, its rivals are not standing still. Google this week announced two new AI chips in the eighth generation of its TPU family: TPU 8tx and TPU 8i Max. They join the TPU 8t and TPU 8i announced in April, and target two different needs of cloud customers.
What does each chip do?
The TPU 8tx is based on the same silicon as the 8t, but works with AMD EPYC processors instead of Google's Axion CPUs. It is designed for secure model serving, with hardware-level protection for data and model weights during processing. According to reports, Google may let select customers install it in their own data centers. The TPU 8i Max combines two 8i chips in one package, with double the compute, memory and bandwidth, and targets long-context models.
Why is it interesting?
The option to install TPUs at the customer's site is a significant change: until now Google's chips were available almost only through its cloud, and that was one of Nvidia's advantages, since it sells to everyone. Banks, governments and healthcare companies that don't want sensitive data to leave are exactly the customers Google is targeting. Still, Google has not published full specifications, and the performance figures vary between reports.
The big picture
Nvidia still dominates the AI chip market, but this week is a reminder that it is not alone. Google, Amazon and Microsoft are developing their own chips, AMD is improving its offering, and large customers are looking for alternatives so as not to depend on one supplier. A leadership crisis at Nvidia only gives them another reason.
