Google and Marvell in Talks to Build Two New AI Chips as TPU Push Intensifies

Google is in talks with Marvell Technology to develop two new chips aimed at running AI models more efficiently, Reuters reported, citing two people with knowledge of the discussions.

The two chips under discussion serve different but connected purposes:

A memory processing unit designed to work alongside Google's existing tensor processing unit (TPU), aimed at improving how AI models handle and move data during inference.
A new TPU built specifically for running AI models, separate from Google's current TPU lineup.

According to the report, the companies aim to finalize the design of the memory processing unit as soon as next year, before handing it off for test production.

Why Google Is Building Its Own Silicon?

Google has been pushing to make its TPUs a viable alternative to Nvidia's dominant GPUs. TPU sales have become a key driver of growth in Google's cloud revenue as the company works to show investors that its AI investments are generating returns.

Experts think that the development of a dedicated memory processing unit points to a specific bottleneck Google is trying to address: moving large volumes of data efficiently between memory and compute during AI inference, a constraint that becomes more significant as models grow larger.

Why Marvell?

Marvell is an established player in custom silicon and data infrastructure chips, making it a natural partner for this kind of specialized hardware development. The partnership, if confirmed, would mark another step in Google's broader strategy to reduce dependence on third-party GPU suppliers and build a full-stack AI infrastructure across its cloud platform.