Marvell Technology (MRVL) stock surged as much as 6.3% in premarket trading following a report that Alphabet's Google is in talks with the chipmaker to co-develop two new artificial intelligence (AI) chips. According to The Information, citing sources familiar with the matter, the companies are working on a memory processing unit (MPU) designed to work alongside Google's existing tensor processing unit (TPU), and a brand new TPU built specifically for AI inference workloads.
The goal is to finalize the design of the memory chip as soon as next year, after which the companies would move to test production. This partnership is a key part of Google's broader strategy to position its TPUs as a credible rival to Nvidia's dominant graphics processing units (GPUs) in the AI compute market.
Google has been building out its semiconductor ecosystem, working with partners including Intel and Broadcom alongside Marvell. Historically keeping TPUs internal, Google shifted strategy in 2022 when its cloud division took over external chip sales. The company has since ramped up production and sales, even selling TPUs directly into customers' own data centers—not just through its cloud platform.
Earlier this month, Google officially launched TorchTPU, a project to make its chips natively compatible with PyTorch, the dominant AI software framework. This move lowers the switching cost for developers considering moving off Nvidia's hardware. TPU sales have become a growing contributor to Google Cloud revenue as the company seeks to demonstrate returns on its massive AI infrastructure spending.
The push comes alongside the launch of Google's new AI model family, Gemma 4, designed for advanced reasoning. The chip development and model upgrades show Google is aligning its software and hardware stack as competition intensifies. Nvidia, meanwhile, is reportedly developing new AI inference chips using technology from Groq. Google's first-quarter 2025 earnings, due on April 29, are expected to provide further clarity on its AI chip investment and production scaling plans.
