Google is in talks with Marvell to build custom AI inference chips as it diversifies beyond Broadcom

April 20, 2026 - 00:55

In a strategic move to diversify its hardware supply chain, Google is reportedly in advanced discussions with semiconductor firm Marvell Technology. The talks center on the development of custom chips specifically designed for artificial intelligence inference tasks.

This potential partnership would introduce a third major design partner into Google's Tensor Processing Unit (TPU) ecosystem, a critical infrastructure powering its vast AI services. The initiative underscores Google's push to reduce reliance on any single supplier, such as Broadcom, and gain greater control over the performance and efficiency of its data center hardware.

The shift toward custom Application-Specific Integrated Circuits (ASICs) is a defining trend in the tech industry. As companies like Google seek competitive advantages in the AI race, off-the-shelf components are increasingly passed over for tailor-made solutions. This demand is fueling remarkable market growth, with forecasts indicating custom ASIC sales could surge by 45% by 2026.

By collaborating with Marvell, a company with deep expertise in data infrastructure silicon, Google aims to further optimize the speed and cost of running trained AI models. This hardware evolution is essential for deploying next-generation AI applications at scale across its cloud and consumer services.