Marvell rises on Google AI chip talks; Broadcom shares slip

Marvell Technology saw its shares rise after reports that Alphabet’s Google is in talks to co-develop custom artificial intelligence chips. The initial coverage, attributed to The Information and later summarized by Reuters, said the discussions focus on chips designed to improve inference efficiency for AI models. Both Marvell and Google offered limited public comment immediately after the reports.

According to the reporting, Google and Marvell explored two projects: a memory processing unit intended to complement Google’s tensor processing unit (TPU) and a new TPU optimized for inference workloads. The reporting noted that design work could advance to initial design completion within the coming year, though Reuters said it could not independently verify those timelines.

The news arrived against the backdrop of a deeper Marvell–Nvidia relationship: Nvidia disclosed a strategic partnership with Marvell and an equity investment of roughly $2 billion at the end of March, linking Marvell into Nvidia’s NVLink Fusion ecosystem for AI infrastructure. The tie-up positions Marvell to supply custom XPUs and networking elements compatible with Nvidia’s interconnect fabric while Nvidia provides supporting hardware and interconnects.

Markets reacted quickly. Marvell shares jumped in pre-market and early trading—reports cited gains in the mid-single-digit percentage range—while Broadcom shares experienced modest declines as investors weighed the potential for Google to broaden its supplier base for TPUs and related components. The moves underline investor sensitivity to shifts in hyperscaler sourcing for AI hardware.

In the broader industry context, major cloud providers and AI developers are pursuing both in-house silicon and multiple external design partners to control costs and mitigate supply-chain concentration risks. Google has long worked with Broadcom on TPU design, but diversification toward additional partners like Marvell would reflect a strategic hedging as demand for inference-optimized hardware grows.

Analysts caution that talks do not equal contracts and that chip development, validation and deployment timelines can span multiple years. Near-term, headlines will likely drive volatility in MRVL, AVGO, GOOGL and NVDA; longer-term implications depend on whether reported design agreements are formalized and scaled into production. Investors should monitor official filings and company statements for confirmation.