Marvell Technology has struck a landmark deal with Google that could reshape the competitive landscape of AI chip development, granting the search giant rights to purchase up to $12.2 billion in Marvell shares in exchange for a custom chip partnership.
The agreement, announced August 19, positions Marvell as a key supplier of chips for Google’s tensor processing unit ecosystem. Google received a warrant to buy up to 58.97 million Marvell shares at $206.58 each, which would make it the chipmaker’s fifth-largest investor if fully exercised. The deal could generate roughly $120 billion in revenue for Marvell through fiscal 2033, depending on whether Google meets the targets tied to its stake option.
What the Deal Covers
The partnership spans a broad range of technologies used alongside Google’s TPUs, including processors that run AI models, manage data storage, and move information across data center networks. The chips are designed to complement Google’s existing TPU lineup, which has become an increasingly important revenue driver for Google Cloud as companies seek alternatives to Nvidia’s dominant graphics processors.
Marvell stock jumped nearly 8 percent following the announcement, while Bloomberg reported that Broadcom, which had been Google’s primary custom chip partner, fell more than 5 percent. Alphabet shares were little changed.
A Growing Trend of Chipmaker-Supplier Partnerships
The deal follows a pattern of Big Tech companies locking in chip supply through equity-like arrangements. In October, AMD agreed to supply OpenAI with AI chips worth tens of billions annually while giving the ChatGPT maker the option to buy up to roughly 10 percent of the chipmaker. More recently, Nvidia agreed to provide a backstop of up to $105 billion for a data-center project OpenAI is leasing in Ohio.
This is a big win for Marvell, said Morningstar analyst William Kerwin, but added that he saw this news as a growing pie at Google for new sources, rather than a competitive displacement of Broadcom.
The surge in demand for in-house chips stems from companies wanting cheaper alternatives to Nvidia’s graphics processors, which are optimized for training AI models but can be less efficient for inference, the process of running trained models. Google’s recent overhaul of its AI division, which shifted power toward executives closer to Google Cloud, has further spotlighted the strategic importance of custom chip infrastructure.
Sources: Reuters; Bloomberg; CNBC; BNN Bloomberg
discussion