Google and Marvell Technology inked a strategic equity and supply deal in which Google commits up to $12.2 billion in share purchases while Marvell supplies custom silicon for Google's AI infrastructure through 2033. The partnership covers TPUs, inference accelerators, networking, storage, and memory controllers, with Marvell revenue potentially reaching $120 billion over the contract term.
The deal signals Google's shift away from sole reliance on Broadcom for custom silicon supply while diversifying its chip ecosystem beyond Nvidia. By taking an equity stake and securing long-term supply commitments, Google is constructing an alternative vendor relationship for high-volume AI chip components—a parallel effort to its in-house Tensor Processing Unit (TPU) roadmap.
For the broader inference-driven market: Google's strategy underscores that supply-chain control and semiconductor partnership diversity are now as critical to AI competitiveness as model capability itself. The 10+ year contract horizon signals confidence in sustained demand for custom inference silicon.
For infrastructure planners: the $12.2B equity commitment reflects Google's bid to secure non-exclusive access to specialized memory, networking, and storage components as inference workloads scale. This pattern—large cap tech companies making equity stakes in suppliers to guarantee supply—is likely to expand as energy and manufacturing constraints drive competition for foundry capacity.