Marvell Grants Google Warrant in TPU Chip Deal
On August 19, Marvell Technology agreed to help develop Google's custom AI chips and granted Google a warrant to buy up to 58.97 million Marvell shares. Reuters calculated that the warrant could be worth about $12.18 billion if fully exercised, with most of it contingent on Google meeting purchasing targets through Marvell's fiscal 2033.
On August 19, Marvell Technology agreed to help develop Google's custom chips and granted the company a warrant to buy up to 58.97 million Marvell shares at $206.58 per share, Reuters reported. If fully exercised, the warrant would be worth about $12.18 billion, according to Reuters calculations.
The agreement covers chips and related technologies designed to work with Google's Tensor Processing Unit, or TPU, ecosystem, which Reuters describes as supporting much of Google's AI infrastructure. Reuters reported that most of the warrant becomes available only if Google reaches agreed purchasing targets through Marvell's fiscal 2033. Based on LSEG data cited by Reuters, a fully exercised stake would make Google Marvell's fifth-largest investor.
A supply-linked equity arrangement
The deal combines custom-chip development with a long-dated commercial incentive. The warrant's value and vesting conditions tie Google's potential ownership position to its purchases from Marvell over several years, rather than representing an immediate $12.2 billion stake purchase.
Reuters reported that Marvell shares rose more than 11% in premarket trading following the announcement, while Broadcom shares fell more than 2%. Broadcom has been Google's principal custom-chip partner and, according to Reuters, has an agreement with Google to develop and supply future generations of custom AI chips and components for next-generation AI racks through 2031.
TPU demand and AI infrastructure
Reuters framed the agreement against growing demand for internally designed accelerators, including TPUs, as companies seek alternatives to Nvidia GPUs and hardware tailored to inference workloads. Inference is the process of running trained models, and its cost, latency, and power requirements can differ substantially from those of model training.
For infrastructure teams, the reported agreement underscores an industry pattern: large AI operators increasingly combine internally specified silicon with specialized external design and supply partners.
Reuters also reported that major technology companies recently reinforced expectations of more than $700 billion in AI infrastructure spending this year, compared with $400 billion last year. The Marvell agreement adds another custom-silicon supply relationship to that broader investment cycle, with Nvidia GPUs and Broadcom's existing Google relationship also part of the competitive landscape.
Key Points
- 1Marvell's Google warrant could reach $12.18 billion, but Reuters reports that most vesting depends on purchase targets through fiscal 2033.
- 2The agreement adds Marvell to Google's custom-chip relationships alongside Broadcom, increasing competition around TPU-related silicon and AI rack components.
- 3Across the industry, supply-linked equity incentives can bind accelerator procurement to multi-year hardware roadmaps and specialized infrastructure integration.
Scoring Rationale
The agreement links a major AI infrastructure buyer with a custom-silicon supplier through a potentially $12.18 billion warrant and multi-year purchasing conditions. It is consequential for practitioners tracking TPU supply, inference hardware competition, and the expanding custom-accelerator ecosystem.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
