Google isn't just buying more custom chips from Marvell. It's earning stock in the company for doing it.
Marvell disclosed Wednesday that it gave Google a warrant to purchase up to 58,970,907 shares of Marvell stock at $206.58 apiece, a stake worth up to $12.2 billion if fully exercised, roughly 7.7% of the company. Marvell shares jumped as much as 14% on the news. Rival custom-chip supplier Broadcom fell.
How the Warrant Actually Works
Google doesn't pay cash upfront for the shares. It earns the right to buy them gradually, unlocking a new tranche of the warrant for roughly every $500 million in custom chip purchases it makes from Marvell.
If Google exercised every tranche, Marvell would have generated around $120 billion in qualifying revenue from the relationship. That figure is a vesting threshold built into the contract, not a spending commitment Google has actually made.
The chips at the center of the deal aren't Google's main AI processors. They're the supporting cast: AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute, all described in the filing as products that "attach to" Google's tensor processing unit ecosystem.
Broadcom Loses Some Ground
For a decade, Broadcom has been Google's primary partner on custom AI silicon. That relationship isn't going away, the two companies expanded their own long-term agreement in April, running through 2031.
But Wednesday's news put a number on how seriously Google is diversifying that relationship. Marvell first signaled a possible Google partnership back in April too, when its stock rose on reports of early talks over a TPU and memory-processing chip. This week's disclosure turned that speculation into a structured, multiyear agreement with real financial stakes attached.
Broadcom's stock drop wasn't dramatic. It was still a signal that investors read this deal as Marvell gaining ground at Broadcom's expense, not just Google spreading its business around evenly.
There's a strategic logic behind having two suppliers rather than one. A single-vendor relationship, even a decade-long one, leaves Google exposed if that supplier hits capacity constraints, pricing disputes, or its own production problems. Splitting the custom silicon work between Broadcom and Marvell gives Google leverage in future negotiations with both.
The Bigger Custom Silicon Race
Google isn't alone in trying to reduce its dependence on Nvidia. Amazon, Meta and Microsoft have all been building or commissioning their own AI chips, chasing lower costs and processors better tuned to their specific workloads than general-purpose GPUs.
Analysts project Google's TPU-related infrastructure revenue could grow from roughly $3 billion this year to $25 billion by 2027. That growth requires exactly the kind of supporting silicon Marvell just agreed to help build.
The equity-linked structure is also becoming a familiar shape in AI infrastructure deals generally, tying a supplier's financial upside directly to a customer's purchasing volume, even when the mechanics run in the opposite direction of deals like Nvidia's recent financing of its own customers.
Prediction markets are still skeptical Google's underlying AI models can catch the leaders, giving the company long odds of having the best model on the market by year's end. Deals like this one are a reminder that Google isn't only competing on model quality. Owning more of the custom infrastructure powering its own AI stack is a separate kind of advantage, one that doesn't show up on a model leaderboard.
What This Means for Miami
Miami's growing base of family offices and private investors evaluating AI infrastructure exposure now has another deal structure worth understanding closely, one where a supplier's stock price is directly wired to a single customer's purchasing decisions.
That's a real concentration risk hiding inside what markets are treating as good news. A near 14% single-day jump on a warrant tied to future orders, not guaranteed revenue, is exactly the kind of headline number that Miami's more cautious institutional investors, the ones already applying real-estate-style due diligence to AI infrastructure bets, will want to look past before treating it as confirmed growth.