Marvell grants Google a warrant worth up to $12.2 billion tied to custom AI chip purchases

Marvell Technology has granted Google a warrant to buy up to $12.2 billion worth of Marvell stock, tying the size of Google's potential stake directly to how much custom AI silicon Google buys from the chipmaker through fiscal year 2033.

What's new

Under the expanded agreement, Google can purchase up to 58,970,907 Marvell shares at $206.58 apiece. If fully exercised, the warrant would be worth roughly $12.2 billion, making Google one of Marvell's largest shareholders. Vesting is conditional rather than immediate: most of the warrant only becomes exercisable as Google hits agreed purchasing targets from Marvell, with tranches tied to every $500 million of chips Google buys, running through Marvell's 2033 fiscal year. Marvell disclosed the arrangement in an SEC filing, describing the expanded partnership as covering products that "attach to the tensor processing unit ecosystem" — including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory compute technology built to work alongside Google's own TPUs.

Marvell shares jumped roughly 10% on the news, while shares of Broadcom — Google's longtime TPU co-design partner — declined, reflecting investor read-through that Google is diversifying its custom-silicon supplier base. Analysts following the filing estimate the underlying commercial agreement could bring Marvell on the order of $120 billion in revenue through fiscal 2033 if Google's purchasing hits the targets the stake option is tied to.

Context

Google has co-designed its Tensor Processing Units with Broadcom for close to a decade, making Broadcom the dominant custom-chip partner behind Google's AI infrastructure. Marvell's expanded role — now with an equity incentive tightly coupled to purchase volume — signals Google adding a second major silicon partner for the broader set of components that sit around the TPU itself: interconnects, memory controllers, and inference accelerators, rather than the TPU compute die. It follows a now-familiar pattern among the largest AI infrastructure buyers: Google, Amazon, Meta, and Microsoft have all been pushing further into custom silicon as a way to reduce dependence on, and cost exposure to, Nvidia's general-purpose GPUs for AI workloads at scale.

Why it matters

Structuring the incentive as a warrant that vests against purchase volume — rather than a straightforward supply contract or cash investment — aligns Marvell's upside directly with how much AI infrastructure spend Google actually routes its way, and gives public markets a rare, dollar-denominated signal of how seriously Google is scaling its custom-silicon roadmap. A stake option sized at $12.2 billion, against a potential $120 billion in Marvell revenue through 2033, implies a purchasing commitment large enough to meaningfully diversify Google's AI hardware supply chain away from single-vendor risk — whether that vendor is Broadcom for custom chips or Nvidia for merchant GPUs. For the broader AI hardware market, it is another data point that hyperscalers are willing to structure increasingly creative, long-duration financial arrangements to lock in chip supply for AI infrastructure build-outs.