Google has cut a deal with Marvell that could hand the chip designer a $12.2bn payday.
Marvell Technology said on Wednesday that the pair will develop specialised hardware for Google’s AI chips, known as tensor processing units.
According to the Financial Times, the deal gives Google a warrant to buy up to 58.9m Marvell shares at $206.58 each, according to a regulatory filing. That puts the possible share grab at about $12.2bn, which is not exactly sofa change.
Marvell helps hyperscalers such as Amazon design their own AI accelerator chips, giving them a route away from Nvidia’s pricey graphics processing units. Nvidia remains the outfit everyone wants to avoid, even as they buy its kit by the truckload.
Marvell’s shares rose eight per cent after the announcement, extending a run that has more than tripled its stock in 12 months. Broadcom, Google’s main TPU supplier, saw its shares fall five per cent.
The expanded Marvell deal lands as Google starts selling TPUs to outside customers, rather than keeping them mainly for its own data centres. Google has built a financing machine to supply more than $150bn of AI chips to Anthropic.
In April, Google signed a deal with Broadcom to develop TPUs and other AI infrastructure components until 2031. The Marvell agreement suggests the search giant wants more suppliers in the tent before the AI spending frenzy gets sillier.
Marvell said it will work with Google on custom semiconductor products tied to the TPU ecosystem. These include “AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute”.
The first 1.3m warrant shares vest in equal quarterly chunks during the first year. The rest vest through “discretionary purchases” from Marvell’s third quarter of 2027 to the end of 2033.
Each $500m in revenue from custom products unlocks another tranche, a tidy incentive to keep the TPU gravy train running. Marvell is already tied into the AI plumbing rush through Nvidia’s $2bn investment and its buyout of Celestial AI photonics.
