Google is deepening its push into custom AI chips with Marvell Technology, giving the chipmaker a role in building hardware around Google’s Tensor Processing Unit ecosystem and for the search giant receiving the right to acquire up to $12.2 billion worth of Marvell shares.
The agreement covers several types of custom silicon, including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory computing. Marvell disclosed the deal in a regulatory filing, describing it as an expanded partnership with Google.
The arrangement also gives Google a warrant to purchase 58.97 million Marvell shares at $206.58 per share. If fully exercised, the shares would cost roughly $12.2 billion.
The $12.2 Billion is Tied to Google’s Chip Purchases
Google did not simply hand Marvell $12.2 billion and become a major shareholder overnight, as the warrant is structured around the commercial relationship between the companies.
About 1.36 million shares will vest over the first year. The remaining shares are divided into 240 tranches, with one tranche vesting for every $500 million in revenue from Marvell’s custom products purchased by Google and its affiliates. And the warrant can be exercised until August 18, 2033.
This means the potential $12.2 billion stake is closely connected to how much custom silicon Marvell sells to Google’s businesses over the coming years.
And Marvell estimates the agreement could generate up to $120 billion in revenue through fiscal 2033 if the relevant targets are reached.
Google is Expanding Beyond its Existing Chip Strategy
Google has spent years developing its own TPUs to handle AI workloads inside its data centers and through Google Cloud. The company has also relied heavily on Broadcom for custom chip development.
The Marvell agreement gives Google another major supplier for custom silicon and expands the technology being developed around its TPU ecosystem. Marvell’s work will cover components beyond the main AI accelerator, including chips responsible for storage, networking, and memory.
This broader approach reflects the growing importance of custom silicon as cloud companies build larger AI infrastructure. Hyperscalers are developing specialized chips because they can be designed around particular workloads instead of relying entirely on general-purpose GPUs.
While Nvidia remains a major supplier of AI accelerators, Google has already built an alternative architecture through its TPUs. The Marvell agreement gives that strategy more room to expand.
Marvell is Becoming More Important to Big Tech
The Google deal also strengthens Marvell’s position in the custom AI chip market.
The company already works with major technology companies on custom silicon, including Amazon and Microsoft. Its expanded relationship with Google adds another major cloud provider to that list.
And investors immediately noticed the significance of this partnership. Marvell shares jumped as much as 13% after the agreement was announced, while Broadcom shares fell more than 5%.
However, the financial payoff for Marvell may still take time. CEO Matt Murphy said the largest contribution from the Google relationship is expected to become more visible from fiscal 2029, even though revenue covered by the agreement through fiscal 2028 was already included in the company’s forecasts.
Google’s latest move, therefore, says more about where its AI infrastructure is heading than about Nvidia’s position today. Google is building more of the silicon underneath its AI systems itself, while bringing in additional partners to design specialized components around its TPU platform.
For Nvidia, this means another hyperscaler is investing heavily in an alternative path to AI computing. For Google, it means greater control over the chips powering its growing AI business.
