In one of the most significant AI infrastructure deals of 2026, Marvell Technology has granted Alphabet’s Google the right to purchase up to $12.2 billion in Marvell shares as part of a sweeping new custom AI chip partnership. Announced this week, the arrangement ties Google’s equity stake directly to its chip purchases from Marvell, potentially delivering as much as $120 billion in revenue to the chipmaker through fiscal 2033. It marks a decisive escalation in Google’s strategy to control more of its AI hardware stack.
What Was Announced
Under the terms disclosed on August 19, 2026, Marvell will issue warrants giving Google the option to acquire nearly 59 million shares at a fixed strike price of $206.58 per share. If fully exercised, the warrants would make Google the fifth-largest investor in Marvell, valued at approximately $12.18 billion. The structure is deliberately performance-linked: roughly 1.4 million warrant shares vest in the first year, with the remaining tranches unlocking incrementally for every $500 million of chips that Google purchases from Marvell.
The deal covers a broad range of Marvell’s technology portfolio, spanning custom silicon that runs AI models, storage controllers that manage vast datasets, and networking chips that move data between accelerators. In particular, the partnership expands Marvell’s involvement in the ecosystem around Google’s Tensor Processing Units (TPUs), the custom AI accelerators that power Google Cloud, Gemini training runs, and internal AI workloads.
Marvell shares surged nearly 10% on the news, reflecting investor confidence that Google’s commitment represents one of the largest and longest-duration cloud silicon contracts ever disclosed publicly. Analysts have described the deal as a vote of confidence from a top hyperscaler that could reshape Marvell’s revenue trajectory well into the next decade.
The announcement also comes at a pivotal moment for the AI infrastructure market, as leading cloud providers race to secure custom silicon capacity ahead of anticipated demand for next-generation AI training and inference workloads.
Technical Details
The partnership focuses on Marvell’s custom application-specific integrated circuit (ASIC) design services, an area where the company has quietly become one of the world’s most important suppliers. Rather than selling off-the-shelf chips, Marvell co-designs silicon tailored to a customer’s specific workload, then manages advanced packaging, high-bandwidth memory integration, and manufacturing coordination with foundry partners such as TSMC.
For Google, this translates into deep support for its TPU roadmap and the surrounding data center architecture. That includes optical interconnects, coherent DSPs (digital signal processors) for high-speed networking, and specialized storage accelerators. As AI training clusters expand into hundreds of thousands of accelerators networked together, the components that shuttle data between them have become as strategically important as the accelerators themselves.
The warrant-based structure is notable in its own right. By tying equity vesting to purchase volume, Marvell aligns its financial incentives directly with Google’s growth, while Google gains a form of long-term supplier lock-in without the operational complexity of an outright acquisition. It is a hybrid model that other hyperscalers may study closely.
Industry Impact and Reactions
The Marvell-Google agreement lands amid a wave of hyperscaler investment in custom silicon and supply chain integration. Nvidia has recently backstopped $250 billion in OpenAI’s Ohio data center financing, Anthropic has expanded its multi-gigawatt compute partnership with Google and Broadcom, and Amazon continues to scale its Trainium and Inferentia chip families. Against that backdrop, Google’s move reinforces a pattern: the biggest AI companies are no longer content to be pure customers of chip suppliers.
For Marvell, the deal validates a strategic pivot the company has pursued for years, positioning itself as the go-to custom-silicon partner for hyperscalers that want Broadcom-caliber engineering without depending on a single vendor. It also underscores growing competitive pressure on Broadcom, which has long dominated the custom AI ASIC market alongside its work with Google on earlier TPU generations.
Industry observers note that the size of the deal, the length of its runway, and the equity linkage together represent a new template for cloud-silicon partnerships. Rather than transactional purchase orders, hyperscalers appear increasingly willing to commit capital, equity, and multi-year volume guarantees to secure priority access to advanced chip design and manufacturing capacity.
What Comes Next
The first tranche of warrant shares vests during the initial year of the deal, with the remainder unlocking over the following years as Google’s chip purchases accumulate. Full realization of the $12.2 billion option, and the projected $120 billion in cumulative Marvell revenue, depends on Google hitting purchase milestones through fiscal 2033. Investors and analysts will be watching quarterly disclosures closely for early indicators of pace.
Beyond the financial mechanics, the strategic milestones to watch include new TPU generations that leverage Marvell-designed components, expansion of Google’s data center footprint, and any parallel announcements from competing hyperscalers seeking to strike similar structural deals with alternative silicon partners.
Conclusion
Google’s $12.2 billion Marvell option is more than a supplier contract — it is a strategic realignment of how leading AI companies think about hardware, capital, and control. As the industry races to build out the compute base for the next wave of AI models, deals like this one signal that the boundary between chip customer and chip investor is blurring fast. Expect more agreements of this shape, and larger, in the months ahead.
Stay updated on the latest AI news at Evolve Digital.
