Marvell’s Google AI Chip Deal: What the $120 Billion Sales Ceiling Really Means

Marvell’s expanded agreement with Google is best understood as a performance-based custom-silicon partnership, not a guaranteed $120 billion chip order. The deal links Google’s purchasing decisions to a potential equity stake in Marvell and broadens the company’s role across Google’s Tensor Processing Unit (TPU) infrastructure. 2

The deal in brief

The companies signed the commercial agreement on July 29, 2026, and Marvell disclosed it in an August 19 regulatory filing. The partnership covers custom semiconductor products designed to attach to Google’s TPU ecosystem, including AI inference accelerators, storage controllers, network-interface controllers, memory-interface controllers, and near-memory-compute products. 11

Marvell also issued Google a warrant to purchase up to 58,970,907 Marvell shares at $206.58 per share. Exercising the full warrant would require about $12.18 billion and could make Google Marvell’s fifth-largest investor, according to reporting based on the filing. 2313

Why the $120 billion figure is conditional

Most of the warrant does not vest immediately. The shares are divided largely into 240 tranches, with one tranche tied to every $500 million of qualifying revenue Marvell recognizes from Google and its affiliates for the covered custom products. At full vesting, those milestones add up to approximately $120 billion in cumulative potential sales through fiscal 2033. 512

That distinction matters. The $120 billion figure represents the purchase volume required for the warrant to vest fully; it is not a firm commitment that Google will place $120 billion in orders. Nor does the warrant force Google to buy Marvell shares. A smaller initial portion—about 1.36 million shares—vests over the first year independently of revenue, while the bulk depends on purchases. 37

For investors, the practical checklist is therefore straightforward:

• Booked revenue: what Marvell has actually recognized.
• Future purchasing milestones: what Google may buy if the programs scale.
• Warrant dilution: the potential effect of issuing additional Marvell shares.
• Execution risk: whether Marvell can deliver the products and Google deploys them at the implied scale.

What Google gets from the partnership

Google has spent years developing TPUs for its own AI workloads and cloud infrastructure. Marvell’s role spans more than a single accelerator: it includes components that support networking, memory, storage, and compute around the TPU system. 19

Adding Marvell gives Google another source of custom-silicon development and supply. That can reduce dependence on a single partner, increase negotiating leverage, and support a broader effort to use workload-specific infrastructure alongside Nvidia’s general-purpose AI GPUs. The available reporting supports the diversification thesis, but it does not establish that Marvell will replace Nvidia or Broadcom across Google’s hardware stack. 19

What Marvell stands to gain—and risk

For Marvell, the attractive feature is the breadth of the opportunity. If Google scales its TPU infrastructure, Marvell could participate in several layers of the system rather than relying only on sales of one processor. Networking, memory, storage, and near-memory-compute products could create a large, recurring data-center opportunity. 111

The risk is concentration. Much of the commercial upside depends on one customer expanding the relevant programs and continuing to recognize purchases under the agreement. The warrant also creates a potential dilution issue if Google exercises its rights, although that exercise would bring cash into Marvell at the fixed price. The headline valuation opportunity should therefore be treated as contingent on execution and demand, not as present-day revenue. 25

Broadcom is still involved—but its position is less exclusive

The Marvell agreement does not, on the evidence provided, cancel Broadcom’s relationship with Google. Broadcom announced a separate long-term agreement in April 2026 to develop and supply future generations of Google’s custom AI chips and other components for next-generation AI racks through 2031. 18

Marvell’s participation nevertheless changes the competitive picture. Broadcom has co-designed Google TPUs for about a decade, while Marvell’s agreement covers a comprehensive range of custom programs attached to the TPU ecosystem. That makes it harder to view Broadcom as Google’s only strategically important custom-silicon partner and could increase competitive pressure over pricing, design wins, and supply allocation.

The precise division of responsibilities between the two companies is not fully detailed in the cited material, so it is premature to describe Marvell’s deal as a direct replacement for Broadcom’s contract.

Why investors reacted positively

Marvell shares rose nearly 10% after the announcement, with other reports describing an intraday gain of as much as 13% to $243.66. The reaction reflected investor enthusiasm about Marvell’s potential role in custom AI silicon and the credibility that comes from a major hyperscaler relationship. 3914

That response also illustrates the central interpretive risk: markets may capitalize a large future opportunity before the associated revenue is booked. The $120 billion ceiling is a useful measure of the agreement’s maximum milestone-linked scale, but it should not be read as confirmed sales or guaranteed earnings.

How it compares with AMD and OpenAI

The Marvell-Google structure resembles AMD’s October 2025 agreement with OpenAI in one important respect: the customer received an equity-linked incentive connected to buying or deploying the supplier’s AI hardware. AMD agreed to provide OpenAI with up to six gigawatts of Instinct GPUs across multiple generations and issued OpenAI a warrant for up to 160 million AMD shares, with vesting tied to specified milestones.

The structures are not identical. Marvell’s warrant is tied primarily to qualifying revenue tranches—one for each $500 million of covered purchases—while AMD’s OpenAI warrant was tied to deployment milestones and other conditions described in AMD’s filing.

The common theme is strategic alignment. A major customer receives a potential ownership interest in the supplier, and the supplier gains a demand anchor that may support product co-design, capacity planning, and financing assumptions. These arrangements can reflect genuine demand for rapidly expanding AI infrastructure, but they also make it more important to distinguish:

• customer commitments from realized purchases;
• planned capacity from deployed systems;
• supplier revenue from end-user AI demand; and
• equity incentives from ordinary commercial economics.

The bigger strategic takeaway

Marvell’s agreement shows how the AI hardware market is moving beyond a simple Nvidia-versus-alternatives narrative. Hyperscalers are pursuing custom silicon, specialized networking, memory, and storage technologies to tailor infrastructure to their workloads and broaden their supply options. 19

For Google, the deal is a way to add capacity and optionality around its TPU ecosystem. For Marvell, it is a potentially transformative customer relationship whose value depends on execution. For Broadcom, it introduces a serious competitor into a supply chain where it has held a long-standing role.

The most defensible conclusion is not that Marvell has secured $120 billion of revenue. It is that Google has created a powerful incentive structure around a broad custom-silicon partnership—and that the resulting opportunity, competition, and financial entanglements will need to be measured against actual shipments, recognized revenue, warrant vesting, and sustained AI infrastructure demand.