Marvell Secures Major AI Chip Partnership with Google

What happened

Marvell Technology secured a multi-year commercial agreement with Google to design custom silicon for Google's Tensor Processing Unit (TPU) ecosystem. The partnership covers AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory computing technologies. As part of the deal, Marvell issued Google a warrant to purchase up to 58.97 million shares at an exercise price of $206.58 per share, potentially worth about $12.2 billion. However, the warrant structure ties most share vesting to performance milestones: tranches vest as Google generates qualifying revenue from Marvell products, with each $500 million in purchases unlocking additional shares through fiscal 2033. This revenue-linked structure means Google is not committing the full amount upfront, and the actual value depends on how much business Marvell wins over time. The agreement follows a similar structure Marvell established with Amazon Web Services in December 2024, which included warrants for 4.18 million shares at $87.78 also tied to revenue performance through January 2030.

Why it matters

This deal significantly expands Marvell's presence in the fast-growing custom AI silicon market and adds Google as a major hyperscale customer alongside existing partnerships with AWS and Microsoft. The revenue-linked warrant structure creates a direct alignment between Google's spending and Marvell's equity value, giving investors a measurable way to track the partnership's development. For Marvell, this validates its custom silicon strategy and positions the company to capture a larger share of the estimated $205 billion Google plans to spend on AI-related capital in 2026 alone. The agreement also demonstrates hyperscalers' strategic shift toward custom chips optimized for specific workloads rather than relying exclusively on general-purpose GPUs. Marvell's technology portfolio, including high-speed SerDes, multi-die XPU designs, CXL memory controllers, and recently acquired photonic memory technology from Celestial AI, likely influenced Google's decision to diversify beyond its existing supplier Broadcom.

Bigger picture

Google's decision to add Marvell as a custom silicon partner reflects a broader industry trend toward supplier diversification among major cloud providers. While Broadcom has been Google's primary partner for TPU development through a contract extending to 2031, Google is now multi-sourcing components from Marvell, MediaTek, and potentially AMD. This shift does not necessarily represent a zero-sum game, as the custom AI chip market is expanding rapidly, but it does introduce competitive pressure on incumbents. Broadcom shares fell over 4% following the announcement, indicating investor concern about potential market share erosion. However, the overall AI infrastructure market is growing fast enough that multiple suppliers can potentially succeed simultaneously. Amazon, Microsoft, and other hyperscalers are pursuing similar custom silicon strategies to reduce dependence on single vendors and optimize costs for their specific AI workloads. The success of these programs will likely influence how much control hyperscalers can exert over their AI infrastructure costs and performance in the coming years.

What to watch

Investors should monitor Marvell's quarterly revenue from Google to track how quickly the warrant tranches vest and whether the partnership generates meaningful financial impact. Marvell's upcoming earnings report on August 27 may provide more detail on revenue targets and timelines for Google-related business. Additionally, watch for any disclosure about which specific technologies Google adopts first, particularly whether Marvell's photonic memory solutions or other advanced IP play a major role. For broader context, track Broadcom's September 2 earnings call for management commentary on customer concentration and whether Google's diversification affects Broadcom's TPU revenue trajectory. Any evidence that Marvell is capturing increasingly larger portions of Google's custom silicon spending would have significant implications for competitive dynamics in the custom AI chip market. Finally, observe whether other hyperscalers follow Google's lead in adding Marvell as a supplier, which would further validate the company's custom silicon strategy.