Google Expands Marvell AI Chip Deal With $12 Billion Warrant

Google has expanded its partnership with Marvell Technology to develop custom AI silicon, a move that includes a warrant allowing Google to purchase up to 58.97 million Marvell shares at $206.58 each [2]. The agreement, signed July 29, ties the potential $12.18 billion equity stake to future purchases of custom AI infrastructure, signaling a long-term commitment to internal chip development as cloud providers race to reduce reliance on third-party hardware [2].

Infrastructure and Supply Chain Strategy

The expanded agreement covers a broad range of custom semiconductor products designed to integrate with Google’s Tensor Processing Unit (TPU) ecosystem [2]. Specifically, Marvell will supply AI inference accelerators, storage controllers, and network and memory interface controllers [2]. By incorporating near-memory computing—which places processing closer to memory to minimize data movement—Google aims to optimize the performance of its AI Hypercomputer architecture [2].

While the deal marks a significant deepening of ties, Google continues to maintain a multi-supplier strategy. Broadcom remains a key partner, with a separate long-term agreement to develop and supply custom TPUs and networking components for future processor generations [2]. This dual-track approach allows Google to manage supply chain risks while deploying its own hardware alongside Nvidia GPUs, which remain a staple of the Google Cloud accelerator portfolio [2].

The Competitive Landscape

Google’s move mirrors a broader industry trend among hyperscalers seeking to control their own AI infrastructure costs. Amazon, Microsoft, and Meta are all aggressively developing proprietary accelerators to handle the intensive demands of generative AI and inference workloads [2]. Amazon has linked its Trainium chips to potential multi-billion dollar reductions in capital expenditure, while Microsoft’s Maia 200 and Meta’s MTIA family are similarly focused on optimizing performance-per-dollar for internal services [2].

The financial structure of the Google-Marvell deal also reflects a growing trend of using equity warrants to align incentives between cloud providers and chip designers. Similar to the agreement between AMD and OpenAI, which ties share vesting to the deployment of specific GPU volumes, Google’s warrant vests in 240 tranches linked to $500 million increments of custom-product revenue [2].

What to watch

• Vesting Milestones: Monitor the first quarterly installments of the 1.36 million shares set to vest within the first year of the agreement [2].
• Fiscal 2027 Ramp-up: Watch for the start of the primary purchase-linked vesting period, which begins in Marvell’s fiscal third quarter of 2027 and extends through 2033 [2].
• TPU Integration: Observe which future generations of Google’s TPU systems incorporate the new Marvell-designed silicon, as current filings do not specify individual program assignments [2].

The success of this partnership will ultimately be measured by whether these custom components can deliver the efficiency gains necessary to sustain Google’s AI infrastructure at scale. With the agreement extending through 2033, the deal underscores a shift toward deep, multi-year vertical integration between cloud giants and their silicon suppliers.