Google Deepens AI Chip Push With $12.2 Billion Stake Deal in Marvell

Introduction
The race to build faster, cheaper AI infrastructure is no longer just about software — it is about who controls the silicon underneath it. This week, Google took another major step in that direction.

The Deal
Google has moved to deepen its position in custom AI chip design through a deal that could give it a stake worth roughly $12.2 billion in Marvell Technology, a company known for its custom ASIC and networking chip work. The move builds on Google's existing Tensor Processing Unit (TPU) program, which already powers a large share of its internal AI workloads.

Why It Matters
Hyperscalers like Google, Amazon and Microsoft have spent the past two years trying to reduce their dependence on a single chip supplier for AI training and inference. Owning more of the chip supply chain gives Google negotiating leverage, better margins, and tighter control over performance-per-watt — a growing concern as AI data centres strain power grids worldwide.

Industry Ripple Effects
Analysts say the deal reflects a broader shift: AI competition is moving from the software layer into capital markets, chip design, and energy infrastructure all at once. Marvell benefits from a guaranteed anchor customer, while Google secures more predictable chip supply for its Gemini model family and cloud customers.

What Comes Next
Expect more such vertical deals through the rest of 2026 as every major AI lab tries to lock down compute capacity years in advance. For businesses relying on cloud AI services, this consolidation could eventually translate into more stable pricing — or new dependency risks, depending on how the chip supply chain shakes out.