Nvidia Invests $2 Billion in Marvell to Build the Light-Speed AI Infrastructure

Two AI chips connected by beams of light on a dark circuit board

Nvidia just made a $2 billion bet that the future of AI infrastructure runs on light, not copper. The company is taking a strategic stake in Marvell Technology and partnering to develop silicon photonics — technology that uses light to move data between chips at speeds copper wiring simply can’t match.

What the Deal Includes

The partnership connects Marvell to Nvidia’s AI factory ecosystem through NVLink Fusion, the technology that allows third-party silicon to plug directly into Nvidia’s proprietary interconnect fabric. This means Marvell’s custom AI chips and networking equipment can integrate natively into Nvidia’s platform.

The companies will also collaborate on leveraging telecommunications networks for AI computing — a nod toward the distributed AI infrastructure future that companies like Starcloud are also betting on.

Why Silicon Photonics Matters

As AI models grow larger and data center clusters scale to tens of thousands of GPUs, the bottleneck increasingly isn’t the chips themselves — it’s the connections between them. Silicon photonics replaces copper interconnects with light-based data transmission, dramatically increasing bandwidth while reducing power consumption and latency.

Market Reaction

Marvell shares surged 9-11% on the news. The company now derives over 75% of its revenue from data centers and expects to hit $11 billion in revenue next fiscal year. Nvidia shares rose 3%.

The Bottom Line

Nvidia isn’t just building chips anymore — it’s building an ecosystem. By investing $2 billion in one of its biggest competitors, Nvidia is ensuring that even the alternatives to its GPUs still connect through its platform. It’s the ultimate platform play: own the interconnect, own the ecosystem.