Nvidia Invests $2B in Marvell for AI Infrastructure Expansion

NVDA took a $2 billion equity stake in MRVL, announced March 31, 2026, in a deal centered on NVLink Fusion — a new framework that opens Nvidia's previously proprietary NVLink interconnect to third-party silicon. The deal sent MRVL surging 12.8% to $99.05 on nearly 50 million shares traded, while NVDA gained 5.6% to close at $174.40, with the broader S&P 500 rising 2.9% that same session. Wedbush attributed the outsized market reaction to recognition that NVLink Fusion could become the dominant rack-scale AI interconnect standard.

NVLink Fusion represents a strategic shift: until now, NVLink was exclusive to Nvidia hardware. By enabling Marvell's custom XPUs and ASICs to plug directly into the NVLink fabric, Nvidia expands the reach of its ecosystem without ceding control of the interconnect layer. The partnership spans three pillars: AI compute via custom XPUs, silicon photonics and optical interconnects, and AI-RAN — converting 5G/6G telecom networks into AI compute infrastructure using Nvidia's Aerial platform.

The investment continues Nvidia's recent pattern of $2 billion strategic equity stakes — it made similar investments in CoreWeave, Nebius, and Coherent in the months prior — building a network of hardware partners tied to the NVLink ecosystem. The key competitive risk is the open UALink interconnect standard, backed by AMD, Intel, and Broadcom, which directly challenges NVLink Fusion's bid to become the AI infrastructure standard. How quickly Marvell ships NVLink-compatible silicon and how broadly hyperscalers adopt the standard will determine whether this investment yields returns beyond the initial stock-price pop.