NVIDIA Takes $2B Stake in Marvell to Expand NVLink Fusion AI Platform

The Marvell Deal

NVIDIA invested $2 billion in Marvell Technology and announced a strategic partnership built on NVLink Fusion, NVIDIA's rack-scale platform for semi-custom AI infrastructure. Under the deal, Marvell will supply custom XPUs and NVLink Fusion-compatible scale-up networking, while NVIDIA contributes its Vera CPU, ConnectX NICs, BlueField DPUs, NVLink interconnect, and Spectrum-X switches.

The two companies will also collaborate on silicon photonics — the optical interconnect technology that's becoming critical as AI cluster bandwidth demands outstrip copper. They'll additionally partner on NVIDIA Aerial AI-RAN for 5G/6G, aiming to turn telecom infrastructure into AI compute fabric.

Jensen Huang framed the move bluntly: "The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories." Marvell CEO Matt Murphy positioned his company's role as the analog, optical DSP, and custom silicon layer that makes those factories scalable.

Marvell shares jumped roughly 13% on the announcement. NVDA itself gained over 3% intraday, which combined with a rebound from the prior session's 2.2% selloff accounts for the 6.4% move tracked on the Hyperliquid perp.

The $2 Billion Playbook

This isn't a one-off. NVIDIA has deployed the same $2 billion ticket across at least five deals in recent months:

• **Nebius** — AI cloud infrastructure (stock popped 16%)
• Coherent — optical networking components
• Lumentum — optical interconnect (included purchase commitments)
• CoreWeave — GPU cloud compute
• Marvell — custom XPUs and silicon photonics

The strategy is clear. NVIDIA is using its financial position to lock up critical supply chain partners across every layer of the AI data center — compute, networking, optics, cloud, and now custom silicon. Each investment comes with a technology integration agreement that deepens the moat around NVIDIA's platform.

The NVLink Fusion angle is particularly interesting. By letting partners like Marvell plug custom XPUs into the NVLink fabric, NVIDIA can position itself as both vertically integrated and horizontally open — offering the full stack while also welcoming heterogeneous components, as long as they run on NVIDIA's interconnect.

What This Changes

NVDA came into March 31 trading at a forward P/E that matched the S&P 500 for the first time in a decade, down roughly 2% on the year despite posting 73% revenue growth in fiscal Q4. The valuation compression reflected real anxiety about whether AI capex could sustain its current pace.

The Marvell deal pushes back on that narrative in two ways. First, it signals NVIDIA has enough conviction in inference-driven demand to keep deploying billions into ecosystem buildout. Second, it expands the addressable market for NVLink Fusion into custom silicon and telecom — sectors that weren't part of the original AI capex story.

For the perp market, the 19-hour move captures both the prior-day dip recovery and the Marvell catalyst. Wall Street consensus sits at $273 with a Strong Buy rating, implying roughly 57% upside from $174. The next major catalyst is Q1 FY2027 earnings expected in late May, where the market will test whether NVIDIA can deliver on its $78 billion revenue guide — a 14.5% sequential step-up from the record Q4.

Trading on Hyperliquid

Trade NVDA on Hyperliquid with up to 20x leverage.