- Nvidia invested $2 billion in Marvell Technology on March 31, 2026, continuing a pattern of $2 billion strategic bets on infrastructure companies including CoreWeave, Synopsys, Coherent, Lumentum, and Nebius Group.
- The partnership centers on NVLink Fusion, a rack-scale platform that allows Marvell’s custom AI accelerators and networking silicon to integrate directly into Nvidia’s high-speed interconnect fabric.
- Marvell shares surged as much as 13% on the news, as the deal validates Marvell’s position in the custom ASIC and silicon photonics markets.
- The companies will also collaborate on silicon photonics — using light instead of copper to move data — and on 5G/6G AI-RAN telecommunications infrastructure.
What Happened
Nvidia announced a $2 billion investment in Marvell Technology on March 31, 2026, opening its NVLink ecosystem to Marvell’s custom silicon and networking products. The deal gives Nvidia a significant equity stake in Marvell while establishing a technical partnership around NVLink Fusion, a rack-scale platform designed to let customers build AI infrastructure using heterogeneous combinations of Nvidia GPUs and Marvell’s custom processors.
Jensen Huang, Nvidia’s CEO, framed the investment around inference demand: “The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories.” Marvell CEO Matt Murphy said the partnership connects “Marvell’s leadership in high-performance analog, optical DSP, silicon photonics and custom silicon to Nvidia’s expanding AI ecosystem through NVLink Fusion.”
Why It Matters
The investment signals that Nvidia views the AI infrastructure market as too large and too varied to serve with GPUs alone. NVLink Fusion allows Marvell’s custom XPUs — specialized processors designed for specific workloads — to communicate with Nvidia GPUs over Nvidia’s proprietary high-speed interconnect. This means cloud providers and enterprises can mix Nvidia GPUs with Marvell’s custom accelerators in the same rack without sacrificing interconnect performance.
For Marvell, the deal is a major endorsement. The company has built a growing business designing custom AI chips (ASICs) for hyperscale customers, competing with Broadcom in that market. Nvidia’s investment and NVLink integration effectively bring Marvell inside Nvidia’s ecosystem rather than positioning it as a competitor.
Technical Details
NVLink Fusion enables full compatibility between Nvidia GPU, LPU (Language Processing Unit), networking, and storage platforms and Marvell’s custom silicon. The platform allows Marvell’s custom XPUs to use NVLink’s high-bandwidth interconnect for scale-up networking, which is critical for large AI training and inference clusters where data movement between processors is a primary bottleneck.
The silicon photonics collaboration is a second major element. Both companies will work on advanced optical interconnect solutions that use light rather than traditional copper wiring to move data between chips and across racks. Optical interconnects offer higher bandwidth and lower power consumption at data center scale, and are considered essential for next-generation AI clusters.
A third area of collaboration involves 5G and 6G AI-RAN (Radio Access Network) infrastructure through Nvidia’s Aerial platform, extending the partnership beyond data centers into telecommunications.
Who’s Affected
Cloud providers and hyperscalers are the immediate beneficiaries. Companies that already purchase custom ASICs from Marvell can now integrate those chips into Nvidia-based racks using NVLink Fusion, rather than maintaining separate infrastructure stacks. This flexibility could reduce capital expenditure on duplicate networking equipment.
Marvell shareholders saw immediate returns — the stock jumped as much as 13% in early trading. The deal fits a pattern: Nvidia has made a series of $2 billion investments in recent months, taking stakes in Synopsys, CoreWeave, Coherent, Lumentum, and most recently Nebius Group. Each investment targets a different segment of the AI infrastructure supply chain.
What’s Next
The partnership’s practical impact depends on how quickly NVLink Fusion-compatible products ship. Marvell will need to deliver custom XPUs and networking silicon that meet NVLink’s performance specifications, and customers will need to validate the heterogeneous configurations in production workloads. The silicon photonics work is longer-term — commercial optical interconnect products at the scale Nvidia and Marvell are targeting remain at least a year from broad deployment.
For investors, the deal raises a question about Nvidia’s broader strategy. The company has now deployed at least $10 billion across equity stakes in its supply chain partners. That level of investment creates deep alignment but also concentration risk — Nvidia’s returns are now tied to the execution of multiple companies simultaneously. Whether NVLink Fusion drives meaningful new revenue or functions primarily as an ecosystem lock-in tool will become clearer as the first joint products reach market.
