Nvidia just made its most revealing strategic move in months - and it is not about a new GPU.
The company announced a $2 billion investment in Marvell Technology and a partnership that integrates Marvell's custom AI chips directly into Nvidia's NVLink Fusion platform. Marvell shares jumped 13% on the news. Jensen Huang told CNBC he had been "dying to say" that Marvell is a marvelous investment.
What the Deal Actually Does
The partnership works in both directions. Marvell contributes custom accelerator chips called XPUs - application-specific chips designed for particular AI workloads - along with NVLink Fusion-compatible networking. Nvidia supplies its Vera CPUs, ConnectX network interface cards, BlueField data processing units, and Spectrum-X switches. Together, the two companies are building a rack-scale AI infrastructure platform that gives customers more flexibility in how they deploy AI compute.
Both companies will also collaborate on silicon photonics - technology that uses light rather than copper wiring to move data between chips at higher speeds and lower power. Marvell recently closed a $3.25 billion acquisition of Celestial AI, which brought photonic fabric technology directly into its portfolio. Nvidia has made similar photonics bets with Lumentum and Coherent in recent weeks, signaling that optical interconnects are becoming the next critical bottleneck in large-scale AI infrastructure.
Why Nvidia Is Doing This
The subtext of this investment is defensive. Nvidia's GPU dominance is real, but an increasing number of hyperscalers - Google, Amazon, Microsoft, Meta - are designing their own custom AI chips to reduce dependence on Nvidia's expensive GPUs. By opening NVLink Fusion to Marvell's custom XPUs and investing $2 billion to formalize the relationship, Nvidia is repositioning itself from GPU supplier to ecosystem platform.
Huang framed it directly: the partnership expands the total addressable market for both companies. Customers who want custom silicon can now build on top of Nvidia's networking, storage, and CPU infrastructure rather than leaving the ecosystem entirely. It is a strategy that trades some GPU revenue for ecosystem lock-in at the infrastructure layer.
Marvell reported fiscal 2026 revenue of $8.2 billion, up 42% year over year, with data center revenue now representing the majority of its business. Nvidia is not investing in an unproven bet. It is cementing a relationship with a chip company that is already winning in AI infrastructure and integrating it more tightly into a platform that Nvidia controls. For business leaders evaluating AI chip strategy, the message is clear: the competitive landscape is broadening fast, and Nvidia's response is to make its ecosystem the unavoidable foundation layer regardless of which chip sits on top.
