Nvidia’s (NVDA) $2 billion investment in Marvell Technology (MRVL) sent the latter’s stock soaring nearly 13% in a single day. On the surface, the deal raises a puzzling question. Nvidia’s GPUs dominate the AI training market, yet Marvell is a leader in custom AI chips (ASICs/XPUs), widely seen as the most significant alternative to the GPU-centric path. Why would Nvidia pour billions into a potential competitor?
The answer lies in two key strengths Marvell brings to the table: custom AI chips and data-center networking technology. Far from conflicting with Nvidia’s business, these capabilities are deeply complementary within the AI infrastructure landscape. Alongside the investment, both companies announced an expanded partnership in silicon photonics and AI-focused communications hardware, signaling Nvidia’s intent to embed Marvell more deeply into its AI ecosystem through a combination of capital and technology.
Marvell designs custom chips for hyperscalers, including Microsoft’s Maia series and Amazon’s Trainium and Inferentia processors. Over the past few months, concerns had grown that its largest customers might shift some orders elsewhere. But management put those fears to rest in its latest earnings report: the custom silicon business is expected to double year-over-year in fiscal 2027, with next-generation programs from its top customer already backed by purchase orders covering the entire fiscal year, and growth set to continue into fiscal 2028. Marvell CEO Matt Murphy emphasized that the relationship with its largest customer remains “strong.”
For Nvidia, investing in Marvell isn’t about propping up a direct GPU rival—it’s a recognition that as hyperscalers increasingly develop their own chips, ASICs will become an indispensable part of the AI computing landscape. Rather than sit on the sidelines, Nvidia is choosing to align itself with the leader in that space.
Marvell’s connectivity chips—including high-speed Ethernet PHYs, switches, and DSPs—are widely deployed in data-center network architectures and play a critical role in enabling GPU clusters to function as a cohesive system. The expanded partnership specifically highlights silicon photonics and AI communications hardware, revealing that Nvidia is looking beyond Marvell’s existing networking portfolio and aiming to jointly develop next-generation optical interconnect technology to tackle the bandwidth and power challenges posed by ever-larger AI clusters.
Nvidia has a track record of taking stakes in companies only to exit them later. This time, however, the investment is just one part of a broader relationship. Along with the capital infusion, the two companies are expanding their technical collaboration, forming a strategic alliance built on both capital and R&D. As Nvidia CEO Jensen Huang put it, Marvell is a “marvelous investment”—a comment that reads as both praise and a hint that the two will be working together across a wider technological frontier.
Nvidia’s heavy investment in Marvell may look like a chip giant funding a potential rival, but it is better understood as a strategic alignment of interests. With its custom AI chips and data-center networking technology, Marvell has already secured a critical position in the AI infrastructure value chain. Nvidia, through a mix of capital and collaboration, is bringing this ally closer into its orbit. For investors, the deal offers a clear reminder: the AI computing story extends well beyond GPUs—and Marvell is one of the key authors of its next chapter.
