Santa Clara, California, March 31, 2026, 14:02 PDT
Nvidia on Tuesday announced a $2 billion investment in Marvell Technology, expanding a tie-up focused on integrating Marvell’s custom AI chips and networking hardware with Nvidia’s systems. Nvidia stock jumped 5.6%. Marvell soared, last up 12.8%.
This shift is significant as the AI infrastructure boom heads in a new direction. While Big Tech’s outlays on AI infrastructure are still forecast to top $630 billion this year, more companies are looking for custom-made processors designed for their own models and data centers, not just Nvidia’s off-the-shelf chips. As a result, Nvidia is leaning into openness as a strategy, aiming to maintain its influence inside those data centers—even if it isn’t supplying every chip.
The news comes just after Nvidia shares stumbled. According to Reuters, Nvidia’s forward price-to-earnings ratio dropped to levels not seen since early 2019—suggesting investors are starting to question if the AI-fueled spending spree can really keep delivering.
Marvell is set to provide custom AI chips and networking gear compatible with Nvidia’s NVLink Fusion, a system designed to connect processors in a unified AI network. Both companies also plan to collaborate on silicon photonics—tech that transmits data between chips using light instead of electricity, a potential fix for bandwidth and power bottlenecks in sprawling server farms.
“The inference inflection has arrived,” Huang said, referring to the moment a trained AI model processes user prompts. Over at Marvell, CEO Matt Murphy described the partnership as an answer to surging demand for speedy connections and optical links, now that customers are scaling up their AI systems. Marvell Technology, Inc.
eMarketer analyst Jacob Bourne flagged the agreement as a way for Nvidia to tap into Marvell’s “semi-custom silicon and advanced optical interconnect capabilities.” He noted that investors probably view this as a move to ease barriers, making it simpler for third-party chips to run inside data centers dominated by Nvidia hardware. Reuters
Marvell’s growth narrative, already closely linked to AI data centers, gets added momentum from the deal. The company posted record fiscal 2026 revenue earlier this month at $8.195 billion, marking a 42% jump over last year. Looking ahead, Marvell said fiscal 2028 revenue could climb to nearly $15 billion.
The landscape is getting more crowded. Earlier this month, Reuters flagged that Nvidia is under increasing pressure in inference as CPUs and custom chips move in. Back in March, Broadcom touted custom silicon as a cheaper substitute for Nvidia and projected AI-chip sales topping $100 billion next year. In February, eMarketer noted Meta’s push toward AMD, with hyperscalers stepping up in-house chip designs.
Still, there’s no guarantee of a steady reward. Nvidia’s decision to keep plowing funds into the AI sector—rather than upping returns for shareholders—has already raised eyebrows among investors. This week, S&P Global Visible Alpha cautioned that if companies rein in AI capex, especially with rising energy bills, stocks could see a “really meaningful correction.” Marvell, for its part, highlighted exposure to partnership risks, shifting demand, and broad market pressures in its latest statement. Reuters
The Marvell agreement on Tuesday marks another move by Nvidia to leverage its financial heft, snapping up more of the infrastructure fueling AI. Earlier this month, Reuters revealed Nvidia was lining up $2 billion for stakes in optical parts suppliers Lumentum and Coherent, signaling the company sees high-speed chip connections as nearly as critical as the processors.
