In a significant move to bolster artificial intelligence infrastructure, Nvidia has announced a $2 billion strategic investment in Marvell Technology. The deal, revealed on March 31, 2026, expands the existing collaboration between the two semiconductor firms, with a particular focus on custom AI processors and advanced networking solutions. Marvell shares experienced a sharp increase in premarket trading following the announcement.
Strategic Expansion and Financial Outlook
This investment strengthens a partnership initially outlined in May 2025, which centered on integrating Marvell's custom cloud silicon with Nvidia's NVLink Fusion technology. The enhanced alliance aims to provide customers with more flexible, semi-custom AI system configurations. Marvell's leadership highlighted the growing criticality of high-speed connectivity as AI models continue to expand in size and complexity.
Financially, Marvell is riding a wave of strong performance. The company reported record fiscal 2026 revenue of $8.195 billion, representing a substantial 42% year-over-year increase. Fourth-quarter data-center revenue alone jumped 21%. Looking ahead, management has provided guidance for first-quarter fiscal 2027 revenue of approximately $2.4 billion, plus or minus 5%. More notably, the company projects that full-year fiscal 2027 revenue could approach $11 billion.
Focus on the AI Infrastructure "Plumbing"
While much market attention remains on graphics processing units (GPUs), this partnership underscores the increasing importance of the underlying infrastructure that powers AI clusters. The collaboration will see Marvell take a leading role in developing custom processors and the scale-up networking—the high-speed interconnects that link massive AI systems together. Nvidia will contribute its Vera central processing units, ConnectX network adapters, BlueField data processing units, and Spectrum-X switches.
A key technological frontier for the partnership is silicon photonics, where both companies plan to co-develop solutions. This technology uses light to transmit data on chips, promising significantly faster and more energy-efficient data movement within and between servers—a critical bottleneck for next-generation AI workloads.
Competitive Landscape and Market Context
Marvell is not alone in targeting the lucrative custom AI chip market for hyperscale cloud providers. Broadcom is also aggressively pursuing this segment, with Reuters reporting earlier this month that Broadcom forecasts over $10 billion in AI-chip sales for the coming year. Both Marvell and Broadcom are working directly with cloud service providers on bespoke processor designs, which serve as alternatives or supplements to Nvidia's standardized AI chip offerings.
Marvell's recent technology launches demonstrate its momentum. The company introduced what it claims is the industry's first 260-lane PCIe 6.0 switch, engineered specifically for scaling AI infrastructure. Demand is also rising for its optical digital signal processors, the chips that drive high-speed optical links, and for its application-specific integrated circuits (ASICs) built for dedicated computing tasks.
CEO Matt Murphy noted an "unprecedented pace of bookings," with fiscal 2026 design wins reaching an all-time high. President and COO Chris Koopmans emphasized that hyperscaler spending "is still growing massively." Analyst Kinngai Chan of Summit Insights suggested the improved financial forecast provided a sense of relief for investors, as Marvell had previously lagged behind some competitors in capturing the AI investment wave.
Risks and Future Execution
Despite the strengthened relationship with Nvidia and a bullish outlook, Marvell faces execution risks. The central challenge is converting its robust pipeline of design wins into substantial shipment volumes across its custom silicon, optics, and switching portfolios. If hyperscale customers slow their data center construction or CapEx spending, defending the ambitious 2027 and 2028 revenue targets would become significantly more difficult.
The partnership signifies a strategic shift in the AI hardware race, moving beyond the core processor to optimize the entire data pathway. As Nvidia CEO Jensen Huang declared, "The inference inflection has arrived," pointing to the growing need for efficient systems to run trained AI models. Marvell's strategic positioning in networking and custom chips places it at the heart of building the essential, yet less visible, plumbing that makes large-scale AI possible.
