Marvell Technology: The Undervalued Pillar of AI Infrastructure's Next Wave

The race to build the next generation of artificial intelligence infrastructure is heating up, and Marvell Technology (MRVL) is positioning itself as a quiet but pivotal player. While giants like NVIDIA and Amazon Web Services (AWS) dominate headlines, Marvell's strategic partnerships, technological breakthroughs, and financial acumen are quietly cementing its role as an undervalued leader in the custom silicon market. For investors seeking exposure to AI's exponential growth without the froth of overhyped stocks, MRVL deserves serious consideration.

Strategic Partnerships: The Backbone of AI Infrastructure

Marvell's partnerships with AWS and NVIDIA are not merely transactional—they are foundational to reshaping how AI is built and deployed. In December 2024, the company extended its multiyear agreement with AWS, a five-year deal focused on co-designing semiconductor solutions for cloud-based AI workloads. This collaboration leverages AWS's cloud infrastructure and electronic design automation tools to accelerate Marvell's silicon design processes, enabling parallelized engineering tasks and reducing time-to-market.

The payoff is clear: AWS's AI chip demand drove a 76% year-over-year revenue surge in Marvell's data center segment to $1.44 billion in Q1 2025.

Meanwhile, a May 2025 partnership with NVIDIA integrated Marvell's custom cloud platform silicon with NVIDIA's NVLink Fusion technology. This union delivers hyperscalers like AWS and Microsoft the ability to deploy custom AI accelerators (XPUs) within NVIDIA's rack-scale systems, achieving 1.8 terabytes per second of bidirectional bandwidth—critical for training large-scale AI models.

Technological Edge: 2nm SRAM and the Future of Memory Efficiency

At the heart of Marvell's differentiation is its first-of-its-kind 2nm custom Static Random Access Memory (SRAM), unveiled in June 2025. This innovation reduces die area by 15% and cuts standby power by 66% compared to standard SRAM, while offering up to 6 gigabits of high-speed memory. For AI chips, which rely on vast memory to handle complex computations, this breakthrough allows for either more compute cores or expanded memory integration without increasing physical footprint.

Marvell's Ultra Accelerator Link (UALink), an open-standard interconnect solution, further solidifies its edge. Designed for rack-scale AI systems, UALink reduces latency and improves compute utilization, enabling hyperscalers to build faster, more efficient data centers.

Financial Leverage: Growth at a Discount

Despite its rapid growth and strategic bets, Marvell trades at a P/S ratio of 10.55x, a stark contrast to peers like NVIDIA (23x) and AMD (16x). This valuation discount appears unjustified given its robust financials:
- Gross margins of 59-60% in Q2 2025, reflecting high-margin custom silicon sales.
- A $1.5 billion buyback program signaling confidence in its stock's undervaluation.
- $2 billion annual R&D spending focused on AI-first architectures and 3nm/2nm technologies.

The company's pipeline is equally promising. Microsoft's Maia project—a collaboration with Marvell and others—will begin contributing revenue from 2026, while AWS's adoption of 3nm-based Trainium3 chips (designed with Marvell) is expected to cut energy use by 40%.

Risks and Mitigants

No investment is risk-free. Marvell's reliance on AWS (76% of data center revenue) and competition from foundries like TSMC or in-house chip teams pose challenges. However, its multiyear contracts and deep integration into AWS's silicon supply chain reduce execution risk. Moreover, its open-standard UALink and custom SRAM designs create barriers to competition, as rivals struggle to replicate Marvell's holistic approach to AI infrastructure.

Investment Thesis: Buy MRVL Before the Market Catches On

Marvell's combination of strategic partnerships, proprietary technology, and undervalued stock makes it a compelling buy. The company is not just a supplier but a co-architect of AI infrastructure—a role that will only grow in value as cloud providers and enterprises invest billions in AI training and inference.

With a June 2025 investor event likely to highlight new products and partnerships, and a buyback program fueling share price support, now is the time to position for MRVL's next leg up. For investors seeking a leveraged play on AI without the volatility of pure-play chipmakers, Marvell's undervalued P/S ratio and robust fundamentals make it a standout opportunity.

Recommendation: Buy MRVL. Set a price target of $45 (20% upside from current levels) based on potential P/S expansion to 14x, reflecting its critical role in AI infrastructure's next wave.

Josh Nathan-Kazis is a technology analyst specializing in semiconductor and AI infrastructure trends.