Marvell Technology's stock surged nearly 30% in pre-market trading on Tuesday, fueled by comments from NVIDIA CEO Jensen Huang at the 2026 Computex event in Taipei.

The core catalyst was Huang's public discussion with Marvell's leadership, where he suggested the company has the potential to become the next publicly-traded entity with a market capitalization exceeding $1 trillion.

Huang emphasized Marvell's crucial role within the interconnect infrastructure systems for large-scale AI data centers, a powerful endorsement from the leading force in AI computing that significantly bolstered market confidence in Marvell's long-term growth trajectory.

Marvell Technology (NASDAQ: MRVL)

In a major keynote speech titled "The Future of AI Scaling Depends on Connectivity," Marvell CEO Matt Murphy outlined a critical industry shift.

He argued that as demand for AI model compute resources and the scale of AI infrastructure clusters grow exponentially, improvements in computational performance are beginning to outpace advancements in data movement and high-speed data center interconnect capabilities.

This dynamic positions connectivity, rather than single-node computing power, as the primary bottleneck for the next phase of AI scaling.

Murphy positioned Marvell's portfolio in optical interconnect, ultra-high-speed DSP, and silicon photonics at the very heart of this new infrastructure imperative.

The subsequent dialogue between Huang and Murphy served to deeply elaborate on this industry-wide upgrade towards a "connectivity-centric" bottleneck within the AI compute supply chain.

It also revealed the fundamental logic behind Huang's strong bullish stance on Marvell and his decision to pursue deep collaboration and strategic investment.

This exchange was more than a technical discussion; it represented a high-level declaration of the dynamics and trends shaping the global AI compute infrastructure landscape.

The Broader AI Revenue Picture

The robust revenue trajectory recently outlined by Anthropic, including an expected doubling of Q2 revenue and its first operating profit, signals a shift in the AI industry narrative from "cash-burning" to "cash-flow generation."

Anthropic's staggering monthly ARR increase of $110 billion highlights the unprecedented monetization potential of frontier AI applications, which are beginning to transform enterprise programming, agent workflows, cybersecurity, and data analytics into high-value token revenue.

The exploding demand for tools like Claude is driving Anthropic toward profitability, with its compute cost per dollar of revenue declining, demonstrating improving scale and inference efficiency.

This context underpins Wall Street's view, exemplified by Morgan Stanley, that the AI compute arms race is entering a phase of systemic expansion.

The firm has significantly raised its capital expenditure forecasts for major U.S. tech giants, predicting that nearly $3 trillion in AI-related infrastructure investment will flow through the global economy by 2028, with over 80% of that spending still ahead.

This outlook underscores that the AI investment thesis is evolving from a singular focus on GPU procurement to a "full-stack compute system" driven by AI agents.

Supply chain bottlenecks are now expanding to encompass the entire AI data center delivery chain, including power equipment, liquid cooling, CPUs, memory, and critically, optical communication and interconnect systems, which are among the current major constraints.

Jensen Huang's Endorsement: High-Speed Interconnect is the Future

Huang emphasized that NVIDIA's strategic $2 billion investment in Marvell and their collaboration is not one-sided but mutually beneficial within the ecosystem.

Under frameworks like NVLink Fusion, Marvell can provide hyperscalers with custom optical interconnect and ASIC solutions, while NVIDIA ensures data center infrastructure balances generality with high performance through its widespread GPU and system platforms.

This synergy of hardware, data center optical interconnect, and ecosystem integration grants Marvell long-cycle growth potential in future AI factories, AI-RAN, and edge AI deployments, supporting its case as a potential trillion-dollar core enterprise in AI infrastructure.

Huang explicitly stated that with the global rise of agentic AI and massively distributed training/inference configurations, performance limits for computation and memory are being surpassed.

The next core constraint is not raw compute or memory capacity, but the efficient movement of data between nodes.

As AI workloads are split across hundreds of thousands of coordinated processing units, only ultra-high-speed, low-latency, power-efficient intra-data center and inter-data center (DCI) interconnects can prevent communication bottlenecks from negating gains in compute and memory.

Huang repeatedly stressed that Marvell's solutions are critically positioned within this interconnect architecture, forming the first layer of his bullish thesis.

While traditional data center networks rely on copper, its physical limitations become apparent at high bandwidths and long distances.

As AI workloads scale across racks and data centers, demand for intra-data center optical interconnects, DCI links, and cutting-edge silicon photonics technology is poised for accelerated growth.

Huang's strategy of "use copper where you can, use light where you must" highlights that at the critical point of breaking traditional bandwidth limits, Marvell is one of the few suppliers capable of providing high-speed optical interconnect, optical DSP, and potential silicon photonics solutions.

Strategic Alignment and Ecosystem Lock-In

Huang's prediction of a trillion-dollar valuation for Marvell is based on a two-layer logic of industrial and ecological synergy within the AI compute chain.

Through the NVLink Fusion platform, NVIDIA has established not just technical compatibility but deep collaboration, extending its AI infrastructure strategy from GPUs alone to rack-level, system-level, and cross-vendor custom ASIC integration.

This partnership allows Marvell to occupy an indispensable position in future AI data center builds, potentially enabling it to transcend traditional chip valuation frameworks.

NVIDIA's $2 billion investment represents more than a financial vote of confidence; it signifies ecosystem lock-in.

By binding capital and technology, NVIDIA ensures its NVLink interconnect standards and AI factory architectures are co-optimized with Marvell's interconnect solutions, reducing the risk of ecosystem fragmentation from customers pursuing independent custom paths.

This "hardware-software compatibility" framework makes it more likely for customers to opt for unified deployment using the NVLink/Marvell network combination when building large-scale AI infrastructure, thereby securing future market share.

The Optical Interconnect Engine Driving Marvell's Rally

Marvell's stock has soared approximately 160% year-to-date, significantly outperforming both the S&P 500 and the tech-heavy Nasdaq 100 Index.

The core engine for this repricing and fundamental reassessment is the resonance of two hard logic themes: "AI ASIC/XPU + Optical Technology," rather than a simple chip demand narrative.

Wall Street is increasingly viewing Marvell not just as a generic data center AI chip company, but as a "dual-wheel asset" that captures both the super-cycle of hyperscaler XPU/ASIC development and the massive upgrade wave in optical interconnect and silicon photonics driven by exploding AI cluster bandwidth demands.

AI ASIC is the first core pillar supporting Marvell's valuation re-rating and fundamental growth, with strong recent earnings from both Marvell and competitor Broadcom quickly providing "earnings-level evidence" to confirm the unprecedented growth logic in this segment.

Optical interconnect forms the second core pillar of the bullish thesis on Marvell's prospects and a key reason for NVIDIA's investment.

For connections beyond approximately 10 meters, copper interconnects cannot meet the bandwidth and distance requirements of large AI data centers, necessitating a shift to optical systems where Marvell's expertise in optical DSP and interconnect products becomes critical.

The company has launched high-bandwidth networking chips like the Teralynx T100 102.4Tbps switch chip, designed to meet the interconnect needs of AI clusters across nodes, racks, and data centers where copper falls short.

Furthermore, Marvell's technology accumulation in optical interconnect and silicon photonics, bolstered by strategic acquisitions, enhances its competitiveness in the high-end optical interconnect and custom ASIC markets—components that are vital infrastructure in the era of scaled AI training and inference.

Looking further ahead, Marvell's leading position in silicon photonics is another key reason for Wall Street's optimism.

Silicon photonics is not a parallel alternative to optical interconnect but a key enabling technology that pushes optical transmission from between cabinets and racks further into the rack, system, and even within the package or chip itself.

This strategic direction explains Marvell's significant acquisition of companies like Celestial AI.

Marvell's technological积累 and product portfolio in high-speed optical DSP, optical interconnect modules, and custom ASIC/XPU place it in some of the most scarce and defensible parts of the AI compute infrastructure stack.

In an environment of rapidly increasing AI inference and training demands, these components are harder to substitute than individual compute chips.

This positions Marvell not merely as an "AI chip stock," but as an architectural core for optical interconnect, DCI, and broader data movement infrastructure, promising longer-term and more stable revenue and profit expansion potential.