Marvell Trillion-Dollar Market Cap Path: AI ASICs, Custom Chips, and NVLink Ecosystem Breakdown

June 2, 2026, at the COMPUTEX Taipei Computer Show stage, NVIDIA CEO Huang Renxun suddenly appeared during Marvell CEO Matt Murphy's speech, and addressed the audience with a single sentence: "The next trillion-dollar company, ladies and gentlemen."

The market reaction triggered by this statement was almost immediate. MRVL's stock price surged over 27% intraday, closing up 32.52%, with a single-day market capitalization increase of approximately $62.4 billion, crossing the $250 billion mark for the first time, setting the company's largest single-day gain in history.

But what basis underpins Huang Renxun’s judgment? To grow from a market cap of about $1.92 trillion to $10 trillion, what objective conditions must Marvell cross?

Current Benchmark: Data Foundation at Huang Renxun’s Speech

Before Huang Renxun took the stage to endorse Marvell, the company had just released an earnings report that exceeded market expectations.

For FY2027 Q1 (ending May 2, 2026), Marvell achieved net revenue of $192B, up 28% year-over-year, up 9% quarter-over-quarter, surpassing the company's previous guidance midpoint by $18 million, setting a new quarterly record. Among this, data center revenue contributed $10k, up 27% YoY and 11% QoQ, accounting for 76% of total revenue.

In terms of operating cash flow, this quarter's operating activities generated a record $639 million, up 91.89% YoY. However, under GAAP accounting, net income attributable to shareholders was only $34.5 million, down 80.61% YoY, mainly due to acquisition-related expenses and stock-based compensation and other non-cash items.

Before Huang Renxun’s speech, MRVL’s market cap was about $1.92 trillion. After the surge that day, the market cap broke through approximately $2.5 trillion, and in subsequent trading after June 5, it further climbed to about $2.768 trillion.

Key Calculation: To grow from about $2.5 trillion to $10 trillion, Marvell’s market cap needs to increase roughly 4 times; from the pre-speech $1.92 trillion, about 5.2 times.

Connectivity Logic: Why Does Huang Renxun Believe in Marvell?

To understand the feasibility and conditions for a trillion-dollar valuation, we first need to dissect the core logic behind Huang Renxun’s judgment.

At COMPUTEX 2026, Huang laid out a clear technological framework—AI computing is shifting from a centralized, monolithic architecture to a distributed, decoupled architecture. Large language models and AI agents no longer perform their computational tasks on a single chip or server but are broken down into many loads distributed across thousands of nodes in data centers, processed in parallel.

“When you break a computing problem into many parts and distribute it across the data center, connectivity becomes indispensable,” Huang explicitly stated in his speech, “This is why Matt does such a good job, and why Marvell is so important.”

From a supply chain perspective, Huang believes Marvell’s networking and connectivity chips are “essential” in AI infrastructure. This means Marvell’s market position has a structural, rigid demand characteristic—regardless of who provides the chips for AI compute power, Marvell’s connectivity products serve as the foundational infrastructure.

This judgment is supported by data. Marvell’s data center growth continues to be validated: in FY2027 Q2 guidance, revenue is expected to be about $2.7 billion, up approximately 35% YoY; the company has raised its FY2027 full-year revenue forecast to about $11.5 billion, up about 40%, and increased its FY2028 revenue target to about $16.5 billion, roughly $1.5 billion higher than last quarter’s guidance.

NVLink Fusion: The Technical Anchor on the Path to $1 Trillion

Huang’s judgment is not solely based on the company’s fundamentals but also on the deepening technical integration between NVIDIA and Marvell.

In March 2026, NVIDIA announced a $2 billion strategic investment in Marvell. Their partnership deepened through the NVLink Fusion platform: Marvell will provide customized XPU and vertically scalable network solutions compatible with NVLink Fusion, while NVIDIA contributes its Vera CPU, ConnectX network cards, and BlueField DPU technologies.

The value of NVLink Fusion lies in providing cloud service providers with a semi-custom route. Customers can balance between NVIDIA GPUs and Marvell’s customized XPU when building AI data centers—neither buying entirely from NVIDIA nor fully developing in-house outside NVIDIA’s ecosystem. Huang openly stated during his speech: “You don’t have to buy everything from us—just a part of it.”

This framework lowers the entry barrier for Marvell in the AI chip market. Before the NVIDIA partnership, the main competitor for custom XPU (ASIC) was Broadcom, which dominated with a large customer base and over 70% of the ASIC market share. But NVLink Fusion offers Marvell a differentiated channel into a large customer base: clients don’t have to choose between NVIDIA’s ecosystem and custom chip solutions.

Additionally, Huang’s technical stance provides another incremental value—insights into the transition timeline from copper cables to optical fibers. He pointed out that in the next 5 to 10 years, “where copper cables can be used, use copper; where optical devices are necessary, use optical.” Marvell is one of the few companies capable of providing complete solutions in both copper and optical interconnects, giving it a systemic resilience advantage as AI data center connectivity demands continue to grow.

Custom Chip Business: Conditions for Achieving the $10 Billion Revenue Target

A few days before Huang Renxun’s trillion-dollar valuation projection, Marvell provided a long-term revenue guidance for its custom chip business in its FY2027 Q1 earnings report.

The company forecasts that its custom chip revenue will exceed $10 billion in FY2029. Morningstar analyst William Kerwin noted that this guidance “implies that a single business could generate an incremental $5 billion in revenue in fiscal 2028 and 2029.”

From a TAM (Total Addressable Market) perspective, Marvell CEO Matt Murphy stated during the earnings call, “We have extensive custom collaborations with all major U.S. hyperscale data center operators.” Currently, Marvell’s share in the custom AI chip market still lags behind Broadcom, but this also indicates significant structural growth potential. If the company can increase its market share to around 20% before FY2029, the $10 billion revenue target for custom chips is financially plausible.

However, achieving this goal depends on several conditions:

Condition 1: Diversification and growth of customer base. Currently, AWS (Amazon Web Services) is Marvell’s largest customer, with its Trainium AI ASIC being a key project. Heavy reliance on a single customer makes the company vulnerable to fluctuations in capital expenditure. Murphy has explicitly stated that the company needs to diversify its customer base, aiming to secure orders from additional clients like Microsoft Azure, Google Cloud, and Meta Platforms. Relying solely on AWS’s growth in FY2027 would significantly reduce the feasibility of reaching the $10 billion target.

Condition 2: Support from TSMC’s advanced process nodes and CoWoS capacity. Marvell’s custom chips are manufactured using TSMC’s 5nm and 3nm process nodes. The allocation of TSMC’s CoWoS advanced packaging capacity directly constrains Marvell’s shipment volume. If AI chip demand remains high, CoWoS capacity bottlenecks could become a limiting factor.

Condition 3: Macroeconomic and CAPEX cycle sustainability. Alphabet, Amazon, and other U.S. tech giants are expected to invest over $700 billion in AI infrastructure in 2026 (about $400 billion in 2025). If AI capital expenditure growth slows in 2027–2029, the growth of custom chip revenue will be directly impacted.

Competitive Landscape: Custom XPU vs. General-Purpose GPU Paths

Understanding Marvell’s path to a $1 trillion valuation also requires evaluating its position within the overall AI chip competitive landscape.

In the custom XPU (ASIC) segment, Marvell and Broadcom are the most representative suppliers, but they adopt very different business models. Broadcom, as an industry leader with strong customer stickiness, mainly serves giants like Google and Meta, whose data center capital expenditures are rapidly increasing. Currently, Marvell’s valuation (PE ratio around 28, PEG close to 1) has some advantage over Broadcom, reflecting a market discount for “chaser” status versus “leader” premium.

Meanwhile, the custom XPU race is undergoing a rebalancing with the general-purpose GPU market. TrendForce forecasts a 45% growth in ASIC shipments in 2026, far exceeding the 16% growth expected for GPUs, indicating broader market acceptance of customized designs. However, in terms of performance requirements and ecosystem maturity, GPUs still hold irreplaceable advantages during training phases. Whether Marvell’s custom XPU business can continue to gain share from GPUs depends on client sensitivity to cost and power consumption during inference.

For Marvell, its collaboration with NVIDIA to build the NVLink Fusion ecosystem could be a strategic differentiator from Broadcom. While Broadcom relies more on self-developed solutions for large clients, Marvell’s position as a “certified partner” within the NVIDIA ecosystem offers a potential advantage. If the industry consensus shifts toward “heterogeneous computing” in AI data centers, Marvell’s position could be more resilient. Conversely, if AWS and other hyperscalers’ anti-NVIDIA tendencies intensify, Broadcom’s independence as a pure supplier could also be a competitive advantage.

External Acquisitions: Strategic Significance of Celestial AI and XConn Technologies

In February 2026, Marvell completed two significant acquisitions, further consolidating its strategic layout in AI data center connectivity.

Celestial AI (acquisition cost at least $3.25 billion) brought a breakthrough optical interconnect platform, Photonic Fabric, designed for intra-system and rack-level optical connections. Marvell expects this technology to help capture the critical transition from electrical to optical interconnects, expanding its total addressable market in connectivity.

XConn Technologies (about $540 million acquisition) focuses on PCIe and CXL switch silicon products. It has established partnerships with over 20 clients, with PCIe 5 and CXL 2.0 switches in mass production, and PCIe 6 and CXL 3.1 sampling stages. These are expected to positively impact Marvell’s revenue in the second half of FY2027 and contribute about $100 million in incremental revenue in FY2028.

Growth Path Risks

Despite the current market enthusiasm, it’s essential to understand the factors that could constrain the trillion-dollar valuation.

Risk 1: Uncertainty in technology and industry evolution. Huang Renxun’s view that AI computing is moving toward a “distributed/decoupled” architecture is central to Marvell’s growth. If AI infrastructure evolves toward highly centralized, ultra-large clusters (reducing connectivity needs), the valuation logic would need to be rethought.

Risk 2: Profit quality and cash flow alignment. While operating cash flow hit a record $639 million, GAAP net profit was only $34.5 million, mainly due to non-cash amortizations, stock-based compensation, and acquisition-related expenses. The gross margin structure of custom chip business (similar to Broadcom, where high revenue growth often pressures margins) warrants ongoing monitoring.

Risk 3: Customer concentration and structural vulnerability. Currently, Marvell’s custom chip revenue is highly concentrated with AWS. Any change in AWS’s AI CAPEX cycle could directly impact quarterly performance and the credibility of the $2.42B revenue target.

Risk 4: Excessive market sentiment premium. As of June 5, MRVL’s stock price has surged 272.36% year-to-date. Cantor Fitzgerald analyst CJ Muse noted that Huang Renxun’s statement of a “trillion-dollar market cap” should be viewed as “symbolic and visionary,” not a precise financial forecast, and cautioned about overbought signals in technical indicators.

Conclusion

Marvell’s journey from approximately $2.768 trillion toward $10 trillion is not a predetermined path but depends on a set of dynamic conditions being continuously met.

Fundamentally, the company’s strong quarterly revenue of $1.83B, $639 million in operating cash flow, the $192B+ custom chip guidance, and NVIDIA’s strategic investment form a solid foundation for growth. Strategically, Huang Renxun’s public endorsement at COMPUTEX 2026 positions Marvell as a “connectivity essential supplier” in AI infrastructure, with NVLink Fusion providing a differentiated market stance. Growth-wise, acquisitions of Celestial AI and XConn expand the addressable market in optical interconnects and PCIe/CXL switching.

However, objective constraints such as customer concentration (heavy reliance on AWS), competitive dynamics with Broadcom, the sustainability of AI CAPEX cycles, and potential market sentiment overvaluation remain uncertainties on the road to a trillion-dollar valuation. Additionally, whether NVIDIA will internalize more network connectivity components or invest in photonic ecosystems remains a variable to watch.

In the industry shift from scale expansion to efficiency optimization in AI infrastructure, whether Marvell can navigate technological divergence, evolving competitive landscapes, and macro cycles to ultimately achieve a trillion-dollar market cap hinges on a core variable: can connectivity performance evolve from an “auxiliary component” to a “core bottleneck,” thereby enabling structural pricing power?