On June 2, NVIDIA (NVDA) CEO Jensen Huang made a surprise appearance at Marvell Technology’s (MRVL.US) keynote during the 2026 Computex in Taipei. In front of hundreds of industry professionals, he delivered a statement powerful enough to rewrite a semiconductor company’s valuation in just 24 hours: "The next trillion‑dollar company, ladies and gentlemen."
Coming from Jensen Huang, these words carry far more weight than any typical investment bank analyst’s rating upgrade.
After the US market opened on June 2, Marvell’s stock price surged intraday to a record high of $291.30, closing up 32.52% from the previous session at $270.79—the largest single-day gain since 2000. Trading volume exceeded 102 million shares, about 257% higher than the March average. Before Huang’s endorsement, Marvell’s market cap stood at roughly $192 billion; by the close, it had soared to around $254.3 billion. In a single day, one sentence added about $62.3 billion in market value.
The bigger question is: Was Jensen Huang’s "trillion-dollar" remark just a sentiment catalyst, or does it signal a structural shift in AI infrastructure investment logic?
The Invisible Infrastructure of AI Compute: From NVDA to MRVL—A Shift in Investment Logic
From 2023 to 2025, the core narrative in AI capital markets has revolved around "compute power." GPUs, HBM memory, and advanced packaging—NVIDIA has monopolized AI compute pricing power thanks to its GPU ecosystem, while HBM suppliers have seen their valuations soar with the rise of trillion-parameter large models.
But at Computex 2026, Jensen Huang spent over 10 minutes systematically carving out an independent narrative space for "connectivity."
Huang’s view is sharply focused: today’s AI data centers are shifting toward "decentralized, distributed, and heterogeneous" architectures. As agent-based computing models spread, the peak compute of a single GPU no longer defines the upper limit of overall data center efficiency. The real bottleneck lies in how thousands of chips achieve ultra-low latency, high-bandwidth data transfer. The core infrastructure enabling this distributed architecture is precisely the networking connectivity chips provided by Marvell—PAM4 DSP optical interconnects, custom ASICs (application-specific integrated circuits), Ethernet switch chips, and the Teralynx T-series AI switch chips.
Marvell CEO Matt Murphy reinforced this point in the March earnings call, stating that Marvell is becoming "a pivotal supplier at the center of AI infrastructure transformation."
Technically, Marvell’s AI business is powered by two mutually reinforcing engines.
Engine One—Optical Interconnect: Marvell’s PAM4 DSP chips occupy the high-value signal processing segment in 800G and 1.6T optical modules for AI data centers. As AI clusters scale up and cross-node data exchange intensifies, optical interconnects take up an ever-larger share of the overall BOM (bill of materials). Huang’s assessment: "Use copper wherever possible; use optics only when necessary." Over the long term, copper cables will dominate in-cabinet transmission, while fiber optics will play a central role in inter-rack, inter-data center, and cross-data center expansion—Marvell is uniquely positioned to benefit from both paths.
Engine Two—Custom AI Chips (Custom ASIC): Marvell designs custom AI accelerators for cloud service providers like Amazon AWS Trainium and Google, focusing on energy efficiency for specific workloads compared to general-purpose GPUs. As cloud providers seek differentiated compute stacks, custom ASICs are moving from "alternative" to "mainstream." Marvell expects this segment to nearly double by FY2027 and has significantly raised its connectivity growth outlook in its earnings guidance.
Huang also highlighted the jointly launched NV Link Fusion technology. This solution allows cloud providers to integrate their in-house custom chips with NVIDIA’s system architecture and networking stack—combining NVIDIA’s platform strengths with Marvell’s expertise in interconnect, silicon photonics, and optical technologies to build a "decentralized, distributed, heterogeneous" data center infrastructure.
When a market features two high-growth, mutually reinforcing product lines, and enjoys top-level endorsement and a $2 billion strategic investment from the world’s largest AI chip company, the market’s one-day reaction is grounded in structural fundamentals—not just short-term sentiment.
MRVL vs. Broadcom vs. AMD: The AI Networking Chip Three-Way Battle
The AI networking chip sector is now a three-way race: Marvell, Broadcom (AVGO), and AMD (AMD.US) each occupy different strategic positions.
Broadcom (AVGO): The Leader in Custom ASIC Market Share. Broadcom currently holds about 60% of the AI custom chip market, with major clients including Alphabet’s TPU series. Marvell, supported by AWS Trainium and Google orders, is rapidly catching up. Their product portfolios overlap significantly, but Broadcom is larger and serves a broader client base; Marvell relies more on deep partnerships with core cloud computing players, offering greater flexibility.
AMD (AMD): The Integrator of GPUs and Networking Chips. AMD’s core strength in data centers lies in its platform synergy between general-purpose GPUs (MI350/MI400 series) and EPYC processors. On the networking side, AMD focuses on integrating its own accelerator clusters, but its third-party networking component market penetration lags behind Marvell and Broadcom. AMD’s edge is in compute density, while Marvell’s is in components—the two are more complementary than direct substitutes.
A more precise positioning would be:
Broadcom = A mature, large-scale ASIC conglomerate with a broad client base;
Marvell = A specialized "AI traffic hub" focused on optical interconnects and custom ASICs;
AMD = An independent acceleration path at the general-purpose GPU and server CPU layer.
These three companies are not locked in zero-sum competition. Major cloud providers, seeking to avoid vendor lock-in, often source ASICs from multiple suppliers and maintain long-term dependency on Marvell at the Ethernet connectivity layer. This means Marvell’s "irreplaceability" in the AI networking space isn’t built on monopoly, but on a combination of technical maturity, first-mover advantage, and client trust—still rare resources in today’s semiconductor industry.
Jensen Huang’s Trillion-Dollar Club: Does Marvell Truly Qualify Structurally?
As of now, most companies with market caps over $1 trillion are AI and tech giants. NVIDIA leads with about $5.4 trillion, while Apple and Microsoft have both surpassed $4 trillion. For MRVL, with a current market cap around $254.3 billion, to reach the trillion-dollar mark in the next 3–5 years, it needs about 3.2x valuation expansion.
This is not an impossible target, but several quantifiable prerequisites must be met.
Prerequisite 1: Sustained expansion in the PAM4 DSP/optical interconnect market, with Marvell maintaining a leading position in each product generation (1.6T, 3.2T, even 6.4T). The 1.6T products are already in large-scale deployment, and 3.2T samples will be available next year. Technological progress must keep this pace.
Prerequisite 2: Custom ASIC orders from cloud clients like Amazon and Google must consistently exceed expectations, not just remain at the MOU stage.
Prerequisite 3: Commercialization of new technologies like CPO must transition smoothly after 2027, contributing incremental revenue growth each year. CPO is still in the validation phase, representing the biggest uncertainty in Marvell’s trillion-dollar narrative.
Prerequisite 4: Sustained macro capital expenditure. Wall Street now forecasts that US tech giants’ AI capex in 2026 will surge from a previous estimate of $433 billion to $805 billion, with 2027 projected at $1.1 trillion. This structural spending trend underpins Marvell’s data center business, but any slowdown in the capex cycle would put significant pressure on lofty valuations.
Morgan Stanley estimates that by 2028, nearly $3 trillion in AI-related investment will flow through the global economy, with over 80% of spending still ahead. For Marvell, this means not just an opportunity for growth, but a race between execution capability and capital constraints.
Trading Marvell (MRVL) Stock on Gate: A New Pathway
For investors looking to gain exposure to Marvell in the AI infrastructure space, platform selection and capital efficiency are key considerations.
On June 1, 2026, Gate officially launched live US stock trading, marking the first unified integration of crypto asset accounts and global stock investment capabilities. As of June 3, 2026, Gate supports over 10,000 stocks and ETFs, covering major US exchanges like NYSE, Nasdaq, and NYSE Arca.
Gate’s stock trading service is fundamentally different from tokenized stocks or CFD contracts:
Direct market access and regulatory compliance. Gate partners with licensed US broker-dealers and clearing firms, giving users direct access to the market. Marvell (MRVL) shares purchased on Gate are real underlying assets traded in sync with Nasdaq, custodied by SIPC-member brokers, and enjoy corresponding asset protection where eligible.
Zero holding costs—no funding rates, no overnight fees. Unlike perpetual contracts with funding rates or CFDs with swap fees, Gate’s spot stock trading incurs zero holding costs, making it ideal for long-term investors in US equities like Marvell.
Users can buy MRVL shares directly on Gate using their USDT balance.
The trading process is streamlined: after completing KYC and regional access, users enter the TradFi stock section in the Gate App, transfer USDT, check real-time quotes, and enter the MRVL ticker to buy or sell. All assets can be managed separately through an independent account system.
With Gate’s stock trading service, crypto investors can, for the first time, hold both digital assets and US stocks in a unified account framework—capturing the growth potential of AI infrastructure leaders like Marvell while retaining the flexibility and 24/7 trading advantages of the crypto market.
Risk Factors and Disclaimer
Any "trillion-dollar" narrative must be grounded in clearly defined variables, not linear and overly optimistic projections.
MRVL’s current P/E ratio is already at a significant high, and the 32% single-day surge has priced in a substantial portion of optimistic expectations. Future stock performance will depend heavily on whether Marvell can maintain mass-production leadership in each product generation, whether cloud providers’ ASIC orders continue to exceed expectations, whether new technologies like CPO commercialize on schedule, and whether there are any unexpected slowdowns in macro AI capex.
The path Jensen Huang outlined is logically sound, but the key to achieving this milestone is not any CEO’s verbal endorsement—it is Marvell’s real-world execution at every product iteration and client delivery point. The market rewards verifiable data, not one-way narratives.
Conclusion
Jensen Huang’s public prediction at Computex thrust Marvell from a "behind-the-scenes AI infrastructure player" into the spotlight. The 32%+ single-day surge and $62.3 billion jump in market cap certainly reflect a sentiment catalyst, but the underlying structural logic is far from baseless. The shift in AI data centers from "compute stacking" to "distributed interconnect" is redefining the value of networking chips. Marvell’s technical positioning in optical interconnects and custom ASICs, combined with the ongoing upcycle in cloud providers’ capital expenditure, gives it the industrial foundation needed for a trillion-dollar narrative.
However, the road to a trillion-dollar valuation is never a simple linear projection. MRVL’s current high valuation already bakes in a lot of optimism. Each future product iteration and every client order fulfillment will serve as a market test of this narrative. For investors, Huang’s endorsement offers a directional signal worth tracking, but the real opportunity lies with those who can distinguish between "industry trends" and "overpricing." In Gate’s unified approach to digital assets and US equities, staying attuned to structural changes in AI infrastructure may prove more valuable in the long run than chasing single-day gains.
