On June 2, 2026, during the COMPUTEX event in Taipei, NVIDIA CEO Jensen Huang and Marvell CEO Matt Murphy appeared together on stage. Huang predicted Marvell would become "The next trillion-dollar company, ladies and gentlemen," forecasting Marvell as the next chip company to surpass a $1 trillion market capitalization. On the day of his remarks, Marvell’s closing price was $290.79, up more than 32% for the day, with its market cap rising to approximately $254.4 billion.
This was not an off-the-cuff statement. Three months earlier, NVIDIA had invested $2 billion in Marvell, establishing a deep strategic partnership based on the NVLink Fusion platform. During their conversation, Huang emphasized that Marvell’s data center switching chips play an "essential" role in handling AI workloads—when computational tasks are distributed across thousands of interconnected chips, "connectivity is everything."
The Evolution of AI Infrastructure Bottlenecks: From "Compute Scarcity" to "Connectivity Scarcity"
In his COMPUTEX keynote, Matt Murphy pointed out that the bottlenecks in AI infrastructure are undergoing a structural shift. The first phase of compute scarcity has been alleviated by large-scale GPU deployments; currently, the main bottleneck is inter-chip bandwidth and latency; the next phase will focus on system-level optical connectivity and heterogeneous integration efficiency.
This assessment is rooted in the exponential expansion of AI cluster sizes. When training AI clusters with tens of thousands or even hundreds of thousands of cards, the efficiency of data exchange between compute chips directly determines the actual system utilization. Huang described this precisely: "When we break down computational problems and distribute them across the entire data center, what truly makes it all work is connectivity."
Marvell’s technical strengths align perfectly with this bottleneck. According to its FY2027 Q1 financial report, Marvell posted $2.418 billion in revenue, up 28% year-over-year. Data center business contributed $1.83 billion, accounting for 76% of total revenue and growing 27% year-over-year. Shipments of 800G PAM4 products continued to ramp up, and 1.6T solutions have entered a rapid growth phase. The data center interconnect (DCI) module business is expected to reach $1 billion in annualized revenue by fiscal 2028. Operating cash flow hit a record $638.8 million for the quarter, providing ample liquidity for future capacity commitments and M&A integration.
NVLink Fusion: NVIDIA’s "Open Closed" Strategy and Marvell’s Ecosystem Position
In May 2025, NVIDIA launched the NVLink Fusion platform, opening its NVLink interconnect technology to custom XPU chips. The core idea is to allow partners like Marvell to integrate custom chips into NVIDIA’s rack-scale systems and end-to-end network platforms, enabling hyperscale cloud providers to maintain compatibility with the NVIDIA ecosystem while gaining the flexibility of custom chips.
In March 2026, the two companies elevated their collaboration to a strategic partnership. NVIDIA invested $2 billion in Marvell, which will provide custom XPU and NVLink Fusion-compatible scale-up interconnect solutions, while NVIDIA offers supporting technologies and ecosystem integration. The partnership also extends to silicon photonics and AI-RAN telecom infrastructure.
From an architectural perspective, the technical linchpin of NVLink Fusion is a small chip providing 1.8TB/s bidirectional bandwidth, allowing cloud providers to scale their own XPU deployments to millions of units. Marvell’s value in this ecosystem is as the "connectivity infrastructure provider"—it does not compete directly with NVIDIA’s GPUs but instead delivers custom XPU and high-speed switching solutions for the interconnect layer of AI data centers.
This strategy contrasts with Broadcom’s approach. Broadcom is also a leader in custom AI chips, with its 3.5D XDSiP packaging platform optimized for AI XPU, but its partnership with the NVIDIA ecosystem is less deep. Marvell, through NVLink Fusion, gains direct access to NVIDIA’s technology and supply chain ecosystem, becoming an irreplaceable component in the AI data center interconnect structure.
Marvell’s Growth Path: Key Milestones FY2027–FY2029
Financial guidance reveals a clear trajectory for Marvell’s growth:
FY2027 full-year guidance: Total revenue of approximately $11.5 billion, up about 40% year-over-year. The company has set aside roughly $1 billion in prepayments to secure additional capacity, with the first payments to begin in Q2.
FY2028 full-year guidance: Total revenue of about $16.5 billion, up roughly 45% year-over-year, and $1.5 billion higher than prior quarterly guidance. Data center business is projected to grow about 50% for the year.
FY2029 custom chip revenue guidance: The custom chip business aims to surpass $10 billion for a single project. Morningstar analyst William Kerwin notes that this means custom chips alone could deliver $5 billion in incremental revenue between FY2028 and FY2029.
Marvell’s current market cap is around $254.4 billion (as of June 2, 2026), leaving about 4x growth potential to reach $1 trillion. Static calculations suggest that at a 35–40x price-to-earnings ratio, a $1 trillion valuation would require net profits of $25–28.5 billion. Based on FY2027 revenue of about $11.5 billion and a non-GAAP net margin of 25–30%, current net profits are still far from this target. Therefore, whether Marvell can sustain high-speed growth—not just a one-off valuation leap—will be the key variable in assessing its trillion-dollar goal.
Strategic Acquisitions: Celestial AI and XConn Complete the Connectivity Puzzle
Between late 2025 and early 2026, Marvell completed two critical acquisitions, strengthening its connectivity strategy from both technical and market coverage perspectives:
Acquisition of Celestial AI (signed December 2025, completed February 2026). Celestial AI boasts a breakthrough photonic transmission platform, supporting chip-level, system-level, and rack-level optical I/O interconnects. AI data center connectivity is shifting from rack-to-rack electrical interconnects to optical interconnects at the chip packaging level—a necessary path for scaling clusters to tens of thousands of cards. The acquisition, completed for about $1 billion in cash, gives Marvell a differentiated technology edge in optical connectivity.
Acquisition of XConn Technologies (signed January 2026). XConn offers industry-leading PCIe 5/6 and CXL switching chip product lines, with PCIe 6 and CXL 3.1 switches already sampling. The deal was valued at about $540 million. CXL technology enables memory pooling and resource sharing in high-performance computing. XConn’s CXL switching chips will complement Marvell’s existing CXL memory expansion controllers.
Both acquisitions focus on "scale-up connectivity," complementing Marvell’s existing scale-out Ethernet switching business. Revenue contributions from these deals are expected to materialize gradually in FY2028–FY2029.
Marvell vs. Broadcom: Comparing Two Custom AI Chip Strategies
Marvell and Broadcom are often seen by investors as the "duopoly" in custom AI chips, but their ecosystem positions are fundamentally different:
Broadcom’s approach: Centers on in-house ASIC customization, targeting AWS, Google, Meta, and other cloud providers with "anti-NVIDIA" custom AI chip solutions. Its 3.5D XDSiP packaging platform emphasizes performance and efficiency but lacks deep integration with the NVIDIA ecosystem. In FY2025 Q1, the semiconductor division (including custom silicon solutions) grew 11% year-over-year, but this is noticeably slower than Marvell’s data center growth (+27%).
Marvell’s approach: Anchored in connectivity technology, it delivers custom XPU and interconnect solutions within the NVIDIA ecosystem. The NVLink Fusion partnership allows Marvell to participate in NVIDIA’s AI Factory and AI-RAN ecosystem, while maintaining comprehensive custom engagements with all major US hyperscale cloud providers. CEO Matt Murphy stated on the earnings call: "We have comprehensive custom business engagements with all US hyperscale cloud providers."
There’s no simple judgment of superiority between the two paths, but Marvell, supported by the NVIDIA ecosystem, enjoys clearer short-term growth certainty during the current AI infrastructure expansion cycle. On the other hand, whether NVIDIA will keep Marvell confined to the interconnect layer rather than the core compute layer, and whether Marvell’s deep integration with NVIDIA will limit its expansion into other GPU ecosystems, are variables investors should continue to monitor.
Conclusion
Jensen Huang’s "trillion-dollar" endorsement has brought Marvell significant market attention, but a fundamental analysis shows that achieving this goal requires four core conditions to be met simultaneously:
First, AI data center capital expenditures must continue robust growth—US tech giants are expected to spend about $400 billion on AI infrastructure in 2025, with forecasts exceeding $700 billion in 2026. This is the foundational anchor for industry momentum.
Second, Marvell’s custom chip and connectivity products must consistently win procurement share among hyperscale cloud providers, enabling FY2028–FY2029 revenue guidance to be met on schedule.
Third, Celestial AI and XConn’s optical connectivity and CXL switching technologies must achieve commercial validation in large-scale data center deployments, fueling sustained growth beyond FY2030.
Fourth, the partnership between NVIDIA and Marvell must evolve from technical collaboration to deeper supply chain and capacity integration, maintaining strong alignment of interests.
As of June 2026, Marvell’s custom chip portfolio includes 18 XPU projects, with revenue contributions expected to accelerate in FY2027. Q2 FY2027 revenue is projected at about $2.7 billion, with a clear pattern of sequential growth. From the perspectives of financial fundamentals, technical layout, and ecosystem positioning, Marvell has the structural conditions to move from a $200 billion market cap toward $1 trillion. However, challenges remain, including capacity constraints, competitive pressures, and uncertainty in technology roadmaps. While investors may focus on the "trillion-dollar" halo, they should pay close attention to the pace of actual FY2027 revenue realization and the evolution of custom chip gross margins.
