Marvell's 2029 $10 billion AI ASIC revenue target: Growth roadmap and competitive risk analysis

A $6 billion growth potential, where will it be filled from.

On May 27, 2026, Marvell Technology, when releasing its FY2027 Q1 financial report, simultaneously provided a long-term guidance that caught the market’s attention: the custom chip business revenue in FY2029 (by January 2030) will surpass $10 billion. Just a week later, at the COMPUTEX conference, NVIDIA CEO Jensen Huang publicly stated that Marvell will become "the next trillion-dollar market cap company," and combined with NVIDIA’s strategic move of investing $2 billion in Marvell, this data infrastructure semiconductor company's attention reached a historic peak.

However, there is a gap between the guidance and the conclusion that needs to be empirically filled. Currently, Marvell’s custom chip business is worth several billion dollars, and surpassing $10 billion in FY2029 means achieving several billion dollars of incremental revenue within about 4-5 years, with a cumulative increase of over 100%. Morningstar analyst William Kerwin pointed out that this guidance "implies that from FY2028 to FY2029, this business alone will bring in an incremental revenue of $5 billion"—meaning nearly $5B of growth must be achieved within FY2028 to FY2029.

Order Visibility: From Short-term Data to Long-term Guidance Deduction Path

The short-term data provided by Marvell’s FY2027 Q1 financial report offers the first empirical anchor for the long-term guidance.

Short-term performance data: Q1 FY2027 (ended May 2026) revenue reached a record $2.42B, up 28% year-over-year, exceeding the company’s previous guidance midpoint by $18 million. Operating cash flow was $638.8 million, also a record high. Non-GAAP net profit was $718 million, with diluted earnings per share of $0.80. CEO Matt Murphy stated in the earnings report that the midpoint of Q2 revenue guidance is $2.7 billion, implying a 35% YoY growth rate, and "it is expected that revenue growth will continue to accelerate each quarter throughout FY2027."

Data center business is the core growth driver. This segment’s Q1 revenue was $1.83 billion, exceeding the expected $1.81 billion, and Marvell expects the full-year data center business growth rate to be about 50%. In specific product lines, Marvell listed strong demand for multiple AI-related products, including 800G and 1.6T scalable optical devices, 51.2T Ethernet switches, custom XPU, and XPU-attached solutions.

Guidance upgrade: Marvell "significantly raised" its revenue outlook for FY2027 and FY2028 in this earnings report, with substantial improvements over the previous quarter’s forecast. The company now expects FY2028 revenue to grow to about $16.5 billion, higher than the previous estimate of $15 billion. The guidance for the custom chip to surpass $10 billion in FY2029 is an even further target based on this foundation.

Design wins (DWs) conversion capability: In the field of custom chips, Marvell disclosed that it currently has 18 design wins at the XPU and XPU-attached levels, with a significant portion already in mass production. These wins include complete custom XPU projects, XPU-attached chips, and integrated I/O chiplets within multi-chip XPU packages, collectively forming a "funnel" with a lifecycle revenue of over $7.5 billion—more than 10% of which is a confirmed revenue, i.e., over $750 million. During the earnings call, CEO Matt Murphy said, "We have comprehensive custom business collaborations with all major U.S. hyperscale cloud providers."

Three Pillars of Strategic Assets

Marvell’s ability to provide a $10 billion revenue target for FY2029 in custom chips is not built from scratch. Behind it are three accelerating strategic pillars.

NVIDIA Ecosystem Entry Ticket: NVLink Fusion and $2 Billion Investment

In March 2026, NVIDIA announced a strategic investment of $2 billion in Marvell, deepening cooperation through the NVLink Fusion platform. The structure of this partnership is: Marvell provides custom XPU and vertically scalable network solutions compatible with NVLink Fusion, while NVIDIA contributes Vera CPUs, ConnectX NICs, and BlueField DPU technologies.

The mutual value of this cooperation lies in: for Marvell, gaining an entry ticket into the global AI chip ecosystem and a shared customer channel; for NVIDIA, leveraging Marvell’s customization capabilities and high-speed interconnect technology to alleviate network bottlenecks in large-scale GPU deployments, while reducing dependence on a single major customer.

Jensen Huang elaborated on the technical logic of the cooperation during a joint discussion with Marvell CEO Matt Murphy at COMPUTEX. He pointed out that as AI infrastructure expands to multi-rack configurations, connectivity is becoming the core bottleneck that determines the limits of infrastructure, and Marvell’s optical connection technology and network chips "play an indispensable role" in this.

Dual Interconnect Acquisitions: Celestial AI + XConn

Alongside NVIDIA’s investment, Marvell completed two key acquisitions at the end of 2025 to early 2026:

• Celestial AI (announced in December 2025, completed in February 2026): The company’s Photonic Fabric optical interconnect technology platform enables full optical connections at the package, system, and rack levels, crucial for realizing scalable architectures connecting thousands of XPUs. After the deal, Marvell’s cash decreased by about $1 billion.
• XConn Technologies (announced in January 2026, acquired for about $540 million): The company offers advanced PCIe and CXL switch chips, with PCIe 5 and CXL 2.0 switches already in mass production, and PCIe 6 and CXL 3.1 switches in sampling stages.

These acquisitions are widely viewed as pre-positioning Marvell’s technological layout for revenue growth in 2028-2029, with substantial contributions expected to begin from 2028. The combined Marvell now offers a complete portfolio covering scale-out (optical modules + Ethernet switches), scale-up (UAIink switches + Celestial optical scale-up interconnects), and scale-across (data center interconnect modules)—a rare configuration among AI infrastructure suppliers.

Capacity-side Pre-investment

Achieving an annual target of over $10 billion for custom chips directly demands capacity expansion. Marvell’s deep cooperation with TSMC on advanced processes (3nm and below) and advanced packaging (such as 3D SoIC, CoWoS) is a prerequisite for meeting customer scale delivery needs. Although specific capacity investment details are limited publicly, the magnitude and speed of Marvell’s upward revision of FY2027-FY2028 revenue outlook imply that capacity arrangements are an implicit prerequisite supporting the guidance.

Logical Verification of Growth Goals: Breakdown and Calculation

Decompose the $10 billion target into verifiable growth units.

Incremental scale estimate: Currently, Marvell’s custom chip business does not have an exact FY2026 baseline figure, but considering FY2026’s total revenue of about $8.2 billion and the significant proportion of the custom chip business within the total data center revenue (FY2026 roughly $6-7 billion), we can roughly estimate the annualized CAGR needed to reach the $10 billion goal from the current baseline. If we take FY2026’s custom chip revenue as about $3-4 billion (based on the pace of 18 design wins translating into revenue), then from FY2026 to FY2029 (a four-year span), a cumulative growth of approximately 150%-230% is needed, with an annualized CAGR of about 25%-35%.

Design win conversion model: Of the 18 existing design wins, some have already entered mass production, while others are in design or validation stages. It typically takes 24-36 months from design start to mass production, meaning the FY2029 target largely depends on whether the design wins already secured can be converted into scaled revenue on time. Murphy mentioned collaborations with "all major U.S. hyperscale cloud providers," providing a first layer of demand certainty.

Synergy contribution from interconnect business: It’s noteworthy that the $10 billion guidance only pertains to the "custom chip business" and does not include Marvell’s interconnect product lines (optical modules DSPs, Ethernet switches, PCIe/CXL switches, etc.). The interconnect business itself is expected to grow over 70% in FY2027, further upward from the FY2026 baseline of 50%. This suggests that Marvell’s actual AI-related revenue could be significantly higher than $10 billion, and the combined sales of custom chips and interconnect solutions—delivering both XPU and full interconnect solutions to the same customer project—are a key way to generate synergistic value.

Comparison with competitors: Broadcom achieved $10.8 billion in AI semiconductor revenue in FY2026 Q2, up 143%, and maintained a full-year AI semiconductor revenue guidance exceeding $100 billion. Although Marvell’s $10 billion guidance is an order of magnitude smaller, it reflects different positioning: Broadcom has volume advantage, while Marvell seeks differentiation through growth speed and ecosystem binding.

Risk Assessment: Logical Verification in Five Dimensions

Customer Concentration and Substitution Risk

Marvell’s current custom chip customer base includes all major U.S. hyperscale cloud providers, but customer concentration remains a variable to watch. These large customers have motivations and capabilities for in-house chip development. For example, Google has collaborated with Broadcom on TPU for years, and Amazon has Trainium/Inferentia products. While Marvell’s NVLink Fusion ecosystem provides a differentiated position with NVIDIA, the stability of this ecosystem relationship also poses a risk—NVIDIA’s strategic investment and the structure of exclusivity clauses will directly influence Marvell’s freedom of customer choice.

Uncertainty in Technical Path: UAIink vs NVLink vs Optical Scale-up

Interconnect technology paths are still in a stage where industry standards have not fully converged. NVIDIA’s NVLink is proprietary, while UAIink is an open standard promoted jointly by Marvell and other industry players. There is competition between these two, but Marvell’s dual role—participating in NVIDIA’s ecosystem via NVLink Fusion and maintaining leadership in UAIink—can be both an advantage and a potential source of conflict. Additionally, optical scale-up interconnects as the next-generation technology still needs to prove its commercial maturity. Marvell’s acquisition of Celestial AI is essentially a strategic bet on future 3-5 year technology paths; if optical interconnect commercialization lags or copper interconnect performance boundaries are delayed beyond expectations, product revenue realization could be affected.

Capital Expenditure Cycle Risks

The four major U.S. cloud providers (Amazon, Microsoft, Google, Meta) are expected to invest over $700 billion in AI infrastructure in 2026, a significant increase from about $400 billion in 2025. However, cloud providers’ capital expenditure is cyclical and influenced by macroeconomic factors and ROI of AI applications. Marvell’s $10 billion guidance implicitly assumes that AI infrastructure investments will maintain high growth through 2027-2029, with a rising share of custom chips in overall AI chip procurement. If macroeconomic conditions or AI commercialization pace change, procurement from cloud providers could slow.

Evolving Competitive Landscape

Marvell faces dual competition from Broadcom (AVGO) and AMD in the custom chip space. Broadcom has six core custom chip customers, with mature 3.5D XDSiP packaging technology offering performance advantages. Currently, Broadcom’s order scale for custom chips is about ten times that of Marvell, a significant gap. However, Macquarie analysts recently downgraded Broadcom’s FY2028 earnings forecast by 21%, warning that increasing competition in AI ASICs could impair Broadcom’s profitability and growth—this is a risk for Broadcom but also an opportunity for Marvell to expand market share through differentiation in the next 2-3 years.

Process and Packaging Dependency Risks

Advanced process nodes (3nm and below) and advanced packaging (CoWoS, 3D SoIC) capacity are hard constraints for Marvell to achieve large-scale delivery. Currently, global advanced packaging capacity is highly concentrated at TSMC, with Samsung and Intel actively expanding but not yet reaching a scale to replace TSMC. If demand for custom chips from cloud providers concentrates in 2027-2029, whether advanced packaging capacity can meet this demand will be a universal bottleneck. If Marvell cannot secure sufficient priority in capacity allocation—especially with NVIDIA and other large customers also occupying significant advanced packaging capacity—its FY2029 revenue targets could face supply chain constraints.

Necessary Conditions for a Trillion-Dollar Market Cap

Jensen Huang’s statement that "Marvell is the next trillion-dollar company"—whether from the credibility of the statement itself or the gambling nature of NVIDIA’s $2 billion investment—must be translated into quantifiable financial conditions. Currently, Marvell’s market cap is around $192 billion; reaching $1 trillion requires about a 5x increase in stock price.

From a valuation logic perspective, to achieve this scale of market cap, Marvell needs to meet three conditions: first, the FY2029 $10 billion custom chip target is achieved or exceeded, supporting a sustained high-growth narrative; second, interconnect business generates substantial revenue in optical scale-up, CPO (co-packaged optics), and other frontier areas, driving both revenue structure and profitability improvements; third, the conversion speed of design wins to revenue and continuous improvement in gross margins provide verifiable financial evidence.

The semiconductor trillion-dollar club currently includes 8 members, such as NVIDIA, TSMC, Broadcom, etc. To join, Marvell not only needs to complete the FY2029 $10 billion custom chip goal but also needs to continuously demonstrate a replicable growth model based on this foundation.

Conclusion

Marvell’s FY2029 revenue target of $10 billion for custom chips is built on three cross-verifiable growth pillars: the short-term data from Q1 FY2027, the demand certainty provided by NVIDIA’s $2 billion strategic investment and NVLink Fusion ecosystem, and the next-generation interconnect moat formed by Celestial AI + XConn acquisitions.

Achieving the $10 billion goal still faces multiple risks, including customer concentration, intensified competition, technological path uncertainty, and capacity constraints. For market participants, assessing the credibility of this guidance hinges not on believing or doubting the macro conclusion of "trillion-dollar company," but on closely monitoring quantifiable indicators: the actual conversion rate of design wins into revenue, whether interconnect revenue growth can continue to outperform expectations, and whether advanced packaging capacity can support large-scale delivery.

Marvell’s next three years will be a critical transition from being "endorsed by Jensen Huang" to "verifying valuation through revenue"—the former can boost stock prices by over 30% in a single trading day, while the latter requires steady performance over multiple quarters. Which side ultimately dominates pricing will be determined by the evolution of these data points quarter by quarter.