Where Will $6 Billion in Growth Come From?
On May 27, 2026, Marvell Technology released its FY2027 Q1 earnings report and simultaneously issued a long-term outlook that caught the market’s attention: the company expects its custom chip business to generate over $10 billion in revenue by FY2029 (ending January 2030). Just a week later at COMPUTEX, NVIDIA CEO Jensen Huang publicly stated that Marvell could become "the next $1 trillion market cap company." Coupled with NVIDIA’s previous $2 billion strategic investment in Marvell, the data infrastructure semiconductor company has found itself in the industry spotlight like never before.
However, there’s a significant gap between guidance and realization that must be bridged by tangible results. Marvell’s custom chip business currently generates several billion dollars in revenue. Reaching $10 billion by FY2029 means the company must add several billion dollars in just four to five years—a cumulative increase of over 100%. Morningstar analyst William Kerwin noted that this guidance "implies a $5 billion revenue increase in custom chips from FY2028 to FY2029 alone"—that is, nearly $5 billion in growth in a single year.
Order Visibility: From Short-Term Data to Long-Term Guidance
Marvell’s FY2027 Q1 earnings report provides the first empirical anchor for its long-term outlook.
Short-term performance: In Q1 FY2027 (ending May 2026), revenue reached a record $2.418 billion, up 28% year-over-year and exceeding the company’s previous guidance midpoint by $18 million. Operating cash flow hit a record $638.8 million. Non-GAAP net income was $718 million, with diluted EPS of $0.80. CEO Matt Murphy stated that Q2 revenue guidance is centered at $2.7 billion, implying 35% year-over-year growth, and that "we expect revenue growth to accelerate each quarter throughout FY2027."
The data center segment is the core growth driver. Q1 revenue from this segment was $1.83 billion, beating the expected $1.81 billion. Marvell anticipates roughly 50% annual growth in its data center business. On the product side, Marvell highlighted robust demand for several AI-related offerings, including 800G and 1.6T pluggable optical modules, 51.2T Ethernet switches, custom XPUs, and XPU-attach solutions.
Upward guidance revision: Marvell "significantly raised" its FY2027 and FY2028 revenue outlook in this report, marking a substantial upgrade from the prior quarter. The company now expects FY2028 revenue to reach about $16.5 billion, up from the previous $15 billion forecast. The FY2029 custom chip target of over $10 billion is set on top of this new baseline.
Design win conversion: In the custom chip space, Marvell disclosed it currently holds 18 design wins at the XPU and XPU-attach level, with a significant portion already in mass production. These wins include full custom XPU projects, XPU-attached chips, and electrical I/O chiplets integrated into multi-chip XPU packages. Collectively, these represent over 10%—more than $750 million—of the $7.5 billion "funnel" in lifetime revenue. CEO Matt Murphy stated on the earnings call, "We have comprehensive custom engagements with all U.S. hyperscale cloud providers."
Three Strategic Pillars
Marvell’s $10 billion custom chip guidance for FY2029 isn’t built from scratch. It stands on three accelerating strategic pillars.
NVIDIA Ecosystem Access: NVLink Fusion and $2 Billion Investment
In March 2026, NVIDIA announced a $2 billion strategic investment in Marvell, cementing a deep partnership through the NVLink Fusion platform. Under this arrangement, Marvell supplies custom XPUs and NVLink Fusion-compatible scale-up networking solutions, while NVIDIA provides Vera CPUs, ConnectX NICs, and BlueField DPUs.
This partnership is mutually valuable: Marvell gains entry into the world’s leading AI chip ecosystem and access to shared customers; NVIDIA benefits from Marvell’s customization and high-speed interconnect expertise, helping to alleviate networking bottlenecks in large GPU clusters and reduce reliance on any single major customer.
At COMPUTEX, Jensen Huang, alongside Marvell CEO Matt Murphy, further explained the technical rationale for the collaboration. As AI infrastructure scales to multi-rack configurations, connectivity becomes the core bottleneck limiting system performance. Marvell’s optical interconnect and networking chips, Huang noted, "play an indispensable role" in overcoming these limits.
Dual Interconnect Acquisitions: Celestial AI + XConn
Alongside NVIDIA’s investment, Marvell completed two key acquisitions from late 2025 to early 2026:
- Celestial AI (announced December 2025, closed February 2026): The company’s Photonic Fabric optical interconnect platform enables all-optical connectivity at the package, system, and rack levels—crucial for building scale-up architectures interconnecting thousands of XPUs. The deal reduced Marvell’s cash by about $1 billion.
- XConn Technologies (announced January 2026, acquired for ~$540 million): XConn provides advanced PCIe and CXL switch chips. Its PCIe 5 and CXL 2.0 switches are in mass production, with PCIe 6 and CXL 3.1 switches sampling.
These acquisitions are widely viewed as Marvell’s preemptive technology bets for 2028–2029 revenue growth, with material contributions expected from 2028 onward. The combined Marvell now offers a complete interconnect portfolio—covering scale-out (pluggable optical modules + Ethernet switches), scale-up (UALink switches + Celestial optical scale-up interconnect), and scale-across (data center interconnect modules)—a rare configuration among AI infrastructure suppliers.
Proactive Capacity Investments
A $10 billion annual custom chip target requires robust manufacturing capacity. Marvell’s deep partnership with TSMC in advanced nodes (3nm and below) and advanced packaging (such as 3D SoIC, CoWoS) is essential to meet large-scale customer delivery needs. While public disclosures on specific capacity investments are limited, the magnitude and speed of Marvell’s upward FY2027–FY2028 guidance revisions suggest that capacity arrangements are a crucial, if implicit, underpinning of its targets.
Stress-Testing the Growth Target: Breakdown and Sensitivity Analysis
Let’s break down the $10 billion target into verifiable growth components.
Incremental growth estimate: While there’s no precise FY2026 baseline for Marvell’s custom chip business, FY2026 total revenue was about $8.2 billion, with custom chips comprising a significant share of the data center segment ($6–7 billion in FY2026). If we estimate FY2026 custom chip revenue at $3–4 billion (based on the 18 design wins’ revenue ramp), then reaching $10 billion by FY2029 would require cumulative growth of roughly 150%–230% over four years, or a 25%–35% CAGR.
Design win conversion model: Of the 18 current design wins, some are already in mass production, with others in design or validation. Custom chip projects typically take 24–36 months from design kickoff to volume production, meaning the FY2029 target heavily depends on timely conversion of today’s wins into large-scale revenue. CEO Matt Murphy’s mention of "all U.S. hyperscale cloud providers" as customers provides a strong base of demand certainty.
Interconnect business synergy: Notably, the $10 billion guidance covers only the "custom chip business," excluding Marvell’s interconnect product lines (optical module DSPs, Ethernet switches, PCIe/CXL switches, etc.). The interconnect business itself is expected to grow over 70% in FY2027, up from a 50% growth forecast in FY2026. This means Marvell’s actual AI-related revenue could far exceed $10 billion, with combined sales of XPUs and full interconnect solutions to a single customer creating significant synergy.
Peer comparison: Broadcom reported $10.8 billion in AI semiconductor revenue in FY2026 Q2, up 143% year-over-year, and expects over $100 billion in AI semiconductor revenue for the full year. While Marvell’s $10 billion custom chip target is an order of magnitude smaller, it reflects a different strategic positioning: Broadcom dominates in scale, while Marvell seeks differentiation through faster growth and deeper ecosystem integration.
Risk Assessment: Five Key Dimensions
Customer Concentration and Substitution Risk
Marvell’s custom chip clients include all major U.S. hyperscale cloud providers, but customer concentration remains a concern. These large customers have both the motivation and capability to develop their own chips. For example, Google has collaborated with Broadcom on TPUs for years, and Amazon has its Trainium/Inferentia line. While Marvell gains a unique position in the NVIDIA ecosystem through NVLink Fusion, the stability of this relationship is itself a risk—NVIDIA’s role as a strategic investor, and the structure of exclusivity clauses, will directly affect Marvell’s customer flexibility.
Technology Path Uncertainty: UALink vs. NVLink vs. Optical Scale-Up
Interconnect technology standards are still evolving. NVIDIA’s NVLink is proprietary, while UALink—championed by Marvell and others—is open. The two compete technically, but Marvell participates in both: it’s deeply involved with NVIDIA via NVLink Fusion, yet leads the UALink initiative. This dual role is both an advantage and a potential source of conflict. Furthermore, optical scale-up interconnects, as the next-gen direction, remain commercially unproven. Marvell’s acquisition of Celestial AI is essentially a strategic bet on the next three to five years. If optical interconnects commercialize more slowly than expected, or if copper interconnects’ performance limits are extended, revenue realization could be delayed.
Customer Capex Cycle Risk
The four major U.S. cloud providers (Amazon, Microsoft, Google, Meta) are expected to invest over $700 billion in AI infrastructure in 2026, up sharply from about $400 billion in 2025. However, their capital spending is cyclical and influenced by both macroeconomic factors and the realized ROI of AI applications. Marvell’s $10 billion guidance implicitly assumes that AI infrastructure investment will continue to grow rapidly through 2027–2029, and that custom chips will take an increasing share of overall AI chip procurement. If the macro environment or AI commercialization pace changes, cloud providers’ purchasing could slow.
Competitive Landscape Evolution
Marvell faces competition from both Broadcom (AVGO) and AMD in custom chips. Broadcom has six core custom chip clients and a mature 3.5D XDSiP packaging platform with performance advantages. Broadcom’s custom chip order volume is about 10 times Marvell’s. However, Macquarie analysts recently cut Broadcom’s FY2028 earnings forecast by 21%, warning that intensifying competition in the AI ASIC market could hurt Broadcom’s profitability and growth—highlighting both risk for Broadcom and a potential opening for Marvell to gain share through differentiation over the next two to three years.
Process and Packaging Dependency Risk
Advanced process nodes (3nm and below) and advanced packaging (CoWoS, 3D SoIC, etc.) are hard constraints for Marvell’s large-scale deliveries. Currently, advanced packaging capacity is highly concentrated at TSMC, with Samsung and Intel expanding but not yet at scale. If cloud providers’ custom chip demand surges in 2027–2029, advanced packaging capacity could become an industry-wide bottleneck. If Marvell cannot secure sufficient allocation—especially as NVIDIA and other large clients also compete for capacity—its FY2029 revenue goals could face real supply chain constraints.
What It Takes to Reach a $1 Trillion Valuation
Jensen Huang’s assertion that "Marvell is the next $1 trillion company"—whether taken at face value or as a reflection of NVIDIA’s $2 billion bet—must ultimately be measured against financial reality. Marvell’s current market cap is around $192 billion; reaching $1 trillion requires roughly a fivefold increase in share price.
From a valuation perspective, Marvell must meet three conditions to achieve this leap: first, it must not only reach but exceed the FY2029 $10 billion custom chip target to sustain a high-growth narrative; second, its interconnect business must generate substantial revenue in cutting-edge areas like optical scale-up and CPO (co-packaged optics), driving both revenue mix and profitability; third, it must provide verifiable financial evidence of faster design win-to-revenue conversion and ongoing gross margin improvement.
There are currently eight members of the semiconductor "$1 trillion club," including NVIDIA, TSMC, and Broadcom. For Marvell to join their ranks, it must not only achieve the $10 billion custom chip milestone by FY2029 but also demonstrate a repeatable growth model beyond that.
Conclusion
Marvell’s guidance for over $10 billion in custom chip revenue by FY2029 is anchored in three cross-validated growth drivers: the short-term data from its FY2027 Q1 earnings, the demand certainty provided by NVIDIA’s $2 billion strategic investment and NVLink Fusion ecosystem integration, and the next-generation interconnect moat built through the Celestial AI and XConn acquisitions.
The path to $10 billion remains challenged by customer concentration, intensifying competition, technology roadmap uncertainty, and capacity constraints. For market participants, the credibility of this guidance hinges not on belief in the "$1 trillion company" narrative, but on close monitoring of quantifiable metrics: the actual speed of design win-to-revenue conversion, whether interconnect revenue growth continues to outpace expectations, and whether advanced packaging capacity supports large-scale delivery.
Over the next three years, Marvell will transition from being "anointed by Jensen Huang" to "proving its valuation through revenue." The former can drive a 30% share price jump in a single trading day; the latter requires several consecutive quarters of solid results. Which side ultimately determines valuation will be revealed by the data, quarter by quarter.
