Marvell’s 40% AI Growth Target Hinges on Open Fabric Adoption vs. NVLink

The numbers are clear. Marvell's recent earnings beat and raised guidance aren't just a bump in the road; they are a validation of its strategic pivot onto the steepening phase of the AI infrastructure S-curve. The company is executing a deliberate shift from being a connectivity supplier to becoming a foundational interconnect layer for the next generation of AI scaling. The proof is in the quarterly results and the ambitious outlook that follows.

Fiscal fourth-quarter revenue surged 22% year over year to $2.22 billion, with adjusted EPS hitting $0.79, slightly ahead of estimates. More importantly, the company is projecting an even steeper climb. For the full fiscal year, it now expects revenue to be "approaching $11 billion," up from prior guidance. The real signal for exponential growth, however, is the forward view: management projects overall revenue in fiscal 2028 to grow close to 40% year over year, reaching approximately $15 billion. That's a compound annual growth rate that aligns with the paradigm shift underway in AI data centers.

This aggressive 40% growth target for FY28 is not pulled from thin air. It is backed by a strategic context of design wins and hyperscaler partnerships. Marvell boasts design wins with more than 20 customers and is a key supplier to major U.S. hyperscalers. Its role in developing custom AI ASICs for Amazon and Microsoft places it at the heart of the next wave of AI hardware. The recent launch of industry-first CXL and PCIe 6.0 switches at OFC 2026 further cements this position, directly addressing the memory pooling and bandwidth bottlenecks that will define the scaling of larger AI models.

The bottom line is that Marvell is moving from a linear growth story to one of exponential adoption. Its raised guidance and strong design win pipeline signal that it is capturing a critical piece of the AI infrastructure stack. Yet, the 40% FY28 target remains a high bar. It requires that these design wins convert into revenue at a pace that validates the company's aggressive projections. For now, the evidence shows the company is on the right trajectory, but the steepness of the S-curve ahead will be determined by how quickly its customers scale.

Building the AI Data Center Plumbing: Technology and Adoption

Marvell's recent push isn't just about selling more chips; it's about building the fundamental plumbing for the next paradigm of AI scaling. The company is targeting the critical bottlenecks that will define the architecture of large AI clusters: memory pooling and interconnect bandwidth. Its strategy is now crystallized in two industry-first products launched at OFC 2026, derived directly from its $540 million acquisition of XConn Technologies.

The launch of the industry's first 260-lane PCIe 6.0 switch and the CXL 3.0 switch is a direct assault on the "memory wall." These switches, built on XConn's IP, offer roughly double the lane density of competing products. This is a game-changer for scale-up fabrics, eliminating the need for complex, costly cascading of multiple smaller switches. As Marvell's VP noted, this architecture is essential for breaking through the memory wall that limits the performance of larger AI models. The company is now positioning itself as a leader in the open fabric-switching portfolio for AI data centers, directly addressing the composable infrastructure demand driven by these bottlenecks.

This move is also a strategic play in the emerging battle for scale-up connectivity standards. Marvell is aligning its new switches with the UALink ecosystem, an open industry standard purpose-built for connecting large numbers of accelerators. The goal is clear: to provide a credible, open alternative to NVIDIA's proprietary NVLink. This positioning is critical, as hyperscaler deployment decisions will be the ultimate gating factor for UALink's market viability. Marvell's acquisition of XConn didn't just add products; it expanded its UALink scale-up switch team with top engineering talent, giving it the in-house expertise to compete in this high-stakes race.

The bottom line is that Marvell is constructing the infrastructure layer for the next exponential growth phase. By launching these foundational interconnect products and aggressively building its UALink capabilities, the company is moving beyond being a supplier to becoming a key architect of the AI data center's internal architecture. The success of this build-out will hinge on whether hyperscalers adopt its open standard and whether its design wins convert into the massive revenue scale needed to hit its 40% growth target. For now, the technological moat is being forged in silicon and open standards.

Financial Impact and Valuation: Leverage vs. Risk

The technological build-out is now translating into a powerful financial story. Marvell's raised guidance for fiscal 2028 implies significant operating leverage, with management projecting non-GAAP EPS to well over $5. That target, set against a projected revenue run-rate of roughly $15 billion, represents a massive step-up from current levels. It signals that the company is not just growing top-line revenue but also scaling its profitability at an accelerated pace, a hallmark of a business hitting the steep part of the S-curve.

This growth trajectory is reflected in the stock's valuation. Despite the recent rally, Marvell trades at a forward P/E of roughly 19x fiscal 2028 estimates. Analysts view this as "particularly attractive" versus broader AI semiconductor peers. The multiple reflects the market's recognition of the high-growth potential embedded in the FY28 outlook, which now sits well above Wall Street consensus. The upgrade from Bank of America to Buy, with a $110 price target, underscores this view, citing improved visibility and the company's solid leverage to AI optical connectivity.

Yet, this attractive valuation is a bet on flawless execution. The path to that $5 EPS target is fraught with risks. First, customer concentration remains a key vulnerability. The company's custom chip programs are deeply tied to Amazon and Microsoft, and recent speculation about shifting partnerships has created noise. The raised guidance must now be validated by sustained, large-scale orders from these hyperscalers.

Second, competitive pressure is intensifying. Marvell is building an open alternative to NVIDIA's proprietary NVLink, but it faces entrenched competition from Broadcom and Nvidia itself. The success of its new 260-lane switches and its UALink ecosystem depends entirely on hyperscaler adoption, a decision point that could be swayed by existing relationships or technical performance.

Finally, there is the execution risk of converting its design wins with more than 20 customers into the massive, sustained revenue needed to hit a 40% growth rate. The company's recent guidance raises have been steep, and the market is now pricing in that success. Any stumble in scaling these design wins or in maintaining hyperscaler relationships could quickly re-rate the stock.

The bottom line is that Marvell's valuation today captures the promise of exponential growth. It is priced for a company that is successfully building the AI infrastructure rails. But the stock's exposure to customer concentration, competitive battles, and execution risk means it is also priced for perfection. For investors, the setup is clear: the potential reward is high if the company navigates these risks, but the downside is significant if it does not.

Catalysts and Watchpoints: The Path to Exponential Growth

The bullish thesis for Marvell rests on a future of exponential growth. But that future is not guaranteed; it will be proven or challenged by a series of specific, measurable milestones in the quarters ahead. Investors must monitor three key catalysts to gauge whether the company is truly scaling the AI infrastructure S-curve.

First, the execution of the XConn acquisition is now a commercial imperative. The recent launch of the industry's first 260-lane PCIe 6.0 and CXL 3.0 switches at OFC 2026 was a technological victory, but its real test is in the market. The coming quarters will show if these new products are being adopted by hyperscalers and OEMs, and if the integration of XConn's engineering talent is accelerating the development of Marvell's UALink ecosystem. Any delay or technical hitch in this rollout would directly challenge the timeline for hitting the 40% growth target.

Second, the company's design win pipeline must convert into tangible revenue. Management boasts design wins with more than 20 customers, but the critical metric is the contribution from custom silicon and the new interconnect products. Subsequent earnings reports will need to detail the ramp of custom chip revenue and the uptake of the new switches. The market will be watching for signs that these products are moving beyond the lab and into the production lines of major AI data centers.

Finally, the trajectory of Marvell's hyperscaler partnerships is the ultimate gating factor. The company is a key supplier to Amazon and Microsoft, helping develop their Maia and Trainium chips. Any major announcement from these partners-whether it's a new chip launch, a shift in architecture, or a change in supplier strategy-will directly confirm or alter Marvell's role. The success of its open UALink standard, which aims to be a credible alternative to NVIDIA's NVLink, depends entirely on these hyperscaler deployment decisions.

The bottom line is that Marvell's exponential growth thesis is a bet on execution. The stock's valuation already prices in success. The near-term watchpoints-the commercial rollout of new switches, the conversion of design wins, and hyperscaler partnership health-are the checkpoints that will prove whether the company is building the infrastructure rails or simply building hope.