The Next $1 Trillion AI Winner May Already Be Hiding Inside Nvidia's Supply Chain

Marvell is forecast to grow revenue at a 41% CAGR and adjusted EBITDA at a 43% CAGR from fiscal 2026 to fiscal 2029 as AI-driven data center upgrades accelerate. Nvidia’s CEO has publicly called Marvell a potential trillion-dollar company and Nvidia has invested $2 billion in the firm, reinforcing the strategic importance of NVLink Fusion and related optical interconnect, silicon photonics, and networking products. The article is fundamentally bullish on Marvell’s long-term growth, though it notes the stock already trades at 24x revenue and 63x adjusted EBITDA.

Analysis

The important read-through is that MRVL is not being valued as a standalone chip vendor anymore; it is increasingly being treated as a “picks-and-shovels multiplier” on the AI buildout. That matters because the market typically assigns much higher terminal multiples to infrastructure layer names that sit inside multiple OEM and hyperscaler designs, especially when a platform leader like NVDA publicly validates the roadmap. The second-order effect is that MRVL’s addressable demand becomes less about unit growth in any one silicon product and more about the duration of capex intensity across the cloud ecosystem.

The bigger winner may actually be the broader optical interconnect and custom silicon supply chain. If NVDA is effectively outsourcing parts of the networking bottleneck to MRVL, then the constraint shifts from GPU supply to system-level bandwidth and power efficiency; that benefits the companies that solve rack-to-rack and chip-to-chip data movement. It also raises competitive pressure on legacy networking incumbents and merchant-switch vendors, because the value pool migrates toward tightly co-designed, custom, hyperscaler-specific hardware.

The main risk is that the market is extrapolating a linear ramp in AI infrastructure spend into the outer years, when these programs tend to be lumpy and procurement-sensitive. MRVL’s multiple already discounts several years of strong execution, so any pause in hyperscaler capex, product timing slip, or margin dilution from customization could compress the stock sharply even if growth remains positive. On a 3- to 12-month horizon, the stock is more vulnerable to guide-down risk than to demand collapse.

The contrarian point is that NVDA’s endorsement may be creating a halo effect that overstates how much economic value MRVL can actually capture versus its platform partner. If NVDA continues to integrate more functionality vertically, MRVL could become strategically important but economically capped, with bargaining power concentrated in the customer relationship rather than the component supplier. That makes this a good name to own on dips, but a poor candidate for a blind momentum chase after a further rerating.

Key Decisions for Investors

• Long MRVL on a 3-6 month horizon, but only on pullbacks: add if the stock de-risks 10-15% from current levels or if the market sells off semis broadly; risk/reward improves materially once the multiple compresses below peak growth expectations.
• Pair trade: long MRVL / short a legacy networking or switching name with weaker AI exposure over 6-12 months; the trade expresses the view that custom AI interconnect will take share from merchant networking as hyperscalers optimize for power and bandwidth.
• Buy NVDA call spreads instead of outright MRVL if the goal is AI infrastructure beta with lower execution risk; NVDA has more control over the ecosystem economics, while MRVL carries higher customer-concentration and design-win timing risk.
• For event-driven traders, fade strength into earnings if consensus already implies 40%+ revenue CAGR: use a defined-risk put spread on MRVL into the next print to express concern that expectations are ahead of shipment timing.