Marvell Technology (MRVL) has more than doubled year-to-date, with the stock surging nearly 109%, an extraordinary rally that naturally raises the question of whether investors are already too late. However, I do not think so. In my view, the market is only beginning to appreciate how important the semiconductor company could become in the next phase of the artificial intelligence (AI) infrastructure buildout.
Claim 55% Off TipRanks
Trade AMZN with leverageYes, the stock is no longer cheap, and the near-term upside looks less obvious after such a powerful run. Still, I remain bullish because Marvell sits at the intersection of two AI infrastructure bottlenecks that are only becoming more important: custom AI silicon and high-speed optical connectivity.
Marvell Is Becoming an AI Infrastructure Company
Marvell is no longer just a broad data-infrastructure semiconductor name. Increasingly, it looks like a specialized AI infrastructure supplier. The company has built a strong position in optical Digital Signal Processors (DSPs), custom XPUs, Ethernet switching, Data Processing Units (DPUs), SmartNICs, and silicon photonics. That may sound technical, but the investment point is simple: as AI clusters get larger, the challenge is not only compute. It is also moving data quickly, efficiently, and reliably between chips, racks, and data centers. That is where Marvell’s portfolio matters.
Data centers are moving from 800G to 1.6T optical connectivity, and that upgrade cycle should be a major tailwind. I estimate that optical already represents more than a third of Marvell’s sales and expect strong sequential growth. Management previously guided interconnect revenue to grow more than 50% this year, and I think that demand may be tracking even better.
In my opinion, this is the cleanest part of the bull case. AI workloads are getting bigger, and the networking layer is becoming more strategic. Marvell is one of the few companies with the product breadth to benefit across DSPs, TIAs, drivers, coherent optics, CPO, and related connectivity technologies.
AWS and Microsoft Could Drive the Next Leg
The second pillar of the thesis is custom silicon. Marvell appears to remain a key supplier to Amazon’s (AMZN) AWS Trainium platform, and recent Amazon Web Services partnerships with Anthropic and OpenAI improve long-term visibility. Anthropic’s expanded agreement with AWS includes up to 5 gigawatts of new capacity, spanning Trainium2 through Trainium4. That is a big deal for Marvell because it supplies important components tied to Trainium, including custom ASICs, optical DSPs, switches, and DPUs.
Near-term upside may be limited by tight wafer supply, especially in 3nm. That is a real constraint, and I would not ignore it. However, this is more of a timing issue than a demand issue. If anything, the tight supply supports the idea that demand exceeds current capacity.
Microsoft (MSFT) is another important part of the story. Marvell’s Microsoft XPU opportunity appears on track for an early calendar 2027 ramp, and management’s revenue assumptions of roughly $600 million may prove conservative if the ramp develops as expected. Put differently, I do not think Marvell needs every AI rumor to be true for the stock to work. Together, AWS, Microsoft, and optical growth already create a strong multi-year setup.
Nvidia’s Investment Was a Major Validation
The Nvidia (NVDA) partnership is not just a headline. It matters. Nvidia’s $2 billion investment in Marvell validates the company’s role in AI networking and custom silicon. The companies are working together across NVLink Fusion, silicon photonics, and AI telecom opportunities.
This strengthens Marvell’s position in AI interconnects. It can participate across multiple architectures and protocols, including NVLink, UALink, Ethernet, optical, and copper. That flexibility is valuable because no one knows exactly which interconnect standard will dominate every layer of AI infrastructure.
I like that positioning. Marvell does not need to make a single binary bet. It can sell into the buildout regardless of whether hyperscalers lean more heavily toward Nvidia ecosystems, custom silicon, or mixed environments.
The Valuation Is Rich, but Not Irrational
This is where I want to be careful. After a 109% year-to-date rally, valuation risk is obvious. Marvell trades at a forward P/E ratio of about 42.2x, above the IT sector median of near 24.2x. Its price-to-operating-cash-flow ratio of about 104x is also far above the sector median of roughly 18x. Those numbers are not easy to dismiss.
However, traditional metrics may not fully capture the slope of Marvell’s AI revenue growth. I expect top-line momentum to continue for the next two to three years, with custom AI application-specific integrated circuit (ASIC) revenues rising from roughly $2 billion this year to around $4 billion in 2027. Optical growth adds another meaningful layer.
So yes, the stock is expensive. Yet I would rather own an expensive company with improving visibility into a large AI infrastructure cycle than a cheap semiconductor name with no clear growth engine. The biggest risk, in my view, is not that Marvell lacks demand. It is that expectations have moved up very quickly. Any stumble in supply, customer ramps, or AI capex sentiment could pressure the stock.
Wall Street’s View
According to TipRanks, Marvell carries a Strong Buy consensus rating, with 22 Buy, four Hold, and no Sell ratings. Based on 26 Wall Street analysts, the average price target is $146.05, implying about 17% downside from the latest price of $176.89.
Conclusion
Marvell’s rally has been huge. However, I still see a strong long-term bull case. Marvell has become a critical supplier in AI optical connectivity, custom silicon, and next-generation interconnect architecture. AWS and Microsoft provide major custom XPU opportunities, while Nvidia’s investment reinforces Marvell’s strategic relevance across the AI ecosystem.
The stock may need to digest its gains, and valuation makes the path bumpier from here. Still, for investors focused on the next phase of AI infrastructure, I remain bullish on MRVL.