Marvell Technology (MRVL) has rallied close to 80% year-to-date as the market rotated back toward AI infrastructure names.
The backdrop is clear. Big tech players, including Microsoft (MSFT), Alphabet (GOOG), Amazon (AMZN), and Meta (META) have collectively guided over $600 billion in AI capital spending in 2026 alone. That spending flows into compute and storage and increasingly into the more specialized compute and the plumbing that connects it all – custom chips and high-speed interconnects. These components are exactly the areas Marvell operates in.
Broadcom, which plays in similar areas, is now a $2 trillion company and has soared almost 10x over the last five years.
Could Marvell, too, see strong upside going forward?
To be fair, Broadcom has a large software business and years more entrenchment across hyperscaler relationships. But on the tech side, custom AI silicon and optical interconnects are core to both companies. Marvell’s market cap today sits at roughly $140 billion – about 7% of Broadcom’s. Overall, the setup is straightforward: a smaller company with a proven product in a market that is still in early innings, trading at a fraction of its closest comparable, and has been seeing strong momentum.
The Interconnect Advantage
As AI clusters scale to hundreds of thousands of processors, internal network speed becomes as critical as compute itself. Electrical signals work well over short distances, but at AI speeds they consume too much power, generate heat, and degrade over distance. The industry is shifting to optical connections, and Marvell is a key supplier behind that transition. Marvell’s optical DSPs hold a sizable share in high-speed segments like 800G modules. See Why Marvell’s Interconnects are a big deal. Its acquisition of Celestial AI positions it for the next step, moving optics closer to compute, which could reduce hyperscaler dependency on costly HBM4 memory. This segment alone is projected to grow over 50% in FY’27, and the data center switching business is targeting revenues above $600 million, roughly double FY’26 levels.
The biggest validation for Marvell’s interconnect business came as Nvidia made a $2 billion strategic investment in Marvell and confirmed it as a key supplier within NVLink, Nvidia’s proprietary interconnect fabric linking GPUs in large-scale clusters. Suppliers integrated into NVLink are designed into the product. They do not compete on the open market for each sale, and that provides a level of revenue visibility for Marvell.
Custom Silicon
Custom silicon revenue reached a $1.5 billion annual run rate last year – that’s under 20% of total revenue, but it is growing fast. The demand is driven by two factors. First, AI workloads are shifting from training to inference. Training a model happens once. Inference, which is the phase where the model responds to actual user queries, runs billions of times a day. At that volume, power efficiency becomes an operating cost problem, and a chip purpose-built for a specific workload consumes significantly less power than a general-purpose GPU doing the same job. Second, every hyperscaler that builds its own chip reduces its dependence on Nvidia, gaining leverage on GPU pricing and supply.
In April 2026, Google was reported to be in advanced negotiations with Marvell to co-develop a Memory Processing Unit and a next-generation Tensor Processing Unit for AI inference work Google had previously done almost exclusively with Broadcom. If finalized, Marvell becomes a lead design partner on one of the most strategically important chip programs in the industry.
Still, Broadcom leads the custom ASIC market and recorded roughly $20 billion in total AI revenue in fiscal 2025. Marvell trails considerably, but hyperscalers routinely qualify two suppliers for critical components to avoid single-source dependency.
How Marvell Stock Can Surge
Let’s run the numbers. Marvell generated about $8.2 billion in revenue in FY’26, and consensus points to about $11 billion in sales for FY’27, a growth of about 33%, rising another 36% to $15 billion by FY’28. However, if revenue actually picks up at a rate of about 50% annually over the next three years driven by stronger interconnect revenue and custom chip sales, then revenue could rise to almost $28 billion by 2029.
Adjusted net margins have been on an improving trajectory. See how Marvell margins compare with peers They stood at close to 30% in FY’26, up from about 24% in the FY’25 period, led by strong operating leverage and a favorable product mix. This trend could continue, taking margins to levels of about 37% in the coming years. For perspective, Nvidia has net margins of roughly 50% while Broadcom has net margins approaching 40%. At these margin levels, that could translate into about $10.3 billion in net income by 2029.
Now Marvell currently trades at levels of around 56x trailing earnings. If Marvell roughly quadruples its earnings between FY’26 and FY’29, the multiple would contract to levels of under 15 for the stock price to stay the same. But that is what Marvell investors are betting will not happen. If Marvell’s P/E stands at about 35x levels, roughly in line with the broader semiconductor industry, that would translate into a market valuation of over $360 billion, which is more than 2.5x current levels. []
Navigating the fast-growing, yet volatile AI infra sector requires balancing these high-conviction bets with a broader strategy anchored by mature cash generators. A smart portfolio helps you stay invested by limiting the impact of market shocks. While consistently beating the market is a challenge, the Trefis High Quality (HQ) Portfolio is designed to make it an achievable goal. The HQ strategy has consistently outperformed its market benchmark since inception, delivering returns of over 105 percent.
