The majority of investors followed the obvious names in the battle to position portfolios for AI infrastructure spending. The main focus was Nvidia. Then came the hyperscalers themselves, revealing projections for capital expenditures totaling more than $600 billion for just 2026. However, Marvell Technology was quietly developing one of the more intriguing positions in the semiconductor business somewhere in the supply chain behind those headline numbers, and the stock has spent April belatedly reflecting that.
Even by the norms of a sector that has delivered remarkable advances, the numbers are startling. In April, MRVL had its best winning run in over a year, gaining 22 percent over five straight sessions. In a market where the S&P 500 is essentially unchanged, it has increased by more than 50% year to date through mid-April. Depending on the date, the stock has increased by between 143 and 151 percent during the last 12 months. However, outside of specialized circles, Marvell is still frequently referred to as a secondary name in AI, a business that gains indirectly rather than being at the center of the development. It is becoming more and more difficult to maintain such framing.
Important Information
| Field | Details |
|---|---|
| Company | Marvell Technology, Inc. (NASDAQ: MRVL) — founded 1995 by Wei Li Dai and Pantas Sutardja; headquartered in Wilmington, Delaware; operations including major engineering centers in Santa Clara, California; approximately 6,000+ employees; fabless semiconductor design company |
| Share Price (April 2026) | Approximately $128–$134; 52-week range roughly $52–$138; up more than 50% year-to-date in 2026; up approximately 143–151% over the past 12 months; market capitalisation approximately $110–$118 billion; 22% gain over five consecutive trading sessions in April — the stock’s best winning streak in over a year |
| Q4 FY2026 Results (March 5, 2026) | Revenue $2.22 billion — a record quarterly high; up more than 20% year-over-year; Data Center segment approximately 75% of total revenue; Data Center revenue growing approximately 42% year-over-year |
| Custom AI Chips (ASICs) | Marvell produces custom AI application-specific integrated circuits (ASICs) for Amazon AWS; on track to add Microsoft as an ASIC client in the second half of 2026; ASIC revenue expected to reach $4 billion in FY2027 and over $10 billion by 2028; Oppenheimer projects total company revenue exceeding $11 billion this year, up from approximately $9 billion six months ago |
| Nvidia Investment | In late March 2026, Nvidia invested $2 billion in Marvell to collaborate on silicon photonics technology — a critical component for ultra-high-speed data transfer within AI data centres; seen as strategic validation of Marvell’s optical interconnect expertise |
| Key Analyst Actions (April 2026) | Oppenheimer: raised target to $170 (from $150), Outperform — April 15; Barclays: upgraded to Overweight, raised target to $150 — April 9; Stifel: maintained Buy, raised target to $140 (from $120) — April 16; 36 of 43 covering analysts rate MRVL Buy or higher |
| Celestial AI Acquisition | $3.25 billion acquisition of Celestial AI announced December 2025 — AI startup specialising in optical interconnect technology for data centres; accelerates Marvell’s silicon photonics strategy |
Custom silicon is the central idea of the thesis. In order to create their own AI accelerators and lessen their reliance on Nvidia’s off-the-shelf GPUs, cloud providers like Amazon, Google, and Microsoft require partners to fabricate and engineer the underlying processors. Application-specific integrated circuits, or ASICs, are specialized processors made for a particular job as opposed to general processing.
In its most recent fiscal year, Broadcom generated almost $20 billion in AI sales, making it the market leader. The recognized second player is Marvell. Being the qualified alternative to Broadcom is also a structurally important position in a market where hyperscalers regularly approve two suppliers for essential components to reduce single-source risk.
The established anchor is the partnership with Amazon. Through several iterations of the Trainium chip initiative, Marvell has been manufacturing specialized AI chips for Amazon Web Services. The Microsoft partnership is more recent and, should it come to pass as anticipated in the second half of 2026, would signify a significant increase in the clientele.
Marvell executives presented ASIC sales forecasts of $4 billion for next year—roughly twice the current level—and a longer-term trajectory of $10 billion or more by 2028 at the Oppenheimer investor meeting in Europe prior to the goal raise to $170. In response, Oppenheimer analyst Rick Schafer referred to the company as “the Switzerland of interconnect,” a term that seems to be intended to convey both its pivotal role in AI data center networking and its relatively neutral relationship with other hyperscalers.
Nvidia, a business that never makes public investments in possible competitors or suppliers unless the strategic rationale is strong, invested $2 billion in silicon photonics-focused Marvell in late March. As AI model training and inference workloads require quicker data mobility within data centers, this technology—which enables optical signals to transfer data at speeds that copper interconnects cannot match—becomes more and more crucial.
Through its own research and development as well as the $3.25 billion acquisition of Celestial AI, which was announced in December 2025, Marvell had been working toward this position. Nvidia’s financing served as outside confirmation that the effort is important.
It’s difficult to ignore how genuinely complex the value discussion surrounding MRVL is. The stock appears pricey based on trailing earnings. The picture significantly shifts on forward earnings that anticipate the ASIC revenue ramp, especially since design victories in custom silicon, once achieved, typically lock in income for several years before a rival can qualify through the same customer’s supply chain. Execution is the risk. Concentration of customers is still a serious issue. Furthermore, there is little margin for error given the discrepancy between current revenue and the goals that support the projected multiple.
The question of whether the outlook and design-win pace reflect the narrative that the stock has been pricing in all spring will be the next test, which is the quarterly earnings release.
