Marvell sees strong 2028 revenue from AI data centers, shares jump

Marvell Technology, a US chipmaker, forecast fiscal 2028 revenue near US$15 billion on rising demand for AI data-center chips and interconnects, and its shares rose about 15% in after-hours trading.

The company now expects fiscal 2027 revenue to rise more than 30% year-on-year to nearly US$11 billion, up from prior guidance of about US$10 billion.

It guided Q1 revenue of US$2.4 billion plus or minus 5%, above consensus of US$2.3 billion, and said the forecast includes Celestial AI and XConn Technologies.

Data-center revenue rose about 21% in Q4, and custom chips account for roughly 10% to 15% of sales, the company’s president and COO said.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Beyond the headlines, Marvell’s growth relies on specific, high-stakes bets

The data center segment anchors Marvell, with over $6 billion in fiscal 2026 revenue, up 46% year over year 1.
In the fourth quarter, that business brought in about 74% of total revenue, raising the stakes for the company’s next phase 1.
Revenue goals lean on the custom chips unit. It helps hyperscalers and cloud providers design silicon. It also lines up manufacturing, with programs widely believed to include Amazon Web Services’ Trainium2 AI chips plus Google’s Axion central processing unit (CPU) 2.
Marvell forecasts its custom business will grow more than 20% in fiscal 2027. It expects custom revenue to at least double year over year in fiscal 2028 1.
The Celestial AI acquisition aims to speed up data transfer between chips in large AI systems. Celestial AI’s optical technology is expected to reach a $500 million annualized revenue run rate in the fourth quarter of fiscal 2028 3.

Marvell’s rise signals a power shift in the chip industry

Marvell’s results track the shift toward custom chips, as tech giants build in-house silicon for AI and reduce reliance on Nvidia 2.
The move also expands demand for partners that handle chip design plus production support. It splinters the market, leaving firms like Marvell in the middle of big tech buildouts.
More attention on optical interconnects (high-speed links between chips and servers) signals a shift in constraints. Data movement is increasingly the limiter, more than raw computation 3.
As AI models scale, more budget flows to the high-speed connections that tie thousands of chips together. Interconnect specialists gain leverage in the next wave of data center construction.