Marvell Technology reported record quarterly revenue on Aug. 28, beating analyst estimates, then watched its shares fall 7.6% in after-hours trading as investors focused on a single line in the outlook: revenue from a custom AI chip built for Google will now be recognized later than expected, in fiscal 2029.
Net revenue for the second quarter of Marvell’s fiscal 2027 rose 37% from a year earlier to $2.74 billion, a company record and above the $2.71 billion analysts had forecast. Adjusted earnings came to 94 cents a share, up from 67 cents a year earlier and ahead of the 93-cent consensus. The company guided third-quarter adjusted earnings to between $1.05 and $1.15 a share.
Marvell also raised its full-year revenue outlook to about $12 billion, a sign that demand for its networking chips and custom silicon remains strong. The stock’s slide shows how far expectations have run: even a beat on revenue, earnings and guidance was not enough to offset concerns about when the Google program will contribute to the income statement.
The Google relationship is one of the most closely watched in the semiconductor industry. The two companies are developing custom AI accelerators, part of Google’s effort to reduce its dependence on Nvidia for the chips that power its data centers. Marvell has said the program will be a significant revenue driver, and investors had assumed the contribution would begin sooner.
Pushing recognition to fiscal 2029 effectively moves the payoff from the Google program more than a year later than some analysts expected. For a company trading at a premium valuation on the strength of that pipeline, the timing change matters as much as the direction of the business, said analysts at several firms, who trimmed near-term estimates after the report.
The rest of the quarter was broadly strong. Data-center revenue, Marvell’s largest segment, grew sharply year over year, driven by demand for optical components and custom compute used in AI infrastructure. The company’s traditional businesses, including carrier, enterprise networking and automotive, stabilized after a long downturn, executives said on the earnings call.
Marvell’s position in AI hardware has improved steadily over the past two years. Beyond Google, the company designs custom chips for a range of cloud and enterprise customers, and its optical products are used in the high-speed interconnects that tie AI servers together. Executives said bookings for AI-related products grew again in the quarter, and that the order book extends well into next year.
The company’s guidance for the current quarter implies revenue growth continuing at a pace near the high end of its longer-term targets. Chief Executive Matt Murphy said demand from cloud customers remained strong and that the company saw no signs of a slowdown in AI spending, even as some investors worry about the durability of the buildout.
The tension in Marvell’s report mirrors a broader debate in the market. Chip suppliers are reporting record orders, while their customers, the cloud giants, are signaling that the pace of AI capital spending will eventually need to slow. Marvell’s custom-chip business sits directly at the intersection of those forces, which makes its revenue timing unusually consequential.
The Google timing shift is not necessarily bad news for the company’s fundamentals. Deferring revenue from a large customer can reflect the schedule of a product launch, the configuration of a chip generation, or the customer’s own deployment plans, and Marvell executives said the underlying demand was unchanged. But for a stock that has re-rated higher on the promise of the AI cycle, a later payoff changes the arithmetic.
Wall Street’s reaction was blunt. In the hours after the report, analysts cut price targets and earnings estimates, and the stock gave back the gains of the session. Some noted that the reaction said as much about positioning as about fundamentals: with expectations so high, any wrinkle in the AI story gets punished.
Marvell’s path through the AI hardware boom is more varied than the Google headlines suggest. The company emerged from a decade of deal-making, including the purchases of Cavium, Innovium and Inphi, with a portfolio that spans networking, storage and optical technology. That history matters now because it gave Marvell something rare among chip designers: products that touch nearly every layer of the AI data center, from the accelerators themselves to the switches and optical modules that connect them.
Competition, though, is intensifying. Broadcom, Marvell’s chief rival in custom silicon, has locked in its own roster of hyperscale customers, and Nvidia has signaled it is willing to build custom chips for cloud providers who prefer alternatives to its flagship accelerators. Analysts said Marvell’s win rate in new custom-chip programs, and its ability to convert design wins into revenue on schedule, will determine whether its AI story holds together.
Marvell’s challenge now is to convince investors that its growth is durable rather than dependent on a single customer’s schedule. The company has diversified its custom-silicon business across multiple cloud providers, and its networking and optical franchises give it exposure beyond accelerators. But the market’s message on Aug. 28 was clear: at these valuations, patience is a scarce commodity.
