Analysis
Marvell Technology chairman and CEO Matt Murphy told investors at Citi's 2026 Global TMT Conference this week that the company now expects roughly $12 billion in AI data-center revenue this fiscal year and about $18 billion next year. That fiscal 2028 figure is a meaningful raise from Marvell's prior target near $16.5 billion.
The stock jumped as much as 6.4% on the update, one of the sharpest single-day moves for a chip name that has already climbed 166% year-to-date and 238% over the past twelve months, according to Benzinga's coverage of the conference remarks.
The timing is what makes this newsworthy for VCs and founders tracking the AI infrastructure cycle, not just chip investors. Marvell beat Wall Street's estimates on its fiscal second-quarter earnings in late August and guided to roughly 50% annual growth -- and the stock fell anyway, down about 6% that day, on worries that hyperscaler capex commitments were getting harder to read and that a single large customer's order timing could swing results. Pulse covered that selloff at the time. Three weeks later, a materially higher guide from the same management team produced the opposite reaction -- a sign that the market's read on AI infrastructure demand is shifting fast in both directions on the same underlying data.
“Pulse covered that selloff at the time.”
A two-horse race for hyperscaler silicon
Marvell doesn't compete with Nvidia's merchant GPUs directly. It designs custom AI accelerators (ASICs) for hyperscalers who want their own silicon instead of, or alongside, Nvidia chips, and the market splits into two camps:
- Broadcom -- the dominant player, holding roughly 60-70% of the custom-ASIC co-design market and design wins with Google's TPU, Meta's MTIA and OpenAI's Titan program. Its AI semiconductor revenue jumped 143% year-over-year last quarter to $10.8 billion, per Motley Fool's analysis.
- Marvell -- the clear #2, at an estimated 20-25% share, with confirmed design wins at Amazon and Microsoft per Murphy's remarks. Combined, the two companies control roughly 95% of the custom-ASIC market between them.
Broadcom's current AI revenue run-rate is already well above Marvell's newly raised FY28 target, a reminder of how much smaller a share of this market Marvell is fighting for even as its own numbers climb.
Murphy also pointed to Marvell's business beyond pure compute ASICs -- optics, switching, analog and custom silicon are all accelerating together, he said, which the company frames as a broader, multi-engine AI infrastructure franchise rather than a single-customer bet. That diversification argument is Marvell's answer to the criticism that dogged the August selloff: that too much of its growth depends on the timing of one or two hyperscaler contracts.
The bear case hasn't gone away. Marvell's own August earnings call proved that a beat-and-raise quarter can still sell off if investors doubt the durability of hyperscaler orders, and nothing about this week's conference remarks is a contractual commitment -- it's guidance, delivered verbally in a fireside chat with CNBC's Jim Cramer, not a filed 8-K with binding numbers. Hyperscalers have shown they will re-shuffle chip orders between Nvidia, Broadcom, Marvell and their own in-house silicon teams as pricing and performance shift, and a guidance raise today does not guarantee the order book looks the same next quarter.
For founders building AI infrastructure or chip startups, Marvell's guide is one of the cleanest available proxies for whether hyperscaler AI capex is actually accelerating or just being reallocated among incumbents -- a distinction that matters for anyone pitching a comparable in a term sheet built off "look how fast this market is growing." Watch Amazon's and Microsoft's own capex guidance next quarter, and whether Broadcom's fiscal Q4 report (due later this year) shows the same reacceleration Marvell just described.
