Marvell Technology raised its fiscal 2028 revenue outlook to roughly $20 billion on Tuesday, extending a series of increasingly aggressive growth targets as artificial-intelligence spending drives demand for the company’s custom chips and data-center connectivity products. The new forecast is about 10% above the $18.2 billion analysts had been expecting and helped send Marvell shares nearly 6% higher, adding to a rally that has more than tripled the stock’s value this year.
The bigger change came further out. At its investor day in New York, Marvell projected fiscal 2031 revenue of $70 billion to $90 billion, implying $80 billion at the midpoint. That figure is far beyond the roughly $46.9 billion consensus compiled by Visible Alpha and signals that management expects spending on custom AI processors, high-speed networking and other data-center infrastructure to remain elevated well beyond the current investment cycle.
Those numbers are forecasts rather than contracted revenue, and hitting them will depend on customers following through with multiyear AI infrastructure programs. Still, Marvell enters the expansion from a stronger operating position than it held a year ago: second-quarter fiscal 2027 revenue reached a record $2.74 billion, up 37% from the prior year, while data-center sales grew 46%. Management said in August that AI-related bookings remained exceptionally strong and guided third-quarter revenue to approximately $3.15 billion.
Custom Silicon Is Becoming a Much Larger Business
Marvell now expects its custom semiconductor business alone to generate about $12 billion in fiscal 2029, up from an earlier $10 billion target. That business has become one of the central pieces of the company’s AI strategy as large cloud operators increasingly develop processors tailored to their own workloads rather than relying exclusively on general-purpose accelerators supplied by companies such as Nvidia.
Marvell does not need to displace Nvidia for the strategy to work. Its opportunity extends across the broader infrastructure surrounding AI computing, including custom accelerators, networking interfaces, storage controllers, memory-related products and high-speed optical connectivity. As cloud companies build larger computing clusters, spending on those supporting components can rise alongside spending on the processors performing the AI calculations themselves.
The company’s recent expansion with Google illustrates both the scale of the opportunity and the need to separate potential business from committed sales. A July commercial agreement covers a range of custom semiconductor programs tied to Google’s TPU ecosystem, including AI inference accelerators, network interface controllers, storage controllers and memory technologies. Marvell subsequently issued Google warrants to purchase as many as 58.97 million Marvell shares at an exercise price of $206.58.
Most of those warrants vest according to purchases rather than time. The agreement divides the performance-based portion into 240 tranches, with one tranche vesting for each $500 million of qualifying product revenue through fiscal 2033. Multiplying those thresholds produces as much as $120 billion of qualifying revenue tied to the full vesting structure, but that should not be treated as a $120 billion firm order or backlog. Google’s purchases are discretionary, and the SEC filing does not guarantee that the maximum revenue threshold will ever be reached.
That distinction becomes especially important as investors assess Marvell’s new long-term targets. The Google relationship provides evidence that hyperscale customers are willing to expand custom silicon partnerships, but the magnitude and timing of the eventual revenue remain dependent on product execution, customer deployment decisions and AI infrastructure spending over several years.
A Faster Growth Curve Raises the Execution Bar
Marvell’s near-term results provide some support for management’s confidence. Revenue in the August quarter exceeded the midpoint of the company’s previous guidance by $39 million, while operating cash flow reached $605.5 million. Non-GAAP gross margin was 58.9%, and management expects revenue growth to accelerate through the remainder of fiscal 2027 as custom programs begin contributing more heavily.
The investment case is therefore shifting from whether Marvell can participate meaningfully in the AI infrastructure buildout to whether it can convert an expanding pipeline of custom programs into revenue at the pace now embedded in its forecasts. Reaching $20 billion in fiscal 2028 would require a substantial step up from the company’s current quarterly revenue base, while the fiscal 2031 outlook implies another dramatic expansion afterward.
Competition will remain significant. Broadcom has established a large custom AI semiconductor business of its own, while Nvidia continues to dominate the accelerator market and other chipmakers are pushing deeper into data-center computing. Qualcomm, for example, said earlier this year that it expects its data-center business to reach $15 billion in annual sales by 2029, reflecting how quickly semiconductor suppliers are repositioning around hyperscale AI spending.
Marvell shares gained 5.8% Tuesday as chip stocks helped push the S&P 500 and Nasdaq to record closing highs, while Broadcom rose about 4%. The broader semiconductor strength means Marvell’s move was not occurring in isolation, although the scale of the company’s new forecasts provided a company-specific catalyst.
The next test is whether the revenue acceleration already anticipated for the second half of fiscal 2027 begins showing up at the scale management expects. Marvell’s third-quarter guidance calls for approximately $3.15 billion in sales, and subsequent results should provide an early indication of how quickly custom silicon programs are ramping. Beyond that, investors will be watching whether the company can turn large customer relationships such as Google into recurring revenue without allowing the long-range opportunity embedded in its $70 billion-to-$90 billion fiscal 2031 forecast to get too far ahead of actual deployments.