Marvell Technology used its October 6 Investor Day to make a much bigger claim about where the AI infrastructure cycle is heading: the company now expects roughly $20 billion in fiscal 2028 revenue, up from its prior outlook, as custom AI silicon and data-center connectivity become larger parts of its business.
Reuters reported that the new fiscal 2028 target is above Wall Street expectations and that Marvell also lifted its custom-chip revenue goal for fiscal 2029 to $12 billion, up from a previous $10 billion target. The same report said Marvell sees a much larger long-range opportunity, forecasting fiscal 2031 revenue of roughly $70 billion to $90 billion.
Custom AI Silicon Is Becoming a Core Growth Engine
Marvell has spent the last several years shifting away from being viewed mainly as a networking and storage chip supplier. Its newer strategy combines custom compute silicon, high-speed SerDes, optical DSPs, advanced packaging and other data-center interconnect technologies into a broader platform for hyperscale AI infrastructure.
That shift is already visible in the hardware. BitcoinVersus.Tech recently covered the company’s 102.4 Tbps Teralynx T100 Ethernet switch, which targets the enormous east-west traffic created when thousands of accelerators communicate inside AI clusters.
Marvell is also pushing deeper into custom accelerators and XPUs designed around the requirements of individual cloud operators. That business can include compute dies, memory interfaces, chiplets, SerDes and packaging rather than a single off-the-shelf processor.
The Interconnect Business Matters Just as Much as Compute
AI systems are limited not only by how quickly a processor can calculate, but by how fast data can move between accelerators, memory and racks. Marvell is positioning itself on both sides of that problem.
Its recent 2 nm optical roadmap for 3.2T AI networking shows how aggressively bandwidth requirements are scaling. Faster optical links, lower-power electrical interfaces and co-packaged technologies become more valuable as AI clusters move toward hundreds of thousands or even millions of tightly connected compute devices.
This is why the Investor Day outlook is broader than a simple custom-chip forecast. Marvell is trying to capture spending on the compute silicon itself and on the communications fabric required to keep those chips useful.
Marvell Is Scaling Manufacturing Partnerships Too
Higher custom-silicon volumes also require foundry capacity, packaging and supply-chain execution. BitcoinVersus.Tech previously reported on Marvell and GlobalFoundries expanding U.S. capacity for AI-networking silicon, another sign that the company is preparing for larger infrastructure demand rather than a short product cycle.
Marvell’s latest reported quarter supports that direction. Fiscal second-quarter 2027 revenue was $2.739 billion, up 37% year over year, while data-center revenue growth accelerated to 46% year over year. Management said AI-related bookings remained exceptionally strong and that custom revenue was expected to accelerate in the second half of fiscal 2027.
The Google Opportunity Raises the Ceiling
One of the largest potential contributors is Marvell’s expanded relationship with Google. Reuters said the agreement announced earlier this year could produce as much as $120 billion in sales through 2033 if specified performance targets are achieved. That figure is not guaranteed revenue, but it illustrates how large custom accelerator programs can become when they are tied to hyperscale cloud deployments.
The opportunity also carries execution risk. Custom silicon programs can take years to design, qualify and ramp, and large customers have enormous bargaining power. Marvell therefore needs its connectivity, optics and packaging businesses to keep scaling alongside custom compute rather than relying on one large design win.
The Bigger Point
Marvell’s new targets suggest that the AI semiconductor market is broadening beyond GPUs. Hyperscalers increasingly want silicon tailored to their own workloads, but those custom accelerators still need memory, switching, optics and high-speed electrical links around them.
At its October 6 Investor Day, Marvell framed those layers as one connected opportunity. If the company executes on the roadmap, it will be competing not just for individual chip sockets, but for a larger share of the full AI data-center silicon stack.
BitcoinVersus.Tech
Advertisement
Editor’s Note
Long-range corporate revenue targets are management forecasts, not guaranteed results. This story distinguishes reported historical results from forward-looking company projections.
We volunteer daily to help ensure the credibility of the information on this platform is Verifiably True. If you would like to support our research initiatives, please donate here: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb
BitcoinVersus.tech is not a financial advisor. Content is provided for informational purposes.
Leave a comment