Marvell’s Investor Day Reset: Why a $70–$90 Billion FY2031 Revenue Target Changes the AI Custom-Silicon Debate

The important change is not one quarter — it is the scale of the long-term model

Marvell Technology’s October 6 Investor Day did more than lift a near-term forecast. It gave investors a much larger framework for what the company believes it can become if custom AI silicon, high-speed interconnects and cloud networking continue to scale with hyperscaler spending.

The headline numbers were unusually aggressive. Marvell lifted its fiscal 2028 revenue target to about $20 billion from the $18 billion outlook it had provided in August. It also introduced a fiscal 2031 revenue range of $70 billion to $90 billion. Against fiscal 2026 revenue of roughly $8.2 billion, that long-range target implies that management expects the company’s revenue base to change dramatically over the next five years.

That is why the stock’s reaction mattered. Marvell shares rose about 6% on October 6, while Broadcom also gained as investors treated Marvell’s targets as a read-through for the broader market for custom accelerators, networking silicon and optical connectivity. The move was not simply enthusiasm about one company. It reflected a wider belief that hyperscalers are becoming large enough to support multiple specialized silicon vendors alongside Nvidia.

What Marvell actually changed

The most concrete revision was the fiscal 2028 revenue target. Management now expects about $20 billion, up from the $18 billion target issued only weeks earlier. The company also said it expects to reach the $18 billion data-center revenue target it had previously associated with fiscal 2029 almost a year earlier.

Custom silicon is central to the upgrade. Marvell said it now expects fiscal 2029 custom revenue to exceed $12 billion, above its prior target of more than $10 billion. The company has emphasized that it is shipping custom silicon to all four major U.S. hyperscalers, giving it exposure to a structural shift in which cloud operators increasingly design their own accelerators and surrounding infrastructure rather than relying exclusively on merchant chips.

The other half of the story is connectivity. AI clusters require enormous bandwidth between accelerators, memory, storage and network switches. As cluster sizes grow, the value of optical links, digital signal processors, switch silicon and other interconnect products can grow faster than the number of accelerators themselves. Marvell’s long-term model therefore rests on both compute customization and the infrastructure that connects that compute.

Why the $70–$90 billion target is such a big deal

A long-range forecast this large forces investors to ask a different question. The debate is no longer whether Marvell can participate in AI spending. It is whether the company can become one of the core platform suppliers to hyperscale infrastructure.

At the midpoint of $80 billion, fiscal 2031 revenue would be almost ten times fiscal 2026 revenue. That is a much more demanding trajectory than a normal semiconductor cycle. It requires multiple design wins to convert into high-volume production, customers to keep expanding capital budgets, and Marvell to maintain share in both custom silicon and networking.

The opportunity is credible because the economics of AI infrastructure are changing. Hyperscalers do not want to pay a premium for every workload if they can design silicon optimized for their own models, inference patterns or internal services. Custom chips can improve performance per watt and reduce dependence on a single merchant supplier. That gives vendors such as Marvell and Broadcom a role that is different from Nvidia’s but still highly valuable.

Marvell is also trying to reduce the market’s tendency to treat it as a single-customer or single-program story. A diversified set of hyperscaler programs can make revenue more durable, but it also introduces execution complexity because each program has its own schedule, qualification process and production curve.

Market impact: the read-through reaches beyond MRVL

The October 6 reaction in Broadcom was important because it showed that investors saw Marvell’s targets as sector evidence rather than a zero-sum share shift. If hyperscaler custom-silicon budgets are large enough, more than one supplier can grow rapidly.

That matters for optical and networking names as well. A custom accelerator still needs memory, switch capacity, optics, power delivery and cooling. As a result, the most important signal from Marvell’s Investor Day may be the implied size of the supporting infrastructure market rather than the accelerator revenue alone.

For Nvidia, the message is more nuanced. Custom silicon does not automatically mean Nvidia loses the AI market. Nvidia’s software ecosystem, merchant GPU scale and rapid product cadence remain major advantages. But a larger custom-silicon market can cap the share of future workloads that default to Nvidia hardware, especially at hyperscalers with the engineering talent and scale to optimize their own designs.

The market debate: credible design wins versus very distant targets

The bullish interpretation is straightforward. Marvell has already delivered record fiscal 2026 revenue, its data-center business has been growing rapidly, and management says the long-term model is based on programs already in development rather than purely hypothetical market share.

The skeptical interpretation is equally important. Fiscal 2031 is far enough away that small changes in customer schedules, AI spending growth or competitive share can produce enormous differences in final revenue. Semiconductor forecasts also have a history of looking strongest near the peak of capital-expenditure enthusiasm.

Another risk is customer concentration. Even with multiple hyperscalers, a handful of buyers can account for a very large share of custom silicon demand. Those buyers are technically sophisticated, price-sensitive and capable of changing architecture decisions. A delayed program or a customer insourcing more design work could create a sharp revenue gap.

Valuation is another constraint. A stock can be correct about the long-term story and still underperform if investors pay too much for that story in advance. After a powerful 2026 rally, the burden of proof shifts toward execution.

What to watch next

The next important checkpoint is not simply the next quarter’s EPS. Investors should track how much of the fiscal 2028 target becomes tied to identifiable production programs, whether custom-silicon revenue remains on schedule to exceed $12 billion in fiscal 2029, and whether interconnect revenue grows in line with accelerator deployments.

Customer diversification is equally important. Evidence that multiple hyperscalers are moving from engineering samples into high-volume production would make the 2031 framework more credible. Investors should also watch gross margin, because rapid custom-silicon growth can carry a different margin profile from higher-value connectivity products.

Finally, watch the competitive response from Broadcom and Nvidia. Marvell’s new targets raise the implied size of the addressable market, but they also raise the level of execution required to defend share.

Conclusion

Marvell’s October 6 Investor Day changed the conversation around MRVL. The company is no longer asking investors to value it as a semiconductor vendor with a fast-growing AI segment. It is asking the market to believe it can become a much larger infrastructure supplier built around custom compute and high-speed connectivity.

The $20 billion fiscal 2028 target is the near-term test. The $70–$90 billion fiscal 2031 range is the long-term option value. Between those two numbers lies the real investment question: how much of the hyperscaler AI buildout will convert into durable, diversified, high-margin revenue for Marvell rather than temporary program-driven growth?