Marvell Stock (MRVL) plunges nearly 30% in one month: has the AI infrastructure bull market ended?

Over the past month, the AI semiconductor sector has gone through a notable pullback. As a representative company in the AI networking chip space, Marvell Technology (MRVL) saw its share price fall sharply from the intra-year high in mid-June. On July 13 Beijing time, Marvell’s US stock closed at $235.81, down about 3.07% on the day. Compared with the 52-week high of $329.88 hit intraday on June 18, the decline across the range is now nearly 30%.

This is not just a problem with Marvell alone. During the same period, AI-related semiconductor names such as Broadcom (AVGO), Micron (MU), and AMD also faced varying degrees of selling pressure. The market is asking a core question: has the growth story of AI infrastructure already been fully priced in? When valuation expansion runs into the threshold of performance verification, how will the market recalibrate expectations? By placing Marvell’s stock performance within a broader valuation cycle for the AI industry chain, we analyze the nature of the current adjustment and potential paths for its evolution next.

AI semiconductor sector retreats across the board: profit-taking or loosening of the underlying logic?

Marvell’s drop this time is not triggered by a single event, but the result of multiple pressures stacking together.

Valuations are the biggest structural pressure. As of the close on July 10, Marvell’s trailing price-to-earnings ratio was about 81.31x, with a market cap of about $211.6 billion. Although the fundamentals have continued to improve—FY2027 Q1 revenue of $2.42 billion, up 28% year over year, and free cash flow of $483 million, up 127% year over year—the market clearly believes the current valuation already reflects fairly optimistic future growth expectations. When any marginal negative information appears, high-multiple stocks tend to experience larger swings.

Scrutiny of the AI capital expenditure return cycle is deepening. Over the past two years, hyperscalers have kept expanding capital spending on AI infrastructure, directly boosting demand for components such as GPUs, networking chips, and storage. But entering the second half of 2026, investors have started asking: when will these massive outlays generate revenue growth that matches them? Have AI data center utilization rates reached expectations? Is there a risk of oversupply? There are currently no definitive answers, but uncertainty alone is enough to trigger valuation adjustments.

Geopolitical risk is intensifying, shrinking risk appetite. On July 13 Beijing time, the US launched a new round of strikes against Iran, followed by Iran launching a large-scale counterattack. As a result, the cryptocurrency market sold off sharply across the board: Bitcoin broke below $64,000. In the past 24 hours, more than 67k people worldwide were liquidated, and the total liquidation amount reached $236 million. All three US stock index futures fell, with Nasdaq futures down 1.33%. The synchronized drop in risk assets indicates the market is currently in a flight-to-safety phase, so semiconductor stocks with high beta are naturally hit first.

The combined effect of these three factors forms the macro and micro backdrop for the pullback in Marvell and the broader AI semiconductor sector.

A shift in AI investment logic: from “GPU-centered” to “network as the bottleneck”

To understand Marvell’s position, it’s necessary to first understand how AI infrastructure investment logic has evolved.

Phase 1 (2023–2025): compute first, GPUs reign supreme. In this phase, the market narrative was highly concentrated: the emergence of large models drove an explosive surge in compute demand. Nvidia GPUs were in short supply, and any stock related to AI chips received valuation premiums. The market traded the logic of “the hotter the AI → the more semiconductors rise,” and investors paid limited attention to the specific revenue mix and earnings quality of individual companies.

Phase 2 (2026 to present): from “telling a story” to “watching performance.” As AI infrastructure construction moves deeper, the market begins to distinguish “who truly benefits” from “who is merely concept-related.” Investor focus shifted from “market opportunity” to “order visibility,” “gross margin trends,” and “free-cash-flow conversion ability.”

In this phase, Marvell’s core value comes from its position in the industrial chain. AI data center expansion is moving from “single-rack compute” to “cluster compute”—the growth in the number of GPUs is only the first step; how to efficiently communicate between GPUs is the key bottleneck limiting the expansion of the cluster scale. Marvell’s data center switching chips, SerDes high-speed interconnects, optical interconnect DSPs, and custom AI ASIC business sit directly in this bottleneck.

In other words, Marvell’s long-term industrial logic has not disappeared because the stock price has fallen. What has changed is the price the market is willing to pay for that logic.

Marvell’s near-term pressure and long-term support

Near-term pressure

Valuation compression is still underway. Even after nearly a 30% pullback, Marvell’s valuation remains at historically high levels. A P/E of 81x implies that the market expects the company to maintain extremely high earnings growth over the coming years. If the macro environment tightens further, or if there are any signals of slowing AI capital spending, there is still room for additional valuation downgrades.

Uncertainty exists in the competitive landscape. Broadcom competes directly with Marvell in custom ASICs and networking chips; Nvidia continues to strengthen its positioning in AI data center interconnect through its networking business (Mellanox). At the same time, some large cloud providers are advancing in-house chip programs, which in the long run could reduce reliance on external suppliers.

Marginal changes in AI capital expenditure need continuous monitoring. The market currently remains fairly optimistic about hyperscalers’ capital expenditure expectations. But if the macro economy weakens or AI commercialization progress falls short of expectations, adjustments to capital spending could directly impact Marvell’s order visibility.

Long-term support

Network expansion demand in AI data centers has structural characteristics. As the scale of GPU clusters expands by one order of magnitude, requirements for network bandwidth, latency, and switching capability do not grow linearly; they exhibit superlinear characteristics. This means the market Marvell serves benefits not only from overall AI growth, but also from the increase in unit value driven by rising cluster complexity.

The data center business mix continues to rise. Data center revenue now accounts for about 76% of Marvell’s total revenue. This indicates the company has largely completed its strategic transition from traditional communications chips to AI data infrastructure chips. FY2027 data center business growth is expected to be about 50%, and FY2028 could accelerate to about 55%.

The custom AI chip market is still in an early stage. Demand from large tech companies for custom ASICs continues to grow. Marvell’s technical accumulation and customer relationships in this area form a long-term competitive moat. Goldman Sachs recently pointed out that Marvell’s visibility for its custom chip pipeline is improving; KeyBanc raised its price target by 48% to $385.

The industry’s long-term growth narrative has not reversed. Nvidia CEO recently said publicly that Marvell could become a company with a market value exceeding $1 trillion. While such statements should be treated cautiously, they reflect recognition by core participants in the industry chain of the strategic value of Marvell.

Conclusion

Marvell’s deep pullback over the past month is, in essence, a normal valuation correction during the AI semiconductor sector’s transition from “expectation-driven” to “performance verification.” The company’s long-term industrial logic—being a key player in AI data center network expansion—has not undergone a fundamental change. But market tolerance for valuation is decreasing, and the requirement for performance visibility is increasing.

For investors, the key question now is not whether AI infrastructure still has room to grow, but whether the current price has already priced in the growth outlook reasonably. There is no standard answer to this; it depends on each person’s different judgment about the AI capital expenditure cycle, how the competitive landscape will evolve, and the macro environment.

What is certain is that AI infrastructure construction is still in an early-to-mid stage of transition. As a key participant in this industry chain, Marvell’s long-term value depends on whether it can continuously convert AI demand into verifiable revenue and cash flow—something the market will closely track next.

FAQ

Q1: What are the main reasons Marvell’s stock fell nearly 30% over the past month?

It’s the result of multiple factors stacking together: the AI semiconductor sector’s overall valuations are at historical highs, triggering profit-taking; the market shifted from “telling a story” to “watching performance,” making scrutiny of high-multiple stocks stricter; and the escalation of geopolitical conflicts tightened risk appetite, hitting high-beta semiconductor stocks first. These factors together triggered a broad pullback in Marvell and similar AI chip stocks.

Q2: Has Marvell’s long-term AI business growth logic changed?

No fundamental change. Marvell is in a key part of AI data center network expansion—demand for switching chips, high-speed interconnects, optical interconnect DSPs, and custom ASICs continues to grow. What has changed is the price the market is willing to pay for this logic. Over the long term, the expansion of AI clusters creates superlinear demand for network capabilities, keeping Marvell’s industry position critical.

Q3: How does the competitive landscape between Marvell and Broadcom look in AI chips?

They directly compete in custom ASICs and networking chips. Broadcom has a larger scale and a wider product lineup; Marvell is more focused on data center interconnect and custom chips, with differentiated advantages in optical interconnect DSPs and switching chips. The market structure has not been finalized yet, and both benefit from AI infrastructure expansion.

Q4: Is Marvell’s current valuation attractive?

As of July 10, Marvell’s trailing P/E is about 81x, and its market cap is about $211.6 billion. Despite the sharp pullback, the valuation is still at an unusually high historical level. The average analyst price target is around $270, and some institutional targets are higher. Whether it is attractive depends on investors’ views on the sustainability of AI capital spending and the company’s ability to deliver on performance.

Q5: What are the key variables to watch for the next steps in AI infrastructure investment?

Three core variables: whether hyperscalers’ capital expenditure guidance shows any marginal changes; the order visibility and gross margin trend for companies like Marvell; and whether AI commercialization progress can support the current speed of infrastructure build-out. In addition, the impact of geopolitical risk on valuations for global risk assets must also be continuously monitored.