Today, let’s talk about Marvell.
First, here’s Uncle Tong’s conclusion: the fundamentals are improving, the management is buying back shares with real money, and the valuation is still discounted; however, customer concentration and intensified competition will bring significant volatility. Those who can withstand volatility and are optimistic about AI infrastructure can focus on research; those who only seek a steady upward trend may not find it suitable.
Why is Marvell worth discussing? This year, AI chips have collectively soared, but Marvell’s stock fell about 30% at one point this year, significantly underperforming Nvidia, Broadcom, and AMD. The turning point came on September 24: the board approved an additional $5 billion buyback and initiated a $1 billion accelerated share repurchase (ASR). The CEO emphasized the company’s strong balance sheet, confidence in intrinsic value, and the commitment to continue investing in AI acceleration infrastructure. The market immediately responded: on the day of the announcement, the stock price surged, and in the following nine trading days, it rose on eight of those days, with a cumulative rebound of nearly 20%. More critically, the valuation. Marvell’s current expected price-to-earnings ratio for the next 12 months is about 24 times, significantly lower than Broadcom’s approximately 39 times, indicating that the market has already reflected some uncertainties, leaving room for recovery.
What exactly does the company do? It manufactures data infrastructure chips responsible for moving, storing, connecting, computing, and securing data. Its business covers cloud data centers, carrier networks, enterprise storage, automotive electronics, and more. Over the past few years, through acquisitions and transformations, it has gradually divested low-growth consumer electronics and focused on high-growth areas such as cloud, 5G, and automotive networks. By the end of fiscal year 2024, about 80% of its revenue will come from data infrastructure-related markets; in fiscal year 2025, cloud data centers are expected to contribute nearly 45% of revenue. This is why the core driver of the stock price increasingly depends on the rhythm of data center and AI-related businesses.
The core of the bullish logic lies here: the rise of custom AI chips (ASICs). In the past, large model training and inference mainly relied on general-purpose GPUs, but cloud giants are rapidly moving towards self-developed or custom accelerators to reduce costs and dependence on a single supplier. Similar to Broadcom, Marvell has the capability to replace GPUs with custom ASICs. The advantage of custom chips is that they can be optimized for specific workloads, offering better performance-to-power ratios and cost per computing power; the downside is the high initial investment and less flexibility compared to GPUs. For example, AWS’s collaboration with Marvell on the Trainium line aims for “cheaper and more efficient” solutions, with the commonly cited target range being a 30%-40% cost/performance optimization. Besides Amazon, Microsoft, Google, and Meta are also advancing their own custom or self-developed solutions. The demand side is very certain: as the scale of large models increases, inference scenarios explode, and energy consumption and cost pressures rise, more workloads will shift from general-purpose to custom.
Marvell has already reflected this strength in its financial reports. In the first quarter of fiscal year 2025, the company reported revenue of $1.895 billion, a year-on-year increase of 63%, setting a new historical high; revenue from data center-related sources accounted for 76%, becoming the absolute main engine. The management has also clearly identified custom AI chips as the core of future growth, specifically outlining the technical roadmap and market goals. In the following quarter, the company disclosed that revenue growth remained strong year-on-year while emphasizing continued investment to solidify its lead.
However, market concerns are also very real and have a significant impact on stock prices. First, customer concentration. The custom business currently relies heavily on a few cloud giants, especially AWS. If major customers adjust their pace or technical direction, revenue volatility will be amplified. During the earnings call at the end of August this year, management candidly stated that data center orders are “lumpy,” and demand will experience fluctuations; on that day, the stock price fell sharply, indicating that investors are very sensitive to changes in pace. Second, competitive pressure. In addition to larger players like Broadcom, new entrants like Alchip are also rumored to be deeply following AWS’s subsequent generations, and cloud vendors generally adopt a “multi-supplier” strategy to both diversify risk and pressure prices, which will squeeze Marvell’s bargaining power and profit margins. Third, the vulnerability of the model. The moat of custom ASICs is more about relationships and project depth; once customers turn to self-development or seek other suppliers, previous large R&D investments may be wiped out instantly, which is completely different from the broad applicability of selling general-purpose GPUs. Coupled with the timeline pull of major customers like Microsoft regarding self-development—while they may still need external partners in the short term, the long-term goal of self-sufficiency remains unchanged—these factors will all lead to discounted valuations.
Looking at the finances, Marvell is in a phase of “high growth + high investment.” On one hand, AI-related demand is driving rapid improvements in revenue and profit: in the last fiscal quarter (Q2 of fiscal year 2025), revenue surged 58% year-on-year, surpassing $2 billion for the first time; net profit turned from loss to profit, reaching $195 million, with a non-GAAP gross margin approaching 60% in Q1 of fiscal year 2025. On the other hand, to secure more high-quality custom projects, R&D expenses have significantly increased: Q1 R&D expenses were $508 million, accounting for 26.8%; management also indicated that subsequent operating expenses would continue to rise. The good news is that the company’s balance sheet is strong, allowing for both R&D investment and share buybacks; this $5 billion buyback and $1 billion ASR essentially represent a vote of confidence in cash flow and long-term prospects.
What does the future hold? The positives include: long-term upward trends in AI computing power and data center upgrades; cloud vendors’ “de-standardization” driving the penetration of custom chips; Marvell has gained a reputation among top customers through its delivery capabilities, and combined with its product synergies in networking, storage controllers, and 5G, it has the opportunity to increase its share in integrated solutions. The risks include: customer concentration and competition for market share will continue to fluctuate; once there are signs of “lost orders/reduced orders,” the stock price may be swayed by emotions again; AI capital expenditures may also slow down in phases, leading to uneven order rhythms. Management’s stance is that the custom business will regain momentum later this year, and they remain optimistic about demand improvement in the fourth quarter, but the market needs to validate this with data over the next few quarters.
In conclusion, Marvell is not a “layup win,” but it is in a favorable position on the fast track of custom AI chips; its valuation is still discounted compared to peers, and the buyback provides a buffer for downside. If you agree with the long-term logic of AI infrastructure and can accept periodic volatility and uneven project rhythms, Marvell is worth placing in your core observation list; if you care more about short-term certainty, then it may be wise to wait for financial reports and major order validations. The above content is for informational sharing only and does not constitute investment advice.