Marvell: Memory-market problems and opportunities

One of the main areas of interest and key challenges for investors and the equity market in recent quarters has been the search for “bottlenecks” in the AI industry. Unprecedented investment, huge earnings growth, and the commitment of technology-sector leaders turn any inefficiency or blockage in the supply chain into a gold mine.

The market has already seen an explosion in the valuations of GPU and CPU manufacturers. At present, the re-rating of memory companies is still underway; relative to their long-term averages, many have already risen by hundreds to several thousand percent. In this context, it is understandable that there is a frantic search for the next businesses with exposure to the next set of constraints and investment waves.

Identifying such companies is not easy, given the technical complexity and the unprecedented pace of development and change in the industry. At the intersection of at least several potential bottlenecks sits Marvell.

Barriers that cannot be overcome

To identify why Marvell might partially benefit from another wave of mounting AI technical constraints, it helps to roughly describe the source of those constraints.

The engine driving the commercialization of AI is, and is expected to be, AI agents. The complexity and capabilities of AI agents are no longer limited strictly by raw compute power, but increasingly by data transfer between compute units and HBM memory.

New generations of HBM memory are already in the prototype phase, but even these ultra-advanced components have physical limits that cannot be crossed. Further miniaturization may prove impossible, which would force optimization in other areas.

HBM memory, and subsequent variants such as HBM4 and HBM4E/HBM5, remain indispensable for the development of AI infrastructure, and shortages are expected to persist to varying degrees until around 2028. However, given the delayed timeline of expected growth across many AI market segments, many investors consider it crucial to seek exposure today to the next bottlenecks.

Why Marvell?

Technical chart of Marvell [MRVL.US] (D1)

Source: xStation5

Optimizing the entire AI ecosystem, rather than improving individual components, helps illustrate the company’s true potential. If miniaturization becomes increasingly constrained over time, the next competitive battlegrounds become:

1. Custom silicon design (HBM/ASIC) - Specialization for specific tasks, software, or projects can deliver competitiveness much more cheaply and much faster than waiting for deliveries of the next generations of memory and chips.
2. Networking solutions (CXL) - Improved data transfer enables better utilization of existing compute solutions.
3. Optics - Removing cables and fiber and replacing them with optical data transmission offers entirely new optimization possibilities. It not only enables faster transfer, but also reduces electricity demand, which may become an increasingly critical advantage.

There are a number of companies, often larger and more specialized in these areas than Marvell. However, Marvell’s advantage is that it is one of the few players that offers and develops advanced solutions across all three aspects.

This positioning can enable:

• Better margins
• Better customer retention and monetization
• Potentially an unrivaled advantage due to synergies between solutions

Risks

As mentioned, Marvell has a range of opportunities and favorable prospects ahead. These scenarios, however, will still depend not only on the company but also on the development of the broader AI market. Further advances in optical computing and greater tailoring of ASIC/HBM to specific use cases will only make sense if the wider AI market allows it. This means that, for example, a valuation collapse driven by a financial crisis, regulatory barriers, or declining competitiveness due to consolidation could negate most of the advantages of companies like Marvell.

In this context, it is also crucial to monitor publications from Nvidia. Nvidia continues to set many of the industry standards and is actively working on data-transfer solutions, for example via integrated optics.