The Magnificent Seven Skew Market Returns, Style Factors

At Harding Loevner, we believe that whether a business thrives largely depends on the competitive structure of its industry, which we evaluate using Michael Porter’s Five Forces framework. NVIDIA has an estimated 95% share of the market for AI chips. Its graphics processing units (GPUs), which were originally used to power video games, have been adapted to become the key engine of AI computation. The customer base for these chips is highly concentrated and primarily consists of “hyperscalers”—companies such as Alphabet, Amazon.com, Meta, and Microsoft that operate large networks of data centers with immense computing power and storage needs. More than half of NVIDIA’s data-center revenue for its fiscal quarter that ended in January came from large cloud-services providers, and one company alone accounted for 19% of last year’s total company revenue.

From the Porter perspective, a key factor in the bargaining power of buyers is their ability to backward integrate. All of the aforementioned customers have this capability, which threatens to diminish their reliance on NVIDIA. In our view, it’s only a matter of time before this manifests. In fact, Amazon, Google, and Microsoft each have unveiled custom data-center chips as potential substitutes for NVIDIA’s general-purpose GPUs. Because each server in a data center performs only a few specific functions that are repeated over and over, these custom chips, called application-specific integrated circuits (ASICs), are tailored to just the tasks the companies need them to perform.

Industry rivalry is also likely to increase. As the frenzied pace of spending on AI infrastructure eventually eases, companies will begin to look for less expensive options from new suppliers, which may include NVIDIA’s own customers.

NVIDIA is a high-quality company led by skillful managers. Its revenue more than doubled over the past fiscal year, and net profit rose fivefold. But the high potential for increased rivalry and backward integration among key customers suggest that it is unlikely NVIDIA can continue growing at these rates with these margins while maintaining its grip on 95% of the AI-chip market. The fragility of its competitive structure limits our valuation tolerance, which is why we exited the stock in the first quarter after holding it for more than five years.

While it is possible that NVIDIA will continue to exceed increasingly aggressive consensus earnings forecasts for some time, we own other more attractively priced businesses that allow us to invest in AI’s potential. For example, Microsoft and other providers of software and services have the advantage of being able to monetize generative-AI technology without regard for whose GPUs or ASICs are used in the future. Microsoft’s customers also don’t appear to have the ability or interest to try to backward integrate; in fact, AI functionality such as its Copilot chatbot are helping to lock in users. In general, as the applications for AI evolve, value is likely to shift from hardware, such as NVIDIA’s GPUs, to the software and services built upon this infrastructure.

Read more about our software and services holdings in the first quarter Global Equity report.

While US chip designer NVIDIA gets most of the AI press, TSMC manufactures virtually all of the high-performance AI chips designed by NVIDIA. The Taiwan-based company uses some of its most sophisticated manufacturing technologies for these chips, enabling them to be at the forefront of transistor density, speed, and energy efficiency. AI chip customers also increasingly depend on TSMC’s innovative capabilities in integrating and packaging the logic, memory, and input/output components involved. TSMC’s CEO recently projected that in four years about 20% of this company’s total revenue would come from AI-specific chips.

Samsung Electronics has received less attention from investors but is also carving out a niche for itself in AI-related semiconductor manufacturing. Samsung anticipates a rebound in their memory chip prices in 2024, in part due to the expected proliferation of on-device AI, a memory-hungry implementation, and is investing for longer term in the next generation of high-bandwidth memory chips integral to AI applications, promising to leapfrog competitors in performance. Samsung Electronics is also positioning itself to compete with TSMC in manufacturing advanced logic chips, using its capability of designing and manufacturing memory, logic, and chip packaging to unlock new opportunities for its foundry business.

Infineon, widely recognized for its strength in automotive and power semiconductors, is less directly exposed to AI themes but does have products supporting growth data centers and AI computing whose high-power demands is a good match for Infineon’s switched mode power supplies and voltage regulator designs that power high performance systems.