Move over, Magnificent 7

2026 has generally been a rewarding year thus far for stock investors, with most major segments of the stock market posting positive returns year to date as of mid-July.

But for certain Asian tech stocks, the ride has been extraordinary. Even after a recent pullback, Taiwanese shares have returned 49% year to date and 76% over the past year as of July 20.¹ South Korean stocks, meanwhile, have returned 71% so far in 2026 and 142% over the past year as of July 20.²

Past performance is, of course, never a guarantee of future results. And as the recent correction in certain Asian tech stocks has shown, rapid gains often go hand-in-hand with bouts of significant volatility. But the pullback has done little to change what many investors see as the underlying driver of returns: the growing role of emerging Asia tech companies in the global AI supply chain. From Taiwan's semiconductor manufacturers to South Korea's tech giants, businesses across the region are increasingly playing a pivotal role in building the infrastructure that powers AI. As tech companies race to expand AI capabilities, the industry's ability to manufacture enough chips and other hardware has become a critical constraint.

That dynamic is changing where value is created across the tech sector. Historically, US companies captured much of the value through software, applications, and chip design, while many Asian companies provided the manufacturing capacity. But as manufacturing capacity has become more valuable, so have the companies that provide it.

That's one reason Fidelity Portfolio Manager Di Chen believes AI infrastructure is "one of the defining investment themes of this century." As manager of the Fidelity^® Emerging Asia Fund (), she has focused on identifying the companies that may be most likely to benefit as the AI supply chain evolves.

A supply chain prone to bottlenecks

The AI infrastructure buildout has been so rapid—with hyperscalers demanding immense quantities of specialized hardware for sprawling data centers—that bottlenecks have become commonplace along the supply chain. For investors, those bottlenecks can be revealing: Companies that control a scarce piece of the AI supply chain often become more valuable as customers compete for limited capacity. Recently, constraints have clustered in a few key areas.

The memory bottleneck

In the past year, advanced memory has emerged as one of AI infrastructure's biggest bottlenecks, Chen says. This refers to the specialized components that allow AI systems to quickly store and retrieve large amounts of data. She says AI's shift from training models to running "inference"—when trained AI models generate responses and perform tasks for users—has spurred demand for advanced memory chips, because systems need rapid access to vast amounts of data. The result has been a roughly 5-fold increase in memory prices, as pricing power has shifted to companies supplying the most advanced memory.

Chen's most recent holdings have included Samsung Electronics³ and SK Hynix (), both among the world's leading memory-chip manufacturers. Both have become major suppliers of high-bandwidth memory (HBM), a specialized type of memory that feeds data to AI processors quickly enough to train and run increasingly sophisticated AI models.

The networking bottleneck

Chen also points to optical networking as a major investment theme. As AI data centers grow larger, thousands of processors must exchange information almost instantaneously over fiber-optic networks—the system of cables and equipment that transmits data using light rather than electrical signals.

As AI systems scale, networking needs can grow alongside—and sometimes faster than—computing capacity itself, driving demand for the equipment that keeps AI systems connected. Chen cites WUS Printed Circuit Co.,⁴ which manufactures high-speed circuit boards and interconnect technologies used in AI data centers and optical communications networks, as an example of this investing thesis.

The manufacturing bottleneck

Taiwan Semiconductor Manufacturing Co. (), the world's largest contract chip manufacturer, is perhaps the poster child of AI bottlenecks. "There are limits to how quickly companies can build chips,” Chen says. Bringing new advanced chipmaking capacity online takes years, while demand for AI computing power has grown far faster than many in the industry anticipated. As a result, customers have been competing for scarce manufacturing capacity as demand has rapidly accelerated.

While scarcity can give companies significant pricing power in the short run, Chen also keeps a long-term perspective. Semiconductor markets have historically been prone to boom-and-bust cycles, with shortages often followed by periods of overinvestment and excess supply. Chen says TSMC has shown discipline in its approach to expanding capacity, which gives her more confidence in the company’s long-term earnings outlook.

The next frontiers: Agents and edge computing

Rather than viewing AI as a single investment theme, Chen sees it as a series of rapidly changing opportunities, with new themes—and new beneficiaries—emerging as the technology matures.

"I've seen more than 10 themes emerge over the past year," Chen says. "Before the AI boom, a new theme typically emerged every 3 to 5 years."

As AI systems evolve in sophistication, businesses and consumers may increasingly have access to AI agents—meaning, AI that not only answers questions or solves problems, but can even execute parts of a “to-do” list. "Last year, everyone was talking about chatbots. Now, everyone wants AI agents," Chen says.

Chen says her team spends considerable time studying the economics of AI—how companies ultimately generate returns from the computing power consumed by AI models. She believes the emergence of AI agents could fundamentally change those economics because, unlike chatbots that answer a few questions before a conversation ends, AI agents can work continuously, replacing hours of human labor and creating far greater economic value. The rise of AI agents might not create a new set of winners, but rather increase demand for the same infrastructure that has already emerged as the backbone of the AI economy.

Chen believes another emerging opportunity may be edge computing, in which some AI processing shifts from the cloud onto devices such as smartphones, allowing applications to perform some tasks locally. Although Chen views it as a theme that is a few years away, she believes the economics could become compelling. The cost of running AI models in the cloud remains high, while many mobile devices already have unused computing power.

Chen says she is hearing growing interest in the concept from consumer-facing AI companies in Asia, particularly in China, where developers are looking for ways to reduce the cost of delivering AI services. If that shift gains traction, Chen believes it could create another opportunity for Asian semiconductor and hardware companies that are already developing the chips needed to enable on-device AI.

Adaptability as a key advantage

For Chen, the prospect of edge computing highlights one of the most compelling aspects of Asia's technology sector: its ability to adapt as new AI opportunities emerge and apply capabilities developed in earlier technology cycles.

One example of this adaptability is MediaTek,⁶ which has been among Chen’s top holdings. Once known primarily for smartphone processors, the company has expanded into custom AI chips, known as application-specific integrated circuits (ASICs). As major cloud providers increasingly seek chips tailored to their specific AI needs, many rely on specialized design firms like MediaTek to help develop them.

More broadly, Chen sees companies across Asia leveraging decades of experience in smartphones, electronics manufacturing, and communications hardware to help build the next generation of AI infrastructure.

Keeping perspective amid rapid innovation and periods of volatility

Although the recent pullback has moderated valuations across parts of the AI ecosystem, comparisons to the dot-com era remain common. Chen disagrees with those comparisons, arguing that today's AI cycle is supported by real earnings, rapidly expanding commercial applications, and the emergence of AI agents that fundamentally change the economics of computing.

At the same time, she believes investors need to be selective as the AI market continues to evolve. For example, she places greater emphasis on companies with strong earnings visibility—such as those with contractual, recurring, or otherwise predictable demand—and the ability to benefit from long-term growth trends.

While recent volatility has fueled investor concerns, Chen sees periodic corrections in these stocks' prices as a healthy part of the AI investment cycle. “It would be unsustainable if the stocks kept going up with no corrections,” she says. “I believe these corrections are extremely healthy.”