Emerging markets' AI story is 'underestimated’ as AI spending pours into EM Asia

Each month at BPM, we offer a slate of articles and content pieces that go deep on a particular topic. This month, we're focusing on Emerging Markets and the sectors and countries within the asset class. 

The emerging markets (EM) index has undergone a rapid transformation, and the force behind it is notably in AI. But unlike the US, where the AI trade centres on hyperscalers and large language models, the EM AI opportunity sits further down the value chain, particularly in the semiconductor hardware that makes the buildout possible.

"The EM AI story is underestimated. We're all focused on the hyperscalers, we're all focused on the Nvidia or where's the next bottleneck but in the US market," said Christine Tan, portfolio manager at Sun Life Global Investments. "Unless you’re truly a global investor, what has been underestimated is how much of this AI infrastructure spending actually has been flowing to EM Asia, specifically North Asia, specifically Taiwan and South Korea.”

Old EM index model misses the AI shift

According to Tan, many institutional investors who’ve maintained EM as a strategic allocation still picture it as roughly a third China, with smaller slices of India, Taiwan and South Korea filling out the rest. Tan underscores that mental model is outdated. She noted that Taiwan has overtaken China as the largest country weight in the MSCI EM index, driven by TSMC's dominance in semiconductor foundry manufacturing, while South Korea has climbed to second on the strength of Samsung and SK Hynix, which now account for about 30 per cent of the South Korean index.

Meanwhile, China has dropped to third - a reshuffling powered by an estimated US$700 billion in global AI capex flowing into North Asian hardware makers.

The investment case, Tan suggests, is not that these names are cheap. Rather, they have rallied hard and sit above their own historical valuations. But relative to US peers, the discount still holds, and for investors who remain committed to the AI capex thesis, the opportunity lies in broadening exposure across the value chain rather than concentrating further in the hyperscalers.

Christopher Knapp, managing director and portfolio manager of Emerging Markets Equities at Jarislowsky Fraser Institutional argued that the AI cycle has done more than generate returns. It’s forced a complete reappraisal of how fundamental EM companies are to the global technology infrastructure.

Still, he cautioned against treating the semiconductor trade as the sole reason to own the asset class, noting the traditional drivers like demographics, urbanization and long-term growth remain intact, and outside technology, valuations across EM look attractive in a way they haven't for some time, he said.

The deeper problem, he suggests, is structural underexposure. Years of EM underperformance coincided with extraordinary US returns, and institutional weightings eroded gradually as a result.

"Weightings or exposures for a lot of global investors have come down in EM over time," Knapp said, emphasizing the current moment is an inflection point where allocators should reassess rather than let that drift continue.

"What you're seeing today in the technology sector is highlighting the innovation occurring in emerging markets and the global importance of many EM countries, companies that maybe people didn't fully appreciate before," he added.

Distinct forces in Asia drive EM AI exposure

Meanwhile, Charlie Dutton, head of emerging market equities at Manulife Investments, sees two distinct investment threads running through EM's AI exposure, each driven by different forces.

The first is the global semiconductor supply chain centred on Korea and Taiwan. Beyond the dominant names like Samsung, SK Hynix and TSMC, he suggests there’s a broad ecosystem of companies involved in capacity buildout.

"Often the cycle has been a one-to-two-year cycle whereas the CapEx plans that we're seeing by the hyperscalers over the next three to four years means that we've been able to extend our duration with regards to this cycle and therefore the valuations we're willing to put on those businesses," he said.

The second thread is China's buildout of domestic wafer fabrication equipment. Beijing has made semiconductor self-sufficiency an explicit policy priority, driven in part by US export bans on advanced chips.

"The drivers of that build out of infrastructure in China are quite independent of the global AI build out. If you have exposures of both of those, you're not doubling up on the same investment thesis," he said.

For portfolio construction purposes, that independence offers diversification that is difficult to find elsewhere within EM's technology-heavy index, Dutton added.

Why Taiwan, South Korea are the current EM index winners

Tan also identified Taiwan and South Korea as the most straightforward country-level expressions of the AI hardware trade but also stressed that each carries distinct risk profiles that investors need to parse carefully.

In Taiwan, the concern is particularly around concentration. According to Tan, TSMC has grown so large within the Taiwan index that investors who already hold it in a global portfolio and then add a separate Taiwan allocation may be unwittingly doubling their exposure to a single stock. She believes that unintended overlap can create far more concentrated risk than the diversification the EM allocation was meant to provide.

Meanwhile, South Korea presents a different challenge: cyclicality. Tan noted how the memory industry - DRAM and NAND - has always moved through boom-and-bust cycles driven by a three-player oligopoly where capacity discipline creates windows of sharp pricing power followed by inevitable corrections.

While Tan acknowledged that the current cycle looks different in both scale and duration, with management teams projecting two to four years before supply catches up with demand rather than the typical 12 to 18 months, she cautioned against treating that extended timeline as permanent. Memory remains a cyclical business, which is why Samsung and SK Hynix trade at lower multiples than TSMC, whose earnings profile is more stable given its unmatched position in advanced chip manufacturing.

Knapp described the South Korea conversation as narrow by necessity, noting Samsung, SK Hynix and related companies account for an estimated 65 per cent of the local KOSPI index, making the country essentially a single-theme bet on memory. While there are governance reform efforts worth watching, he doesn’t think they’re what’s driving capital flows.

"The discussion on South Korea is fairly one dimensional," he said.

Investors need to maintain caution around EM’s AI opportunity

Still, while Tan believes there’s potential opportunity around AI in EM, she’s also ultimately cautious, pointing to three factors. The first is whether corporate AI spending is translating into real bottom-line gains. According to Tan, some large US companies have already pulled back on deployment after finding the costs difficult to justify and investors should watch for that gap between promise and profitability to narrow before assuming the capex cycle is self-sustaining.

The second is the US-China rivalry over the AI supply chain. Export controls on chips and equipment flow one direction; restrictions on rare earths flow the other. That creates risk, but Tan argued it also opens doors for companies that might otherwise have been shut out of the value chain.

"Some of these companies will actually do well because of this competition, because maybe they wouldn't have gotten as much exposure to the supply chain if it wasn't because of the fact that there were restrictions," she said.

The third point Tan raised is a fundamental divergence in how the two AI superpowers are building. The US is pouring resources into compute-heavy, complex models and worrying about applications later, while China is taking an applications-first approach, embedding cheaper, fit-for-purpose models into sectors where cost matters more than sophistication.

While Tan sees room for both, the implications for pricing power and profitability cut in opposite directions. On the US side, she warned, the ability to charge more for increasingly complex models has a ceiling.

"At some point you will hit a wall. When does that wall hit and what does that mean for profitability and what does that mean for the valuations at which you should be trading?" she said. “I do think there's room for both, but those are some of the considerations as we look at how all of this AI investment plays out and how the landscape develops and who ultimately ends up being the top global players. It's not clear yet.”