Global insights

AI scarcity in Asia: beyond broad benchmarks

South Korea’s KOSPI more than doubled in the year to mid-June¹. It then entered a volatile slide. The index has triggered seven market-wide circuit breakers this year, compared with six in the previous 26 years combined². Five of the seven have occurred since single-stock leveraged products began trading on May 27, following rule changes that took effect on April 28. Across Korea’s 16 single-stock leveraged ETFs, combined assets had fallen 43% from their peak by July 21 but remained more than twice their level at launch, according to Bloomberg data. South Korean regulators have responded by raising the minimum cash deposit to 30 million won and pausing new single-stock product listings until market conditions stabilize, according to the FSC.

The turbulence helped shape our decision to lock in gains and cut Korean equities to neutral from overweight in our Midyear Global Outlook at the end of June. Asia is central to AI scarcity, but Korea shows why scarcity is not automatically a broad equity call. Strong demand for high-bandwidth memory has driven extraordinary earnings growth, yet it has also drawn capital into an increasingly concentrated market. By July 15, Samsung Electronics and SK Hynix together accounted for 52% of KOSPI market capitalization, up from 34% at the end of 2025.³ This is the tension at the heart of our Midyear Outlook, Scarcity vs. abundance.

Scarcity without broad exposure

Asia supplies several of the inputs most constrained in the global AI buildout. Taiwan anchors advanced foundry and packaging capacity. South Korean suppliers are projected to account for about 78% of global high-bandwidth memory bit output in 2026.⁴ These bottlenecks remain critical, and the strength of memory demand shows that the scarcity is real. Yet the volatility in Korean equities illustrates why identifying a scarce resource is not the same as buying the broad market that contains it.

This distinction underpins our broader move to neutral on emerging market equities. It is not a retreat from the AI theme or a claim that Korea and Taiwan have lost their strategic importance. It reflects a higher bar after strong gains in AI-linked markets. Broad benchmarks can combine strong earnings exposure with heavy concentration in a few companies and sensitivity to the AI capital-spending cycle. Strong demand alone is not enough. We focus on where it can support durable margins and earnings, and where valuations still leave room for companies to exceed elevated expectations.

The bottleneck is moving

The main constraints on AI development are rolling through the value chain. They have moved from advanced logic and foundry capacity towards memory. We see the next constraints emerging in deployment: power, grids, datacenters, cooling, sensors and industrial systems. This creates opportunities beyond the chipmakers that have led the theme so far.

Japan offers exposure through factory automation, robotics, machine tools, sensors and high-precision components. Asia accounted for nearly three-quarters of global industrial-robot installations in 2024, with China the largest market and Japan second. The International Federation of Robotics expects installations across Asia and Australia to grow at an annual rate of about 8% from 2025 to 2028, faster than in Europe or the Americas.⁵ In Japan, we favor companies that combine technical leadership with improving governance and more disciplined capital allocation (including the unwinding of cross-shareholdings).

China offers scale in robotics, industrial automation, power equipment and manufacturing supply chains. Falling model costs and open-source development could accelerate adoption, but policy priorities, intense competition and U.S. technology controls complicate the path from adoption to profits. We favor companies with defensible technology, diversified customers and a clearer path from rising adoption to profit growth.

The ASEAN region is attracting investment in datacenters and electronics manufacturing. Malaysia, Thailand and Indonesia are drawing projects as companies look for land, power and supply-chain capacity outside more constrained hubs. DC Byte estimates that Southeast Asia’s total datacenter capacity grew at a 72% annualised rate from 2021 to 2024, versus 32% in Northeast Asia. As of May 2025, it had less live capacity—2.1 GW versus 3.1 GW—but a larger development pipeline: 14.0 GW versus 11.3 GW. The pipeline includes capacity under construction, committed and at an early stage.⁶ The critical test is not the size of announced projects. It is whether sites can secure reliable power, grid connections, cooling, water and permits at commercially viable cost.

Investment implications

Asia remains central to the AI buildout, but we favor selective exposure over broad regional or country benchmarks. We still see opportunities in compute bottlenecks where pricing power and earnings durability remain intact. We also see opportunities in automation, power, grids, cooling and digital infrastructure.

This supports our beyond-labels approach. The opportunities may sit in listed technology suppliers, infrastructure assets or the credit financing new capacity. We assess each through its pricing power, demand visibility, funding structure and sensitivity to the AI capital-spending cycle. We prefer companies and assets with durable pricing power, visible demand or contractual revenues, rather than assuming AI scarcity will lift an entire country or regional benchmark.