AI infrastructure investment has moved upstream. The advent of ChatGPT, Claude and other AI applications fueled demand for semiconductor chips that enable the software to “think.” The demand concurrently brought about record capital expenditures to build out hyperscale data centres housing those chips. Now the bottleneck is even more basic: power. For AI, electricity is no longer a utility input; it is strategic infrastructure.

Data centre growth needs energy – a lot of it

That shift is colliding with a US grid whose expansion is constrained at multiple points: new generators are stuck in interconnection queues; interstate transmission still requires approvals across multiple jurisdictions; transformer shortages are delaying grid upgrades; and local opposition is increasingly slowing or cancelling data centre projects. North American Electric Reliability Corporation’s 2025 long-term reliability assessment warned that 13 of 23 North American assessment areas face resource-adequacy challenges over the next decade, underscoring that the issue is not only energy volume, but deliverability and reliability.

Electric Power Research Institute’s Powering Intelligence 2026 report makes the same point from the data centre side. Its “Generation and Capacity Impacts of Data Center Load” analysis finds that data centre growth could require large additions of generation and transmission capacity, but that supply-chain, siting and permitting constraints may limit how fast those additions arrive. In least-cost scenarios, incremental data centre load is met primarily by new and existing gas generation rather than carbon-free resources.

Getting power to where it’s hard to get

That naturally explains the recent order flow into large reciprocating engines. In April, the Finnish vessel engine manufacturer Wärtsilä Oyj Abp announced a 790 MW off-grid power solution for a new Texas data centre facility, using its 50SG natural gas engines. Wärtsilä explicitly framed the order around fast access to reliable power in a region where the grid cannot adequately meet urgent AI-infrastructure demand. Around the same time, the Korean shipbuilder HD Hyundai Heavy Industries Co. Ltd. disclosed that it had signed a US data centre power generation equipment contract based on its 20 MW-class HiMSEN engines, citing total capacity of 684 MW.

The appeal is straightforward. Large reciprocating engines are modular, dispatchable, fast-starting, scalable in increments and deployable closer to load than central-station plants. Compared with combined-cycle gas turbines, nuclear projects or major transmission upgrades, they can often be installed in shorter phases and avoid waiting years for grid interconnection. For a data centre developer, speed-to-power can be as important as cost-of-power.

Maintaining engine power at sea and on land

HD Hyundai Marine Solution Co. Ltd. (443060 KS) in our Emerging Markets Small Cap Strategy is the sole authorized provider of maintenance, repair and overhaul (MRO) aftermarket services to HiMSEN engines worldwide. As a HD Hyundai-affiliate, the company benefits from having HD Hyundai Heavy Industries – the world’s second largest shipbuilder and the largest manufacturer of medium-speed 4-stroke vessel engines – as a captive market. Of approximately 17,000 HiMSEN units in operation globally (most of them generating power for over 4,000 ships at sea), roughly 2,000 units are generating power on the ground.

Could data centres move offshore?

Mitsui O.S.K. Lines and Karpowership’s Kinetics have already signed a memorandum of understanding to develop what they describe as the world’s first integrated floating data centre platform, hosted on a retrofitted vessel and supplied by a powership capable of using LNG. In that scenario, vessel-engine makers are also powering the physical layer of AI.

China’s semiconductor ambitions returned to the spotlight following the recent meeting between President Trump and President Xi. While the US administration reportedly signaled willingness to permit exports of certain downgraded or older-generation AI GPUs into China, the more notable takeaway may have been China’s relatively muted reaction. Rather than relying on controlled access to foreign technology, China appears increasingly focused on accelerating the development of its own semiconductor ecosystem.

While the AI cycle continues to demonstrate remarkable strength, China’s push toward semiconductor self-reliance increasingly appears to represent an additional structural driver for the industry – one that could persist largely independent of the pace or duration of the current AI infrastructure cycle.

A focus on a domestic opportunity for self-reliance

The scale of the opportunity remains significant. China is already the world’s largest semiconductor consumption market, representing well over USD200 billion of annual chip demand and likely continuing to grow meaningfully over the coming decade. Yet domestic self-sufficiency across many semiconductor categories remains relatively low, leaving substantial room for domestic substitution over time. Even within analog and power semiconductors – categories generally viewed as more achievable for domestic suppliers – the opportunity remains large. Industry estimates suggest China’s power management semiconductor demand alone already represents a multi-billion-dollar market, while domestic suppliers still account for a relatively modest share. If China materially increases domestic semiconductor content over the coming years, tens of billions of dollars of annual value could gradually shift toward Chinese suppliers.

China still faces important technological bottlenecks. Advanced EUV lithography remains effectively inaccessible, while gaps persist across certain leading-edge manufacturing equipment, inspection and metrology tools and advanced materials. However, recent developments suggest China continues to make incremental progress across multiple parts of the semiconductor stack despite these restrictions. Domestic memory players have advanced meaningfully in NAND and DRAM, while progress in high-bandwidth memory (HBM), advanced packaging and other areas continues to evolve. More broadly, as the saying goes, necessity is often the mother of invention, and technological constraints themselves can become catalysts for accelerated domestic innovation.

Lessons from solar, batteries and EVs

Importantly, China has already demonstrated an ability to achieve global scale and competitiveness in industries once dominated by foreign incumbents. The country now holds leading positions across solar panels, batteries and electric vehicles, while also becoming increasingly competitive in industrial automation and advanced manufacturing more broadly. Regardless of one’s geopolitical perspective, China’s long-term willingness to commit capital, engineering talent and policy support toward strategic industries should not be underestimated.

As the AI infrastructure cycle evolves, bottlenecks have gradually expanded beyond AI accelerators and memory into broader areas of the semiconductor supply chain. More recently, power management integrated circuits (PMICs) have emerged as an area experiencing tighter supply-demand dynamics, driven by rising demand from data centres and AI infrastructure. AI servers require increasingly sophisticated power architectures, translating into higher semiconductor content and more advanced PMIC requirements. These products typically command higher pricing and more attractive margins, while stronger AI-related demand may also help stabilize pricing conditions across broader analog semiconductor markets.

Against this backdrop, we believe companies positioned within China’s domestic semiconductor ecosystem could benefit from these longer-term trends. One example within our Emerging Markets portfolio is Silergy Corp. (6415 TT), a China-based analog semiconductor company and one of China’s leading domestic suppliers of PMICs.

Why PMICs are important

PMICs are semiconductors responsible for regulating and distributing electrical power within electronic systems, helping ensure that processors, servers, vehicles and industrial equipment receive power efficiently, reliably and safely. Unlike leading-edge AI accelerators, analog and power management semiconductors are embedded across a broad range of everyday electronic applications.

Headquartered in Hangzhou, Silergy designs analog and mixed-signal semiconductors serving industrial, automotive, consumer electronics, computing and communications applications. While the company is gaining increasing exposure to AI servers and data-centre-related applications, its business remains diversified across multiple end markets, which in our view provides a more balanced way to participate in both semiconductor self-reliance and broader electronics content growth.

Silergy is already one of China’s leading domestic PMIC suppliers, yet its market share within China’s broader analog and power semiconductor market likely remains relatively small, suggesting a potentially long runway for continued share gains over time.

While market attention remains concentrated on the most visible AI beneficiaries, some of the more durable investment opportunities may emerge deeper within the semiconductor supply chain and away from the headlines. China’s semiconductor ambitions could ultimately prove to be one of the more important long-term trends still unfolding beneath the surface of today’s AI cycle.

Earnings remain resilient, but growth is concentrated, macro risks are building and selectivity is becoming critical.

Resilience in a tense environment

The Q1 reporting season underscores a growing divergence in global earnings. While US earnings growth remains robust, it is increasingly concentrated in AI-related industries. In contrast, Europe remains in a low-growth, late-cycle environment, while Japan continues to benefit from structural tailwinds. At the same time, a gap is emerging between the AI narrative and broader earnings. While AI-related sectors are seeing strong growth, the benefits have yet to spread across the wider economy.

The conflict in Iran has driven a sharp rise in oil prices and renewed volatility across equities and bonds, reflecting concerns around inflation and energy supply disruptions. It has also led markets to reassess the path of interest rates, with higher energy costs reducing the likelihood of near-term policy easing. This could test the resilience of corporate earnings through 2026.

So far, corporate earnings in developed markets have been more resilient than expected, despite successive macro shocks. Part of this resilience reflects lessons learned over the past five years. The pandemic period, in particular, has led to improved inventory management, stronger cost discipline and a greater willingness to implement cost optimization programs. More broadly, companies appear better equipped to manage their cost base, and in some cases, have demonstrated persistent pricing power. This has been particularly evident in industrials and technology, where contract structures and product differentiation have enabled effective price pass-through. These factors have helped preserve margins even as demand has plateaued or softened.

However, without a swift resolution to the conflict in Iran, global growth could decelerate further, exposing more vulnerable areas of the market. Discretionary spending, manufacturing and energy-intensive sectors such as transportation and logistics are likely to be most at risk. Rate-sensitive sectors, including residential real estate and REITs, could also face valuation pressure.

Looking at the broad small-cap market, balance sheets are structurally more fragile today than they were a decade ago when companies were deleveraging following the Global Financial Crisis. In the current environment, smaller companies are more exposed to rising interest costs and refinancing risk, particularly at the lower end of the quality spectrum.

Positioning for resilience

In this environment, a quality-focused approach centred on sustainable EPS growth remains critical. Our strategy continues to prioritize companies with strong balance sheets and high returns on equity.

As illustrated by our portfolio characteristics, our Global and International Small Cap strategies exhibit the following attributes:

Source: IDA, Bloomberg, MSCI

The lower leverage of these strategies points to less balance-sheet risk and better ability to navigate higher-for-longer rates. At the same time, the higher operating margins and stronger ROE, alongside faster forward EPS growth, are indicative of higher-quality businesses with more durable profitability and earnings power than the benchmark.

In addition, we continue to focus on companies exposed to structural growth drivers. Themes such as electrification, automation, health-care innovation, defence and reshoring offer improved visibility over the medium term. These areas can provide both defensive characteristics in a slowdown and operating leverage in a recovery.