This is a two-part series on small caps. This week, we look at historical small caps performance, recent small caps performance and the reasons behind the unexpected. Next week, we’ll dive into what this means for allocators and skeptics.

The case for a meaningful small-cap allocation in institutional portfolios

Small caps have just done something the textbook says they should not have. Since the Middle East conflict began at the end of February 2026, small caps have outperformed large caps on both sides of the Atlantic – through an oil-price spike, a sharp re-pricing of European front-end rates and deeply negative eurozone economic surprises. Inside those headline numbers, that is not the behaviour of an asset class to be avoided.

Source: Bloomberg

Small caps have outperformed large caps on both sides of the Atlantic for the last two years.

Source: Bloomberg

Crowded at the top, despite a universe of options

MSCI’s classification methodology defines small caps as roughly the bottom 14% of free-float market capitalization in each country it covers. This creates a universe that spans over 12,000 listed companies and the institutional allocation to that universe has, if anything, contracted. Even within the S&P 500, long-only funds remain overweight the largest quintile by market cap and underweight the smallest. The result is a public-equity allocation that, for all its sophistication, is structurally tethered to the same fifty or so mega-caps that everyone else owns.

As we observed in our September 2025 article, “Why small caps are built for what’s next,”* three of the four episodes of extreme S&P 500 top-ten concentration over the past century – the Go-Go conglomerate years, the Nifty Fifty, and the dot-com boom – were each followed by extended periods of small-cap leadership. Top-ten concentration at the dot-com peak reached 37.0%; today it stands at roughly 40%. The fourth episode, today’s AI-and-Mag-7 configuration, has not yet resolved. An allocator who treats the current setup as permanent is implicitly betting that this time is different, despite repeated historical evidence.
 

Why small caps matter through the cycle, not just at the turn

Small caps earn a place in the policy mix on three grounds independent of timing the next rotation.

1. Breadth of opportunity.
   The small-cap universe is where most listed companies actually live, offering diversified factor and theme exposure that a mega-cap-dominated large-cap book does not provide. Further, the sector composition is materially different from large caps; US small caps carry more industrials, financials and real estate than US large caps, against an underweight in technology.
2. Direct, undiluted exposure to the themes that matter.
   Reshoring, European fiscal expansion, defence, grid and AI-infrastructure build-out and Japanese corporate reform all get expressed more purely through small caps than through the large-cap index, because the pure-play, picks-and-shovels companies in these themes are typically not listed at large-cap market capitalizations.
3. Differentiated economic exposure.
   According to research by Kepler Cheuvreux, European small caps generate roughly 60% of revenues from within Europe, versus roughly 33% for the blue-chip 50. This domestic tilt has become a feature rather than a bug in a world of recurring trade frictions, and the pattern broadly extends across EAFE.Bloomberg data indicate that Japanese small caps generate roughly 65–75% of their revenues domestically, compared with about 45% for TOPIX large caps. That domestic exposure leaves them well positioned to benefit from the emerging global order.

Unexpected outperformance: How are small caps doing it?

The cyclical setup we identified previously – technological disruption, top-of-market concentration, valuations well above long-run averages – remains in place. The conditions since end-February 2026 have been textbook unfavourable for small caps. Per Kepler, European front-end rate expectations re-priced sharply higher, eurozone economic surprises turned deeply negative, and oil spiked before partially retracing. Kepler’s analysis of monthly relative returns since 1994 shows European small caps have, on average, underperformed large caps by roughly -0.29% per month in OECD-defined “downturn” phases and -0.43% per month in “slowdown” phases. That headwind has not materialized.

And yet, small caps are outperforming large caps. Two features of the outperformance are worth flagging:

1. Outperformance has been broad-based across sectors rather than carried by a narrow theme: Kepler’s sector-level data since February 27, 2026 shows positive median small-cap performance against negative median large-cap performance, with European small caps outperforming large caps in nearly every sector.
2. Small caps’ historical sensitivity to bond yields has visibly weakened – per Kepler, the relative performance of European small caps versus large caps has materially decoupled from the German Bund yield since early 2025, breaking a relationship that had held for most of the post-2021 period.

The asset class is no longer waiting for rate cuts.

The valuation picture is the cleanest piece of the case. On Kepler’s data, European large caps trade at roughly 15.2x forward earnings against a long-term average since 2009 of 13.2x. European small caps trade at 14.4x against an average of 14.8x. The US picture is similar: BofA Global Research reports the relative forward P/E of the Russell 2000 versus the Russell 1000 is approximately 0.82x, and historically the relative forward P/E explains roughly 46% of the variability in subsequent 10-year relative returns.
 

Next week: turning the case into allocation

Small cap performance has been unexpected, but not entirely surprising. As investment managers specializing small caps, we know what small caps are capable of. Now that this market segment is regaining the attention of investors, what does this information mean for portfolio allocations? How can allocators adjust their underweight? We’ll discuss those details in next week’s commentary.

Global Alpha was founded on the conviction that small caps are an inefficient asset class in which an experienced team can generate alpha across a full cycle. We are happy to discuss how a small-cap allocation can be sized within a particular plan’s constraints, and how our Global and International Small Cap strategies have navigated the recent environment.

* Contact us for a copy of this article.

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.