We debated for months whether the investment community truly needed another article on artificial intelligence. Given the sheer volume of commentary already published, that question was fair. Ultimately, we concluded that the answer is yes – not due to a lack of coverage – but because we realized our perspective is a differentiated one, as we were unable to find another article approaching the topic through the same lens as our investment methodology focused on quality small-cap companies.

In addition, the unprecedented scale of the current artificial intelligence (“AI”) spending program now makes it one of the largest infrastructure investment cycles in modern history. Spending on data centers, power, networking, and enabling infrastructure has accelerated sharply in recent years as hyperscalers and enterprises race to deploy AI at scale, potentially creating outsized growth opportunities for well-positioned participants across the value chain. Yet for long-term, quality-focused investors like ourselves, the central question is not how large this capital expenditure (“CapEx”) wave may become, but how sustainable and investable it truly is over a long horizon.

At Van Berkom, our investment philosophy has always emphasized bottom-up fundamental analysis, a strong focus on business quality and durable competitive advantages, in addition to disciplined valuation. In the context of the AI CapEx boom, our philosophy leads us to a simple conclusion: avoid at all costs novel, unproven, or speculative AI business models, no matter how exciting or convincing their story is.

Our decades of experience investing through several cycles have reinforced the importance of staying true to our principles and invest only in the best, proven, and most enduring franchises of our investment universe. Over time, this has led us to deploy our clients’ capital in time-tested, serial compounders with long-lasting fundamentals through cycle, and we apply the same lens to the AI landscape.

 

Framing the AI CapEx Boom

The rapid advancement of AI models has dramatically increased the infrastructure required to support them – turning data centers from conventional IT environments into some of the most complex and power-intensive industrial assets being built today. This shift has triggered humongous investments across multiple layers of the ecosystem: In a recent industry report, Deloitte projects AI data center CapEx for 2026 near $450B globally, potentially rising to $1T in 2028[i]. Most of that spending will occur in the United States; for perspective, we have seen estimates that AI CapEx represents today more than 1% of U.S. GDP, with the potential to reach 3%+ by 2028, which would be one of the largest private infrastructure buildouts in history on a relative basis.

While there is a great deal of uncertainty around these numbers, an even more important debate is whether today’s spending represents the early stages of a long-duration infrastructure cycle, or a front-loaded surge that could normalize more quickly than expected. In other words, is this a secular or cyclical market? The optimism around AI has certainly propelled U.S. equities to new highs for several quarters now, but increasingly, industry pundits and respected business leaders and investors question the return on investment that developers will achieve over time. Something not discussed as much is also the classic risk associated with aggressive and booming fixed assets investment cycles in capital-intensive industries, when short-term optimistic demand assumptions can drive excess capacity and an oversupply if demand fails to materialize. We certainly recognize the enormous disruptive potential of this technology and nearly infinite theoretical horizontal applications, but most AI products and associated demand and revenue models are still being explored or in early commercialization phases, without much history or certainty as to how this will all develop. We also note that even if optimistic demand assumptions do materialize, not every AI company will become a successful business or a successful investment, especially after the strong gains generated in recent years on stocks of the obvious listed AI “winners” and pre-revenue business concepts.

 

Ripple Effects

Rightfully, servers and chips get most of the attention given their critical role in training and running AI models, in addition to the significant wallet share of AI CapEx they capture. However, looking through the obvious well-publicized beneficiaries such as Nvidia or other mega cap tech companies, the AI CapEx cycle creates many second-order investment opportunities across several attractive sub-industries.

For example, demand for electrons in the United States is emerging as one of the most limiting constraints on the AI buildout. After nearly two decades during which U.S. electricity demand grew at roughly ~0.5% annually[ii], recent revisions toward ~2%+[iii] forward growth represent a four-fold+ acceleration in the pace of annual consumption growth, driven largely by data centers. While previous data centers built typically ranged from tens of megawatts (“MW”), developers are increasingly planning large campuses exceeding 1GW (or 1,000 MW), equivalent to the electricity usage of a mid-sized city. Published estimates for incremental U.S. data center electricity demand associated with new facilities planned by 2030 is estimated between 85GW[iv] to 90GW[v], more than doubling the sector’s existing electricity consumption. On the supply side, there are very few quick fixes: utility-scale solar and wind are fastest to deploy, but natural gas turbine lead times have stretched to 5+ years, and nuclear (particularly small modular reactors) remains a longer-dated and more controversial solution. This dynamic is also causing a multi-year investment cycle in the electrical transmission and distribution infrastructure, with U.S. electric utilities entering a CapEx super cycle totaling >$1T through 2029[vi] – roughly matching what was spent on the grid over the entire prior decade, but compressed into just five years. This buildout is likely to benefit a broad ecosystem of specialized electrical and construction services providers, in addition to power developers in a position to deliver new generation.

Beyond power availability, water access is emerging as a parallel infrastructure challenge for the data-center buildout, particularly as facilities scale in size and density. Large data centers can consume 1 to 5 million gallons of water per day[vii] for cooling needs, with the upper end comparable to the at-home water use of roughly 50,000 people. Importantly, the water is sourced from municipalities; estimates suggest nearly all the water used by major data center operators is purchased from municipal drinking water systems, many of which are already under strain. The AI value chain also extends well beyond data centers and is fundamentally dependent on water, with water consumption growing across power generation, semiconductor manufacturing, and critical mineral mining. Crucially, this incremental demand is being layered onto a U.S. water infrastructure that is already old and underinvested, amplifying the need for upgrade and expansion. Water access being regional in nature, the implication for communities is very localized, having an uneven impact across the nation. This tension is most evident in regions experiencing rapid data center development and an existing water scarcity, including parts of the U.S. Southwest and Texas. Reflecting this pressure, a growing number of states have begun pushing for enhanced monitoring and disclosure of data center water use to protect local water resources. Recent policy responses also extend to funding, with Texas voters approving a $20B investment in water infrastructure[viii] to secure supply in part for industrial growth. As these pressures continue, our research has identified a set of likely beneficiaries, including specialized water engineering and advisory firms, manufacturers of advanced water infrastructure equipment, and a broader set of companies that sell into municipal water utilities.

These are select representative examples of ripple effects rather than an exhaustive list. Being small-cap specialists, the breadth and fragmentation of our large investment universe provide a particularly fertile hunting ground for identifying companies exposed to these attractive themes, with their participation in these niches potentially translating into a needle-moving impact on their financials. Furthermore, given the complexity of large-scale AI data centers, time-sensitive nature of commissioning schedules, and projects often realized in remote locations, products and services sold into data centers tend to be specialized, highly engineered, or only supplied by a handful of players, generally resulting in pricing power relative to other industries. During several conversations we’ve had with industry executives on this topic, we heard that revenue from data centers tends to generate margins above fleet average, meaning that AI-related revenue can both be a growth accelerant while being accretive to profitability, something we believe is often underappreciated.

 

Quiet Winners

Van Berkom’s investment process is focused on identifying reasonably priced, high-quality companies with the following attributes:

  • Durable competitive advantages in established industries
  • High returns on invested capital
  • Proven management teams and sound capital allocation
  • Resilient business models across cycles
  • Attractive growth and profitability expansion prospects
  • Strong and sustainable cash flow generation

Our investment process applied to the AI CapEx boom has inevitably led us to established enablers, not cycle-dependent beneficiaries.  The typical business profile of relevant existing portfolio holdings exhibits a durable business model with strong competitive advantages that predate the current AI cycle, coupled with an emerging but fast-growing direct AI exposure generally representing somewhere between low to high-single-digit percentage of total revenue, given the more recent expansion in the adjacent data center vertical for these long-standing operators. In essence, we are invested in diversified businesses that also sell into data centers without depending on them. This structure allows us to benefit from incremental growth tied to AI infrastructure investments while limiting downside risk if spending slows or pauses. Consistent with our approach, these businesses are market share leaders in their industry, enjoy recurring or predictable revenues with attractive incremental margins and strong free cashflow conversion rates, and generate high returns on invested capital protected by entrenched moats.

Take Ormat Technologies (Ticker: ORA US), covered by Alex Innes (Partner & Senior Analyst), a vertically integrated market leader in geothermal energy and long-standing investment in our client portfolios. For over 50 years, this company has designed, manufactured and operated geothermal power plants providing baseload renewable power to utilities. A substantial majority of its revenue is highly recurring thanks to 20-year+ power purchase agreements signed by customers, while generating extremely high profitability. Geothermal development has significant barriers to entry, including a proven track record as a consistent operator required by customers and specialized expertise needed in geology, reservoir mapping and plant design; case in point – you can count on two hands the number of geothermal developers globally. Since initiating a position in 2018, their services have experienced strong demand in the U.S. driven by state-level mandates to shift electricity generation to baseload renewables. These mandates drove a material increase in pricing for Ormat, rising from $55/MW when we first invested to $85/MW a few years ago. The more recent AI CapEx wave has only added fuel to the fire, with Ormat recently disclosing a pipeline of 250MW negotiated directly with hyperscalers building energy-intensive data centers, materially accelerating demand when compared to Ormat’s existing U.S. portfolio of ~900MW of installed geothermal power. This recent AI-driven surge has also helped push PPA prices closer to $105/MW – driving higher revenue, profitability and plant-level returns.

Another example, Gates Industrial Corporation (Ticker: GTES US), covered by Mathieu Sirois (Partner & Senior Portfolio Manager), is a global manufacturer of innovative, highly engineered power transmission and fluid power solutions. The company sells critical components into several industrial and automotive applications, such as belts, hoses and couplings, where the cost of product failure and equipment downtime is very high relative to the insignificant cost of Gates’ products. The company has built a strong brand reputation over more than a century, holding the #1 market share globally across all products in its Power Transmission segment and is the largest player in North America in its Fluid Power segment. Its leading “first fit” position with original equipment manufacturers maintained over decades has resulted in a vast installed base, driving >2/3 of total revenue being recurring, replacement-driven demand due to normal wear and tear of its products. Owning the broadest distribution footprint globally across 130 countries is also an important component of its moat, allowing a strong presence in key replacement channels. The business model boasts strong economics, generating high margins, best-in-class returns on invested capital and elevated free cashflow conversion. While long-term trends such as industrial automation, infrastructure development and fleet electrification are expected to drive mid-single digit organic growth through cycle, the business has also established a nascent but rapidly growing presence in the data centers market by leveraging existing intellectual property to address liquid cooling needs specific to AI and high performing computing conditions. The company anticipates scaling this vertical from <$10M in revenue today to $100M+ by 2028, adding potentially ~1 point of organic growth per year while being margin accretive due to limited incremental costs associated with this market expansion.

Last but not least is SPX Technologies (Ticker: SPXC US), covered by Charles-Antoine Germain (Analyst), a leading supplier of highly engineered niche infrastructure equipment mostly in commercial heating, ventilation and air conditioning (“HVAC”). The company holds a #1 or #2 position in 90% of products sold, operating in markets that range from oligopolistic structures to more fragmented categories with ongoing consolidation potential. SPX’s brands have been around for a long time, are well recognized in the marketplace, and are trusted by engineers, owners and contractors, having amassed a large installed base driving durable replacement demand representing ~2/3 of total revenue. In addition to leading specification positions, SPX benefits from high switching costs given the disruption involved in switching from existing SPX systems to a competitor’s product, creating a high barrier to entry and providing pricing power.  The business is also well diversified by distribution channel and end markets, selling to a wide range of industrial, institutional, healthcare, commercial and government customers. In addition to generating healthy and expanding margins, the firm’s asset-light model drives attractive free cashflow conversion to fund their inorganic growth program. The firm’s organic growth is underpinned by secular tailwinds such as electrification, decarbonization and reshoring, and more recently, SPX has been a natural participant in the AI CapEx cycle, selling equipment required for data center cooling needs given its already established strong position in commercial markets. Data centers currently represent 9% of total revenue growing double digits.

In addition to the three case studies outlined above, we also see exposure across other portfolio holdings tied to the electricity and water infrastructure expansion themes, including names such as Primoris Services (Electrical and power generation contractor) and Tetra Tech (Water consulting services), as well as additional second order beneficiaries of AI-related investments including APi Group (Commercial fire protection), Armstrong World Industries (Commercial ceilings), DigitalOcean (Data center infrastructure), EPAM Systems (Software engineering services), Modine Manufacturing (Thermal solutions), and RadNet (AI-enabled diagnostic imaging).

 

Risk management

Our investment team is composed of seasoned generalists holding over 80 years of combined U.S. small-cap investing experience. Our institutional knowledge coupled with a continuous scanning across sectors are key to unearth attractive yet underappreciated long-term enduring opportunities, rather than prioritizing momentum over fundamentals. We source all new ideas internally based on in-depth research, which includes “boots on the ground” type due diligence to fully research investment opportunities and leave no stones unturned. We develop relationships with executives running the business, but also with relevant stakeholders across the industry such as competitors, customers, and experts. We don’t trust anyone and look to validate information across separate, independent sources to build our investment thesis and conviction. While time consuming, we think rolling up our sleeves and thinking independently have driven our excess returns over time while mitigating investment risks.

While the U.S. large-cap universe offers several high-quality avenues to participate in AI, the U.S. small-cap opportunity set is far more uneven. Many of the most visible pure-play AI opportunities in small caps are early-stage, pre-revenue, or lack proven business models, making them highly speculative and poorly aligned with our investment discipline. Empirical evidence across the economy shows that less than 10% of initial market pioneers prevail as the final winners in their market, with most failing outright. As a result, we are selective in how we add exposure to long-term AI-driven themes, favoring companies with established operating histories over those whose prospects are closely tied to short-term hype or spending cycles.

Consistent with our risk management framework, we avoid highly cyclical business models, including those with revenue streams concentrated in direct AI-related capital expenditure. Instead, we focus on businesses with durable competitive positions, demonstrated profitability, strong balance sheets, and the ability to self-fund growth across cycles. Valuation discipline remains central to our process, with investment decisions grounded in detailed, multi-year, three-statement discounted free cashflow models and supported by comparative valuation work. We believe this emphasis on quality, resilience, and valuation – applied consistently over time to every investment candidate considered at Van Berkom – allows us to participate in secular growth themes through quality business models while mitigating downside risk.

 

Why the backdrop is attractive today

Recent equities market performance has favored lower-quality and speculative companies, while the stocks of many high-quality compounders have lagged. We believe this divergence has created an opportunity to invest in durable businesses at attractive valuations, with AI potentially acting as an underappreciated incremental tailwind rather than the core investment thesis. Although there might be fears of a bubble in the market, outside of speculative AI-related investments, valuations of quality small-cap stocks are quite reasonable now and offer a favorable risk/reward profile in our view.

For that reason, although the AI CapEx cycle has the potential to endure, we prefer to invest through high-quality, diversified businesses that can compound regardless of how this cycle ultimately unfolds.

 

[i] Deloitte: https://www.deloitte.com/us/en/insights/industry/power-and-utilities/power-and-utilities-industry-outlook.html

[ii] U.S. Energy Information Administration: https://www.eia.gov/totalenergy/data/browser/index.php?tbl=T07.01#/?f=M&start=200001

[iii] ICF Consulting: https://www.icf.com/insights/energy/electricity-demand-expected-to-grow

[iv] Grid Strategies: https://gridstrategiesllc.com/wp-content/uploads/Grid-Strategies-National-Load-Growth-Report-2025.pdf

[v] S&P Global: https://www.spglobal.com/energy/en/news-research/latest-news/electric-power/101425-data-center-grid-power-demand-to-rise-22-in-2025-nearly-triple-by-2030

[vi] Edison Electric Institute: https://www.eei.org/-/media/Project/EEI/Documents/Issues-and-Policy/Finance-And-Tax/Financial_Review/FinancialReview_2024.pdf

[vii] Environmental and Energy Study Institute: https://www.eesi.org/articles/view/data-centers-and-water-consumption

[viii] Office of the Texas Governor: https://gov.texas.gov/news/post/governor-abbott-signs-largest-generational-water-investment-in-texas-history-in-lubbock