Not Missing In Action: How Small Funds Are Playing The AI Investment Boom

Photo credit: Canva

AI is attracting capital on a scale not seen since the dot-com era. Hyperscaler capital expenditure approached US$450 billion in 2025 and is slated to increase significantly in 2026, driven largely by investment in compute, data centres and supporting power infrastructure. These multibillion-dollar deals naturally dominate the headlines, and at that scale, small and mid-market funds can look all but missing in action.

With smaller AUMs, they've to be strategic about how and where they deploy their capital, locking into positions that are significant enough to provide meaningful control or influence.

While these funds seldom headline major platform deals or build facilities at hundreds-of-megawatts scale, they are actively participating in the AI boom as well. Their capital is appearing in localised infrastructure platforms, earlier-stage companies, overlooked assets, data-centre suppliers, enabling technologies, and applications and services above the physical layer.

That said, these are not simply smaller versions of the same investments. They reflect how the market is sorting different capital pools according to scale, mandates and capabilities. Infrastructure funds, private-equity managers and venture investors are all participating in the same growth story, but through very different assets and risks.

Staying Within the Digital Infrastructure Game

Smaller infrastructure funds choose to remain within this asset class by scaling down or entering at an earlier stage.

This includes buying into regional or independent data centres, smaller telco players, alternative network operators, edge facilities, and development platforms. Cordiant’s digital-infrastructure fund, for example, targets mid-sized datacentre, tower and fibre platforms, while the US$250 million Convergence Partners fund invests across similar assets in Africa.

These funds can also take on bigger development risks that more established investors prefer to avoid. They can assemble assets, improve operations, and create the platform that larger funds eventually acquire.

But in the infrastructure business, a lower entry ticket doesn’t necessarily mean lower risks. Smaller platforms can have higher revenue concentration risks, weaker purchasing power, greater dependency on a few markets, and more limited access to financing. Earlier-stage assets carry further risks on regulatory licensing, construction, execution, and commercialisation that an established platform would have already resolved.

Nevertheless, this route allows smaller funds to secure an ownership position that is substantial enough to influence the outcome. This is more meaningful than holding a marginal interest in a flagship asset that is largely controlled by other parties.

The Overspill into Traditional Data Centres

AI is also creating a less obvious opportunity within existing data-centre capacity.

While newer facilities are designed for the power density and cooling requirements of AI, traditional cloud, storage and enterprise workloads don't necessarily require those specifications, nor justify the price premium.

According to JLL, AI workloads accounted for approximately one-quarter of data-centre workloads in 2025, and could reach half by 2030. Even with that kind of growth trajectory, traditional workloads continue to represent a substantial share of the demand. Enterprises continue to move their applications and data to the cloud.

This less glamorous but persistent demand is keeping conventional data centre capacity relevant.

As facilities are built or upgraded to prioritise high-density computing, customers with lower-density requirements could be squeezed out towards more economical capacity. An older data centre can remain investable because scarce AI-ready capacity makes conventional facilities a more economical home for other workloads.

Buildings with existing grid connections are particularly valuable in markets where new power is difficult to secure. Some facilities can also be selectively retrofitted to support higher densities or distributed AI inference without attempting to compete head-on with purpose-built training campuses. JLL has observed similar refurbishment of old and small facilities to extend their useful lives.

However, the investment case still depends on equipment age, energy efficiency, tenant demand and the capital required for upgrading. Legacy is not the same as obsolescence, but continued operation does not guarantee long-term competitiveness. Facilities that fail to keep pace may face weaker demand and greater pricing pressure as conventional capacity becomes more commoditised.

Investing Around the Infrastructure

Capital is also appearing throughout the ecosystem surrounding AI infrastructure, although this ecosystem contains several distinct investment markets.

The first comprises companies that supply, enable or support data centres throughout their lifecycle. Cooling and electrical-equipment manufacturers, construction contractors, maintenance providers, and specialist engineering firms can all benefit from facilities being developed, operated and upgraded.

Some address immediate constraints in equipment or delivery, while others serve the day-to-day requirements of an expanding installed base that must be powered, cooled, secured, and maintained continuously.

However, their underlying economics, revenue models, and market dynamics vary considerably. Equipment manufacturers depend on order volumes, cyclical demand and production capacity. Maintenance and engineering companies may benefit from more recurring demand, but still rely on technical capabilities, service contracts and customer retention.

Power is also emerging as a standalone investment theme as it becomes a binding constraint on data-centre development and operations.

Renewables generation, behind-the-meter solutions, battery storage, and backup power can suit specialist mid-market infrastructure or energy-transition funds. Nevertheless, power segment is not readily investable for every smaller investor. Regulatory requirements, development capabilities, grid access, and capital intensity remain significant barriers to entry.

Exposure to data centre demand doesn’t remove the complexities of investing in the energy market.

A further category comprises nascent technologies that could fundamentally change how digital infrastructure is designed or operated. These include next-generation cooling, new thermal materials and alternative energy technologies that may reduce power consumption, support higher computing density, or expand the terrains where capacity can be deployed.

Larger investors are also active in these areas through venture, growth and specialist strategies. The opening for smaller funds lies in earlier-stage companies that might accommodate smaller cheques while still offering meaningful ownership and influence.

An investment that is immaterial to a megafund can be significant to a specialist fund.

The corresponding risks are very different from those of established infrastructure. Returns depend heavily on technical validation, commercial adoption and the ability to scale production consistently.

More broadly, although the investments across the surrounding ecosystem benefit from the same underlying demand, they are not equivalents. Each carries a different revenue model, risk profile and capital requirement.

Moving Up the Value Chain

Smaller venture, growth and technology-focused funds are also going beyond the physical layer by investing in companies that offer AI-related products and services.

The application layer offers smaller entry tickets and more direct exposure to the commercial value created by AI.

Companies that own the customer relationship or become embedded within important workflows have potential to capture significant value through subscriptions, transaction fees or outcome-based pricing. Successful applications can also scale without requiring the same capital investment for every additional customer.

The investment proposition is very different as compared to physical infrastructure, where the assets and competitive landscape are more reasonably defined. Investors can assess the site, power access, permits, replacement cost, customer contracts and competing capacity within a particular market. Even so, infrastructure does not retain its scarcity value indefinitely. As capacity expands and technical specifications become more standardised, less differentiated parts of the infrastructure layer may become commoditised, shifting value towards what remains scarce, including power, location, connectivity and customer relationships.

The application layer is much less orderly. It ranges from thin interfaces built on third-party models to products embedded deeply within industry workflows. Competitive boundaries move quickly, and functionality can be replicated by new entrants, incumbent software providers or the model owners themselves.

Validation becomes the central challenge. Early adoption can reflect novelty rather than sustained demand, while apparent differentiation can disappear as the underlying technology improves.

The more defensible businesses tend to control an important workflow, possess established distribution, use proprietary industry context, or sit within systems that customers cannot readily replace. Those reaching sufficient scale might also gain bargaining power over the infrastructure providers supporting them, although this is a consequence of success rather than the basis of the original investment.

Moving up the value chain gives smaller funds meaningful ownership closer to where AI is commercialised, and where greater value may ultimately be captured. In exchange, they move from a relatively defined physical market into a far less settled landscape carrying greater validation, product and competitive risks.

Building Exposure Across the Stack

These investment routes are not mutually exclusive. Some managers are building exposure across digital infrastructure, power, enabling technologies and applications. For managers with broader mandates, these investments may sit within the same fund. Others pursue them through separate investment vehicles with different risk and return expectations.

Specialised GPU cloud providers, or neo-clouds, sit between the physical and application layers. They lease high-density facilities, operate GPU fleets and sell computing capacity to customers. Their demand supports new data-centre development and creates opportunities for infrastructure, growth and credit investors.

Where these exposures sit within the same portfolio, they spread the investment across several potential sources of value. Where they are held through separate vehicles, the breadth exists at the manager level rather than within any individual fund.

A portfolio spanning the stack doesn’t need to be an integrated ecosystem either. It can be made up of companies that benefit from the same structural trend without sharing customers, capabilities or commercial opportunities. In such cases, the value comes from portfolio diversification and thematic exposure, rather than operating synergies.

In my experience, portfolio synergies are easy to articulate, but their value can be elusive. Realisation depends heavily on execution, integration and the ability to identify, capture and attribute the resulting benefits.

Not Every Investor Needs to Be an Owner

Some investors are participating in the AI boom without owning or operating the underlying assets.

Private-credit and infrastructure-debt funds provide financing in exchange for interest, fees and contractual protections.

They do not participate fully in the equity upside, but neither do they require operational control to generate returns. Their focus is on repayment capacity, collateral, seniority and the strength of the financing structure.

This doesn’t make credit an automatically accessible route for smaller funds. Financing the largest AI infrastructure projects can require commitments as substantial as the assets themselves. Smaller credit investors are more likely to gain exposure where the financing requirements are correspondingly smaller, including regional assets, equipment and companies serving the wider ecosystem.

This is a less active role in value creation, but the underwriting is not passive. Lenders continue to monitor performance and enforce contractual rights when borrowers fall short of expectations.

Fund interests, listed infrastructure companies and REITs provide other forms of participation without direct ownership or operational involvement. These also extend financial exposure to retail investors, although without meaningful influence over the underlying assets.

The AI boom is stratifying capital by asset, investment stage and form of participation. Some investors own and operate the infrastructure. Others finance it, supply it or back the companies expected to monetise it.

The opportunity extends beyond ownership, but the source of return changes with the investor’s position in the stack.