Executive Summary
- As AI investment accelerates, the infrastructure needed to support it is becoming increasingly capital intensive.
- The sheer investment required to build AI infrastructure means traditional funding sources cannot carry the load alone.
- Marc Harris, Head of Real Assets, EU&ME at Vistra Fund Solutions, explains how private capital can play a growing role in turning AI investment into the physical infrastructure needed to power it.
The AI investment story is often told through the lens of chips, models and software. But behind every advancement sits a far more fundamental requirement, and that’s whether there’s enough physical infrastructure to run it.
Leading hyperscalers are expected to invest a combined $5.3 trillion between 2025 and 2030, with a significant share directed towards technology and data centres. Turning that investment into operational capacity, however, requires more than corporate spending alone. Sites need to be acquired, power needs to be secured, and facilities need to be built and installed with specialist equipment, often across projects with long and complex development timelines.
This is creating a growing role for private capital. Infrastructure, private equity, private credit funds and real estate can provide capital tailored to individual projects, offering the flexibility needed to navigate the different stages of development and cash-flow requirements. Constraints around power and grid capacity are only going to intensify, meaning flexibility will also become increasingly important.
Traditional financing is coming under pressure
The largest hyperscalers have strong cash flows and broad access to public debt, but borrowing on the scale envisaged creates practical constraints. Bond investors can become heavily exposed to the same small group of issuers, while banks must manage concentration risk and the long duration of loans attached to major developments. Goldman Sachs expects these pressures to increase as AI-related borrowing rises.
Private real-asset funds offer developers another substantial source of financing. Infrastructure funds held just over $1.7 trillion in assets as of September 2025, including almost $400 billion of dry powder, while private real estate funds held $2.1 trillion with a further $600 billion available to deploy. This capital is not reserved for data centres, but it demonstrates the depth of the markets able to support them.
The range of needs this capital can address is also important. Equity can absorb early development and construction risk; infrastructure funds can support long-duration operating assets; real estate strategies can finance land and buildings; and private credit can fund equipment, expansion and refinancing. A data centre’s capital needs evolve from site acquisition through construction to stabilised operation.
This is already visible in the US. Private equity-backed investment in data centres reached $45.7 billion in 2025, representing 72% of the $63.35 billion invested in the sector. Private credit is also expanding: more than $200 billion of loans to AI-related companies were outstanding by early 2026, according to the Bank for International Settlements. Private credit can support projects whose timing or risk profile does not fit a standard corporate loan.
A broader pool of capital for a complex buildout
Data centres are well suited to private financing because they do not fit neatly within a single asset class. The site and building have real estate characteristics; power and network connections resemble infrastructure; and specialist equipment may need upgrading on a much shorter cycle. Financing must therefore accommodate assets with different useful lives, risks and return expectations within the same development.
Private funds can match the form of capital to the project’s stage and risk. A stabilised facility leased to an established hyperscaler presents a different proposition from a development awaiting a grid connection or serving a newer cloud operator. Private financing enables the cost and duration of funding to reflect these distinctions more closely than a standardised instrument may allow.
Power makes this flexibility particularly valuable. Data centres now represent 41% of private digital infrastructure deals, up from 26% a year earlier, while limited grid capacity is directing investment towards secondary locations and projects with dedicated generation. Financing a data centre increasingly means financing the wider system required to operate it, including grid connections, generation, storage, cooling and, in some cases, transmission infrastructure.
Private capital can be structured around this integrated power-and-compute ecosystem. An infrastructure fund may back a data centre platform while a credit fund finances individual developments, introducing capital as projects reach agreed milestones. This gives operators access to a broader financing base without placing excessive pressure on a single market.
Building the infrastructure to support AI growth
AI’s next phase will depend on whether the infrastructure needed to support it can be built at the pace that investment is being committed. Hyperscaler spending, bank lending and public debt will remain important, but they will not necessarily be sufficient for every project or every stage of development.
Private markets can help fill that gap by providing capital that is patient, adaptable and structured around the specific characteristics of data centre projects.
Developments are becoming more and more complex and increasingly intertwined with power and other supporting infrastructure, so having the right financing model will be just as important as having the capital itself.
This should be viewed as a real opportunity for the industry, not just to fund more data centres, but to actually build a financing ecosystem capable of supporting them from development through to operation.
The sheer investment required to build AI infrastructure means traditional funding sources cannot carry the load alone. Private markets can provide the scale of capital, longer investment horizons and flexible financing structures needed to meet the shortfall between hyperscalers’ own spending and the total cost of bringing projects online.
Turning AI investment plans into fully powered, operational data centres will therefore require not just substantial private capital, but the fund structures and operating infrastructure capable of deploying it efficiently.
