Infrastructure funds are pricing AI data centers like long-lived assets, but power supply, tenant concentration and chip cycles will test the returns.
A consortium of the Artificial Intelligence Infrastructure Partnership (AIP), Abu Dhabi's MGX and BlackRock's Global Infrastructure Partners (GIP) completed its purchase of 100% of Aligned Data Centers in July at an enterprise value of approximately $40 billion. The buyers committed another $5 billion in growth capital at closing. Funds managed by Macquarie Asset Management sold the business.
The additional capital funds expansion of AI-ready capacity, which signals the buyers plan to build as well as hold. Aligned is AIP's first investment. The partnership aims to deploy $30 billion in equity and support up to $100 billion in total investment, including debt.
Infrastructure funds treat data centers as long-term holdings
The Aligned deal ranks among the largest private investments in digital infrastructure, the companies said. Blackstone set an earlier marker in 2021, when it agreed to buy QTS Realty Trust for about $10 billion.
The appeal is familiar to institutional investors. Pension funds and sovereign wealth funds have historically sought long-duration income from assets such as pipelines and toll roads. Data center leases fit that template: cloud and AI companies sign multiyear agreements for capacity, and owners collect rent.
But the infrastructure comparison has limits. Pipeline and toll roads typically serve established patterns of demand. Data centers, by contrast, are being built to support an AI market whose long-term spending and infrastructure requirements remain uncertain. Even when leases provide contractual revenue, investors must consider tenant creditworthiness, renewal terms and the possibility that customers will need less capacity than anticipated.
The immediate demand outlook remains strong. Alphabet, Amazon, Meta and Microsoft have each forecast more than $100 billion in 2026 capital spending, according to their earnings disclosures, much of it for data centers and servers.
The budgets give owners a visible source of leasing demand. They also concentrate that demand among a few tenants, so a change in one company's plans would reach landlords quickly.
Power availability could determine how fast capacity grows
Electricity is one of the sector’s most physical constraints. Data centers used about 415 terawatt-hours (TWh) in 2024, or 1.5% of global electricity consumption, according to the International Energy Agency (IEA). Its base case projects 945 TWh by 2030, with the United States accounting for nearly half of the growth in demand.
IDA estimates that a typical AI-focused data center uses as much electricity as 100,000 households. The largest under construction will use 20 times as much. That makes access to electricity, grid connections and reliable generation central toa data center’s investment value. Constellation Energy agreed in 2024 to supply Microsoft with output from Three Mile Island Unit 1 in Pennsylvania under a 20-year contract, contingent on a reactor restart.
Yet the scale of future demand remains uncertain. The IEA's scenarios, which account for efficiency gains and energy-sector constraints, put 2035 consumption between 700 TWh and 1,700 TWh. Owners that finance capacity against the high end face weaker returns if demand lands near the low end.
The IEA has called for investment in new generations, stronger grids and more efficient data centers. Regulators will also influence project economics through decisions about how utilities recover the costs of grid upgrades and who ultimately pays for additional infrastructure.
Capital moves into markets with energy and chip access
Investors are placing bets beyond the United States, but regional investment strategies differ. MGX's participation in the Aligned acquisition connects Gulf capital with U.S. digital infrastructure. In the UAE, Stargate UAE, a project involving Abu Dhabi's G42, OpenAI, Oracle and Nvidia, plans a 1-gigawatt computing cluster in Abu Dhabi.
Google, meanwhile, has announced a $15 billion AI hub in Andhra Pradesh, India, where the draw is a large user base. The logic differs by region: the Gulf pairs capital with energy, while India offers scale. Both projects need advanced chips, and U.S. export policy governs their sale. Investors cannot control that variable.
Investors therefore face risks that extend beyond construction costs and electricity prices. Export restrictions, changes in technology policy and access to advanced hardware can affect when facilities become operational and which workloads they can support. Capital and energy may be available, but neither can substitute for restricted or delayed computing equipment.
AIP's structure ties investors to the technology supply chain
BlackRock, GIP, Microsoft and MGX formed AIP in 2024, and Nvidia and xAI later joined. The partnership says its members give it insight into the technologies that define AI demand.
That access could help investors understand where computing capacity is needed. It also raises questions about how independently the partnership can assess demand. Some members supply technology or may lease capacity from the kinds of assets AIP finances. Their participation could strengthen the commercial ecosystem around its investments, but it also makes it important to distinguish demand supported by independent customers from demand linked to members’ own activities.
For outside investors, the key issue is not simply who participates in the partnership. It is whether the underlying assets can attract a broad enough customer base and generate durable cash flow under different AI spending scenarios.
Chip cycles and valuations test long-term returns
Timing is the central mismatch. Advanced chips can be superseded within a few years, while buildings and power equipment are financed over decades. The International Monetary Fund and the Bank of England have both warned that valuations of AI-linked assets could fall sharply. Buyers are responding with take-or-pay leases, long-term power contracts and staged capital commitments. These tools limit exposure but do not remove it. They move risk among tenants, utilities and fund investors.
Whether AIP reaches its $30 billion equity target will show how deep institutional appetite runs. The harder question is whether lease terms and power contracts will outlast the chip generations they host. The answer will emerge slowly, in lease renewals, utility rate cases and grid connection queues, well before it appears in earnings.
