Every conversation about the AI trade eventually circles back to chips and the companies that make them. But increasingly, the more interesting — and more accessible — way to play the buildout is the physical infrastructure underneath it: the data centers, the power substations, and the land that hosts both. This has become its own real asset category over the past year, distinct from general real estate and distinct from the semiconductor trade, and it’s worth understanding on its own terms.
The scale of capital involved is genuinely unusual. Dell’Oro Group projects that global data center IT capital expenditure will exceed $1 trillion this year alone, and separate analysis puts cumulative AI-related capital spending at roughly $7.6 trillion between 2026 and 2031. That kind of spending touches nearly every corner of the economy — utilities, materials, industrial construction, and of course real estate — which is part of why asset managers have started building dedicated vehicles around it rather than treating it as a subset of a broader tech or REIT fund.
Power, Not Land, Is the Real Constraint
The single most important thing to understand about this sector right now is that the bottleneck isn’t construction — it’s electricity. Ropes & Gray’s mid-2026 industry outlook estimated that U.S. data center power demand could reach 35 to 45 gigawatts by 2030, roughly double 2024 levels, and utilities simply can’t build substation capacity fast enough to keep pace. That mismatch is exactly why some announced projects are being delayed or scaled back even as demand looks insatiable on paper — a widely cited estimate suggests roughly half of the data centers originally scheduled for delivery this year have slipped or been canceled outright due to power interconnection delays.
For investors, that power bottleneck cuts both ways. It’s the biggest risk to any individual project — a facility that can’t secure grid access or a long-term power purchase agreement is effectively a stranded asset no matter how well-located or well-built it is. But it’s also the source of the sector’s pricing power. Operators who locked in interconnection rights and power agreements before this surge in demand are sitting on something genuinely scarce, and that scarcity is showing up directly in valuations.
Three Ways to Get Exposure, With Very Different Risk Profiles
The most straightforward entry point remains publicly traded data center REITs, which trade with daily liquidity and standard REIT reporting. Equinix, for instance, operates more than 260 data centers across 33 countries and focuses on the interconnection and retail colocation side of the business rather than pure wholesale hyperscale capacity, which has historically supported a premium valuation — the stock has traded in the 25 to 30 times forward funds-from-operations range, well above smaller wholesale-focused peers trading closer to 12 to 15 times. For investors who want broader exposure than a single name, sector-focused ETFs holding a basket of data center and digital infrastructure companies offer the same liquidity with less single-stock risk.
The second tier is the middle ground that’s grown fastest over the past year: non-traded REITs and interval funds that hold private data center portfolios, typically offering quarterly rather than daily redemptions and minimums in the $25,000 to $50,000 range rather than the six-figure minimums common to fully private funds. These vehicles trade some liquidity for access to assets — and often to the private credit side of the trade, such as construction loans and sale-leaseback financing to hyperscalers — that public REITs don’t hold directly.
The third tier is fully private, institutional-scale capital. Blackstone describes itself as the largest financial investor in data centers and digital infrastructure globally, with more than $150 billion invested in the category since 2018, and in May it launched a new publicly listed vehicle, BXDC, designed to focus exclusively on AI-boom data center assets. Notably, that fund launched with no existing assets — essentially a blank-check structure that will deploy its roughly $1.75 billion IPO raise into new acquisitions, which is a meaningfully different risk than buying into an established, income-producing portfolio.
The Honest Risk List
None of this is a low-risk trade dressed up as real estate. The construction pipeline is running ahead of deliverable power capacity, which means announced projects and actual operating capacity can diverge sharply — and that gap is the single biggest thing to watch when evaluating any data center-focused fund or REIT. Community opposition and local permitting fights are becoming a real headwind in markets that once welcomed this kind of development without much friction. And because so much of the value in this sector is tied to long-term contracts with a small number of hyperscale tenants, concentration risk at the tenant level is worth as much scrutiny as concentration risk at the asset level.
The Financing Side Deserves Its Own Look
It’s not just equity investors who are finding a way into this trade. A growing slice of the capital funding data center construction is coming through private credit — loans and sale-leaseback structures financing new builds and existing facilities, often with the hyperscale tenant’s own credit backing the arrangement. That private credit angle can generate yields in the high single digits to low double digits, generally with better underlying credit quality than a typical commercial real estate loan, precisely because the tenant covenant behind it usually belongs to a company like a major cloud provider rather than a smaller regional operator. For income-focused investors who find the equity side of this trade too speculative, that lending layer is worth understanding as a separate, more conservative way to participate in the same buildout.
The Timing Question Nobody Can Fully Answer
The single hardest question in this sector right now is how much of the current capital spending pace is durable versus how much reflects a temporary overbuild driven by competitive pressure among a handful of hyperscalers racing not to be left behind. Contracts signed today, at today’s construction and power costs, are a very different economic proposition than the contracts signed a few years ago at lower costs — new builds have to compete against pricing set before this surge in demand, which compresses the margin available to newer entrants even if underlying demand for compute capacity keeps growing. That’s not a reason to avoid the sector, but it is a reason to weight recently signed, fully-priced contracts more heavily than aggregate demand forecasts when sizing a position.
For investors who want exposure to the AI buildout beyond chipmakers, this is a legitimate and growing avenue — but it rewards understanding the specific structure you’re buying into. A public REIT with an interconnection moat, a private interval fund financing construction loans, and a blank-check vehicle waiting to deploy capital are three very different bets wearing the same data center label.
