The AI Real Estate Boom (2026): Data Centers, Power, and How a Cross-Border Investor Actually Gets Exposure

Published on: July 9, 2026

Last verified: 9 July 2026. This article is informational. It is not investment advice, and JanusHermes is not a financial adviser. Named companies appear as examples of vehicle types, not as recommendations.

The largest real estate story of the decade is a building most people will never enter, in a place they will never visit, leased to a tenant they already use fifty times a day.

Here are the numbers, from JLL's 2026 Global Data Center Outlook, published January 2026:

• Global data center capacity: 103 GW today, rising to roughly 200 GW by 2030
• New capacity added 2026-2030: nearly 100 GW, effectively doubling the installed base
• Sector growth: 14% CAGR through 2030
• Total capital required: up to $3 trillion by 2030, comprising $1.2 trillion in real estate asset value creation, roughly $870 billion in new debt financing, and a further $1 to $2 trillion of tenant spend on GPUs and networking

For scale: $1.2 trillion of real estate value creation, in one asset class, in five years.

And almost none of it is available to you directly.

That is the real subject of this article.

Part 1: This is not an office building with servers in it

Data centers have stopped behaving like real estate and started behaving like infrastructure. Three structural facts explain why.

Power is the constraint, not land and not capital

This is the single most important sentence in the sector, and it inverts everything a property investor knows.

JLL reports grid connection lead times exceeding four years in primary markets. Average equipment lead time is around 33 weeks, up roughly 50% from pre-2020. A 50 MW facility takes about 18 months to build.

"Speed to power" is now the primary site-selection criterion, ahead of community support, latency, and proximity to customers. Leases are increasingly negotiated on power capacity, not price per square foot.

The consequence: the value of a data center site is a function of its electrical interconnection queue position. A perfect parcel with no power is worthless. A mediocre parcel with a signed interconnection agreement is extraordinarily valuable.

Property people are not usually good at valuing that. Utilities are.

The tenants are the strongest credits on earth

Hyperscalers, Microsoft, Amazon, Google, Meta and peers, occupy more than 50% of all global data center space. JLL notes their leases commonly run ten years or more, with backstops protecting landlords in the event of default.

JLL estimates hyperscalers will allocate around $1 trillion to data center spend between 2024 and 2026 alone, pursuing a dual strategy: leasing capacity and self-building. On JLL's forecast, hyperscalers could construct 41 GW of owner-occupied capacity between 2026 and 2030, roughly doubling that segment to 70 GW, while leased colocation and build-to-suit contribute a further 62 GW.

Read that carefully. The best tenants are also your competitors. A landlord class whose largest customers are simultaneously the largest developers is unusual, and it caps landlord pricing power in ways that do not show up in occupancy statistics.

The workload is changing under the buildings

AI represented roughly 25% of data center workloads in 2025. JLL projects 50% by 2030.

More consequentially, JLL anticipates an inflection in 2027, when inference workloads overtake training.

This is a real estate fact disguised as a technical one. Training concentrates demand in a few enormous, remote, power-hungry campuses, AI training facilities are cited as carrying roughly 10 times the power density of traditional data centers and commanding around 60% lease-rate premiums. Inference distributes demand toward regional centres close to end users, because inference happens every time someone actually uses the model.

If that shift lands as forecast, the geography of data center demand redistributes from a handful of megasites toward many regional ones, within eighteen months.

Part 2: Is it a bubble?

JLL's answer is explicit: "the fundamentals for the sector remain healthy and property metrics do not point to a bubble."

Their evidence is decent:

• Global occupancy around 97%
• 77% of the construction pipeline already pre-committed to tenants
• Lease rates forecast to grow at a 5% CAGR through 2030 (Americas leading at 7%)
• Rising equipment lead times reflecting demand, not distress

Now the counterweights, which matter more than the reassurance.

JLL is a broker. It transacts, leases, and advises in this sector. Its outlook is well-researched and it is also produced by a firm with a commercial interest in the sector's expansion. Data Center Dynamics, reporting on the same document, noted that other research firms, including Omdia, have provided more conservative estimates through 2030. Read both.

Construction costs are climbing. JLL forecasts average global build cost at $11.3 million per MW in 2026, up 6% year-on-year, and that figure covers shell and core only. Tenant technology fit-out for AI infrastructure can run to $25 million per MW on top.

JLL's own risk list includes economic headwinds, persistent energy constraints, supply-chain limits, restrictions on technology trade for geopolitical reasons, and disruptive technologies, quantum computing is named explicitly, that "may necessitate changes to current infrastructure models."

Community opposition is now an underwriting variable. JLL frames it directly: community acceptance "is no longer a secondary concern; it is an operational variable that directly influences schedule certainty, capital deployment, and long-term project returns." Data centers consume water and power in communities that vote.

And the financing has grown exotic. ABS and CMBS issuance in the sector has roughly doubled every year since 2020 and is projected to reach $50 billion in 2026. Core investment strategies now represent about 24% of sector fundraising, up from under 10%. Global M&A has exceeded $300 billion since 2020, the largest single transaction being a reported $40 billion deal led by BlackRock's GIP for Macquarie's Aligned Data Centers platform (50 campuses, over 5 GW), which was still pending completion as of early 2026.

Rapidly securitising an asset class with four-year power queues and a technology roadmap nobody can see past 2028 is not automatically a bubble. It is, historically, the setting in which bubbles have been found.

Part 3: How an individual actually gets exposure

This is where most articles wave vaguely at "REITs" and stop. Here is the fuller map, with the trade-offs.

Route 1: Listed data center REITs

The most liquid and accessible route. The global pure-plays and near-pure-plays include:

Listed as examples of vehicle types, not recommendations.

The trade-off nobody mentions: data center REITs typically yield substantially less than traditional REITs, commonly in the low single digits. If you buy them expecting income, you have misunderstood the instrument. You are buying AFFO growth, priced on the assumption that the buildout continues.

That makes them more rate-sensitive, not less, and correlates them more closely with technology equities than with property.

Route 2: Sector ETFs and digital-infrastructure funds

Diversify single-name risk across REITs, operators, and sometimes semiconductor exposure. Convenient. Also means you may be buying chip stocks under a real estate label, read the holdings, not the ticker. The broader menu of indirect routes is covered in our guide to investing in global real estate without buying property.

Route 3: The power stack

If power is the binding constraint, then the scarcity is not the building. It is:

• Generation and grid interconnection
• Natural gas as bridge and permanent on-site power (JLL expects it to play a "major role" in the US)
• Battery energy storage systems, which shorten interconnection timelines
• Solar-plus-storage, which JLL expects to become a core data center energy strategy by 2030

In EMEA, JLL notes hybrid models combining renewables with private-wire transmission can cut power costs by up to 40% relative to grid supply.

This route is not real estate. It is utilities and energy infrastructure. It is also, arguably, where the actual moat is.

Route 4: Land and interconnection rights

Available only to sophisticated or institutional buyers with local counsel. The asset is not the acreage. It is the position in the utility's queue.

Route 5: The second-order residential effect

This is the route a cross-border residential buyer can actually use.

A 100 GW buildout is a construction programme, a permanent-employment programme, and a local-tax event, concentrated in specific corridors. The Americas will hold roughly half of global capacity and post the fastest growth. Asia-Pacific rises from 32 GW to 57 GW. EMEA adds 13 GW, growing around 10% annually, with expansion concentrated in established European hubs and emerging Middle Eastern markets pursuing digital transformation and sovereign AI capability.

Where those campuses land, three things follow: construction employment, permanent technical employment, and a municipal revenue shock.

And a fourth thing follows too: local opposition, over power prices, water use, noise, and land. JLL flags it as a schedule risk for developers. For a residential buyer, it is a planning and price risk, in both directions.

Buying residential property in a data center corridor is a bet on the first three effects outrunning the fourth. It is a real bet, it is legible, and it is available to individuals in a way that a 300 MW campus is not.

Part 4: What we would actually check

If you are evaluating exposure, through any of the routes above, these are the questions that separate an underwriting from a narrative:

1. What is the power position? Not "is there power." What is the interconnection queue position, and what is the contracted capacity?
2. What is the pre-lease percentage, and to whom? Sector-wide it is 77%. A specific asset at 40% pre-lease with speculative tenants is a different asset.
3. Training or inference? Assets built for 2025-era training economics may face a different demand curve after the 2027 inflection.
4. Where is the AFFO growth coming from, rent growth, development yield, or acquisition? These have very different durability.
5. What is the rate assumption? A low-yield, growth-priced REIT in a rising-rate environment is a duration trade wearing a real estate costume.
6. Who is the counterparty on the debt? With ABS/CMBS issuance doubling annually, know whether you are the equity or, indirectly, the lender.
7. What does the local community think? In 2026, this is a hard number in a development schedule, not a soft concern.

The uncomfortable synthesis

Data centers are the fastest-growing real estate asset class in the world, with 97% occupancy, ten-year leases, and the strongest tenant credits available anywhere.

They are also an asset class where the tenants are the competitors, the constraint is electrical rather than physical, the technology roadmap has an inflection point eighteen months away, the buildings cost $11.3 million per megawatt before anyone puts a computer in them, and the primary forecast used across the industry comes from a firm that earns fees on the transaction.

Both of those paragraphs are true. Neither cancels the other.

For a cross-border residential investor, the practical conclusion is narrow and unglamorous: you are almost certainly not buying a data center. You are deciding whether to hold a listed vehicle that owns them, or whether the housing market next to one is worth owning.

Those are legitimate questions with real answers. "AI is booming, so buy data center exposure" is not a question. It is a slogan, and slogans are what get sold at the top of supercycles.

Frequently asked questions

How big is the data center buildout?
JLL forecasts global capacity roughly doubling from 103 GW to about 200 GW by 2030, requiring up to $3 trillion of total investment, including $1.2 trillion of real estate asset value creation and around $870 billion of new debt financing.

Can individuals invest in data centers?
Not directly, in practical terms. The accessible routes are listed data center REITs, digital-infrastructure ETFs, energy and utility exposure, and, for property buyers, residential markets in data center corridors.

Do data center REITs pay good dividends?
Generally no, relative to traditional REITs. Yields are typically low single digits. The investment case rests on AFFO growth rather than income, which makes them more sensitive to interest rates and more correlated with technology equities.

Is the data center market a bubble?
JLL argues property metrics do not indicate one, citing ~97% occupancy and 77% of the pipeline pre-committed. Other researchers, including Omdia, publish more conservative forecasts. The genuine risks are power availability, four-year grid queues, community opposition, technology disruption, and rapidly growing securitised debt.

What is the biggest constraint on growth?
Power. Grid connection lead times exceed four years in primary markets, and "speed to power" has replaced traditional site-selection criteria.

Why does 2027 matter?
JLL expects AI inference workloads to overtake training in 2027, which would redistribute demand from a few centralised megasites toward distributed regional facilities, changing where data centers get built.

Related reading

• How to invest in global real estate without buying property: REITs, funds and crowdfunding
• Currency risk in international real estate
• How the 2026 tariff war is redrawing global property capital flows

Compare the markets around the boom

If your exposure to the AI buildout is going to be residential rather than a data center itself, the question is which market. JanusHermes brings verified listings and country-level intelligence across 50+ markets into one place. Browse listings on JanusHermes.

This article is informational and reflects publicly available information as of 9 July 2026. It is not investment, financial, tax, or legal advice. JanusHermes is not a licensed financial adviser and does not recommend any security. Companies and funds named appear as illustrations of vehicle categories. Past performance and forecasts are not indicators of future results. Consult a licensed adviser before making investment decisions.

Sources

• JLL, 2026 Global Data Center Outlook: Navigating AI demand, power constraints and global opportunities (published 6 January 2026), and accompanying JLL newsroom release
• Data Center Dynamics, coverage of the JLL outlook, including reference to more conservative Omdia forecasts
• Data Center Frontier, analysis of the JLL 2026 outlook
• ION Analytics / Infralogic, coverage of data center capital markets and the GIP-Aligned transaction, January 2026
• Capacity, coverage of JLL occupancy, pre-leasing, and regional capacity forecasts
• JLL, Global Real Estate Perspective (May 2026)

A note on the numbers: where no source is named, the market figures in this article (prices, yields, costs) are indicative estimates compiled from publicly available market data and industry reporting at the time of writing. Markets move and rules change, so treat them as a starting point and verify current figures with official sources before acting on them.