AI Data Center Investment: A Plain Guide for Private Investors

What an AI data center investment actually is, the four ways a private investor can get exposure, what the money is made from, and what can go wrong. Sourced, and honest about what you cannot buy.

Someone has told you that AI data centers are the infrastructure trade of the decade, and you are wondering whether you can own a piece of one. This site answers that question without the pitch.

Here is the short version. The electricity that American data centers consume went from 58 TWh in 2014 to 176 TWh in 2023 — about 4.4% of all US electricity — and the Department of Energy's 2024 report puts 2028 somewhere between 325 and 580 TWh, or 6.7% to 12% of national consumption (DOE/Lawrence Berkeley National Laboratory, December 2024). Buildings that use that much power have to be financed. That financing is the investment, and most of it is not available to you.

The four doors, and which ones are open

There are only four realistic ways a private investor gets exposure to AI data centers. They differ enormously in minimum, liquidity and what you are actually exposed to.

The first two are open to anyone with a brokerage account. The fourth is closed unless you are an accredited investor. The third is the interesting middle, and the one most new "invest in AI data centers" marketing is selling.

Where the money comes from

A data center is a landlord business wearing a technology costume. The operator builds a shell, brings in power and cooling, and rents it out — not by the square foot, but by the kilowatt. A tenant contracts for, say, 5 MW of power capacity and pays a monthly rate per kilowatt for the right to draw it. CBRE put the global weighted average asking rate at $217.30 per kW per month in the first quarter of 2025, up 3.3% year over year (CBRE, Global Data Center Trends 2025).

Multiply that out and the appeal is obvious: 10 MW at $200 per kW per month is $24 million of annual contracted revenue. Subtract operating costs and you have net operating income; divide that by what the project cost to build and you have the yield that offering documents advertise.

Everything that can go wrong with an AI data center investment is a way that arithmetic fails to happen. How the leases work and why power is the real constraint are the two pages that explain the inputs. The risks page explains the failure modes.

Why demand is real but that does not make a deal good

Vacancy in North American data center markets hit an all-time low of 1.6%, and 74.3% of the capacity still under construction was already leased before completion (CBRE, North America Data Center Trends H1 2025). That is about as tight as a real estate market gets.

It is also the single most over-used slide in the sector. A tight market tells you the category is in demand. It tells you nothing about whether a particular building will get a tenant, whether its power will arrive on schedule, or whether the sponsor priced the deal so that the returns go to you rather than to them. The 1.6% is a fact about Northern Virginia, Dallas and Chicago — the markets with fiber, substations and hyperscaler campuses. A greenfield site in a rural county is not that market until it has a signed lease, and the gap between those two states is where private investors lose money.

Start here

• What is an AI data center? — how an AI facility differs from the ordinary kind, and why that difference costs money.
• How to invest in AI data centers — the four routes in detail, with the minimum, the paperwork and the exit for each.
• AI data center REITs — the liquid route, what you get and what it dilutes.
• How data center leases work — kilowatts, not square feet; triple net; escalators; the difference between a letter of intent and a lease.
• Why power is the constraint — interconnection queues, substations, and why a "10 MW site" often is not one yet.
• Risks — what actually goes wrong, in rough order of how often.
• REITs vs private deals — the honest comparison, including the fee drag on both sides.
• Investing with a few thousand dollars — the realistic options at that size.
• Glossary — the forty terms you need to read an offering circular.

Going further

Once you know which route you are considering, these go into the detail that decides an outcome.

• What it actually costs to start — the real minimum for each route, and why advertised minimums mislead.
• Data center ETFs and funds — how to work out what a fund actually holds, which is usually not what its name suggests.
• Returns, and why the headline is not yours — the six places a projected yield leaks before it reaches you.
• Data center markets — published rents differ by a factor of three; why location is an investment variable.
• Liquid cooling and what it means for your money — where the extra capital goes, and obsolescence risk stated honestly.
• Accredited investor requirements — the exact SEC tests, and how the net worth calculation treats your house.
• Regulation A vs Regulation D — the two exemptions behind almost every private offering, and what disclosure each gives you.
• How to read an offering circular — a section-by-section walkthrough, and the phrases to search for.
• Questions to ask a sponsor — 27 of them, with what a good answer sounds like and what the usual evasions sound like.

And then the structural questions, which decide more outcomes than any individual deal term.

• The AI capex cycle — four companies' capital budgets underwrite this entire category. Their actual spending, from their own SEC filings.
• Single-tenant risk — what the big REITs disclose about tenant concentration, with the numbers, and the same arithmetic for one building.
• Non-traded REITs — the SEC's own warnings: fees up to 15% of the offering price, and distributions that may be your own money.
• Exit and liquidity — the three ways out, the Rule 144 holding periods, and why "a planned liquidity event" commits nobody.
• What a hyperscaler lease contains — powered shell or turn-key, who reimburses operating costs, and the lease-expiry figures a large REIT publishes.
• Cap rates and yield on cost — three numbers all called "the yield", and the exit cap rate that quietly decides your return.
• Behind-the-meter power — co-locating next to a power plant, and the federal proceeding rewriting the rules for it.
• Data center tax incentives — what the exemptions are worth, and the investment and jobs conditions attached.
• Checking a sponsor on EDGAR — nine free searches that tell you what a sponsor has done before, and what happened.

On the other guide

If you want the engineering — siting, cooling topologies, network fabrics, commissioning — read The Definitive Guide to AI Data Centers, a free, vendor-neutral reference of 166 chapters. It is better at that than we will ever be and we are not trying to compete with it.

This site is for the other half of the problem: you are not building one, you are deciding whether to put money into one.