The Poorest Risk-Adjusted Bet in Real Estate Right Now

Steuart AI's latest sector rankings place them at the bottom of the real estate opportunity set.

Every investment comes with a story.

In data centers, the story is almost too easy to tell: AI needs compute, compute needs power, and power needs buildings. Abilene is proof that the boom is real.

That is exactly where Steuart AI is useful. It starts after the story has done its job and asks a colder question: after entry cap rate, NOI growth, exit cap, liquidity, and sector risk are marked to market, are you being paid?

In Q2 2026, the answer is no. Steuart AI screens data centers at a 6.2% projected four-year forward unlevered return against a 10.0% risk hurdle. That is -3.8 percentage points of Cap-Rate Alph, the weakest risk-adjusted score in our cross-asset real estate screen.

Data centers are being sold as the cleanest real estate expression of the AI build-out.

Steuart AI marks them as the poorest risk-adjusted real estate sector in the Q2 2026 cross-asset view.

The demand story is not the problem. The problem is that the market has already capitalized the cleanest version of that story. At today's pricing, allocators are not being paid enough for power delivery risk, tenant concentration, technology obsolescence, and exit liquidity.

The Evidence

Part 1: The story has already been distributed

Private equity deployed $45.70 billion into U.S. data centers in 2025, a five-year high and 72% of the $63.35 billion invested in the sector that year. Infrastructure funds raised a record $221 billion in 2025, with average fund size rising to $1.8 billion. Data center asset-backed securities issuance has grown from $2.6 billion in 2020 to more than $14 billion annually by 2026.

That is not early-cycle behavior. That is a fully financialized thesis. When a trade moves from operator decks to private equity consortiums to ABS shelves, the allocator is no longer underwriting undiscovered demand. The allocator is underwriting exit pricing for a story the market already believes.

Part 2: Steuart AI does not give the story a free pass

Cap-Rate Alpha is Steuart AI's cross-asset hurdle test. It compares each sector's estimated forward unlevered return with the return required for that sector's risk. Positive alpha means the market is paying more than enough for the risk. Negative alpha means the allocator is overpaying.

The contrast is the point. Industrial / Logistics, Retail, and Multifamily all clear their risk hurdles by more than two percentage points. Data centers do not clear the hurdle at all. Steuart AI is not saying AI demand is fake. It is saying the return no longer compensates the allocator for the risks attached to owning that demand at today's price.

Part 3: Abilene proves the demand story, and the execution risk

OpenAI's Stargate campus in Abilene, Texas is already operating at an estimated 0.3 GW, with four of eight buildings operational. The project brought roughly 6,000 workers into a city of just over 100,000 people and intensified an already constrained local housing market. That is a real economic footprint.

But the same case study cuts the other way. OpenAI had planned to expand Abilene to 2.1 GW, then reversed course and redirected that capacity elsewhere. The capacity did not disappear; it moved. That is the underwriting lesson: the demand is real, but the siting, power, tenant, and timing decisions are messier than the headlines imply.

Part 4: The pipeline still has to clear power, equipment, and exit markets

Steuart AI's supply work shows why the exit matters. Operational data center capacity in primary U.S. markets has nearly tripled since 2021, reaching roughly 9.4 GW in 2025. Another 5,994 MW is already under construction in those markets, and the construction pipeline has grown roughly 14x over the past decade.

Power is now the binding constraint. Goldman Sachs Research projects data centers' share of U.S. peak summer power demand to rise from 4.1% in 2025 to 5.3% in 2026 and 8.5% in 2027. Bloom Energy's 2026 Data Center Power Report identifies power availability as the dominant development constraint, with construction costs and community scrutiny becoming additional barriers.

The pipeline charts look linear. The actual delivery curve does not. A project that is announced, pre-leased, or capitalized still has to secure power, interconnection, equipment, community tolerance, and an exit buyer who believes the same rent and tenant story in 2028.

The Steuart AI Hurdle Test

The mistake in crowded infrastructure trades is to underwrite demand certainty and skip price. Steuart AI uses a different order of operations:

• Start with the forward unlevered return, not the theme.

• Compare that return with the sector's risk hurdle.

• Measure the gap in percentage points:

• Cap-Rate Alpha.Then ask what has to go right at exit for the number to survive.

The Obvious Trade Discount is what shows up when a sector becomes so obviously right that everyone pays for it in advance. In Steuart AI terms, it is negative Cap-Rate Alpha attached to a positive demand story.

Data centers are the current example. The demand story clears. The return does not.

What This Means for Allocators

Do not debate whether AI demand is real. It is. Debate whether the current basis pays you for the risks that sit between the demand story and the allocator return: power delivery, tenant concentration, lease re-trade risk, technology obsolescence, and exit liquidity.

Right now, Steuart AI marks data centers at 6.2% projected forward return against a 10.0% hurdle. The same screen marks Industrial / Logistics at 10.4%, Retail at 9.7%, and Multifamily at 10.6%, each with positive Cap-Rate Alpha. Less obvious does not mean less attractive. In this market, less obvious may be the point.

The capital that entered data centers in 2019 and 2020 captured the repricing. The capital entering today is more likely funding that exit. Wait for the reset: power delays, overbuilt secondary markets, tenant-credit scrutiny, or a wider-cap-rate environment can all reopen the spread. The demand will still be there. The price does not have to be this tight.

Decision Snapshot

One question before signing a data-center term sheet in 2026:

What does the deal earn if the exit cap widens by 50 basis points and the anchor tenant is still burning cash in 2028?

If the answer depends on a hyperscaler paying peak rents indefinitely, you are not underwriting a margin of safety. You are underwriting the cleanest version of someone else's story.

If you are reviewing a data-center, AI-infrastructure, or real estate fund this week, send us the package. We'll return an independent Decision Snapshot, no sponsor fee, no placement-agent relationship.

Sources & Methodology

• Steuart AI Q2 2026 Cross-Asset Real Estate Report, EM Capital, May 2026: model-generated forward return, hurdle, Cap-Rate Alpha, and cross-sector comparison. Cap-Rate Alpha = estimated four-year forward unlevered return, net of estimated exit costs, less the risk-adjusted hurdle.

• Epoch AI, “OpenAI Stargate: where the US sites stand,” April 2026 / updated 2026: Abilene capacity, building status, and expansion reversal.

• Texas Standard, February 12, 2026: Abilene Stargate construction worker count and local housing impact.

• S&P Global Market Intelligence, May 12, 2026: U.S. data center private equity investment and share of total investment in 2025.

• Goldman Sachs Research, June 2026: private infrastructure fundraising, average fund size, and data-center power-demand share of U.S. peak summer demand.

• Institutional Investor / Charles River Associates, March 2026: data center asset-backed securities issuance growth.

• CBRE North America Data Center Trends H2 2025: Class A data center cap-rate spread to the 10-year Treasury and supply / construction context.

• Bloom Energy, 2026 Data Center Power Report: power availability, construction costs, community scrutiny, and development constraints.

• Deutsche Bank Research, DB CoTD, December 2025: OpenAI projected cumulative negative free cash flow between 2024 and 2029.

• Federal Reserve / FRED and MarketWatch, July 2026: 30-year Treasury yield near 5%.

For discussion purposes only. Not an offer to sell or solicitation of securities. Not legal, tax, regulatory, or investment advice. Model-generated, forward-looking estimates are hypothetical and not guarantees of future results. EM Capital Management does not act as a placement agent.