IREN · WULF · CIFR · APLD · NUAI
AI Infrastructure: Cap Rates Versus Long-Term Bond Yields
Why risk-free yields, debt pricing, NOI growth, duration, and equity risk premiums should be connected without being conflated.
Thesis
A practical framework for understanding when higher long-duration yields should compress infrastructure values and when contracted NOI growth can offset part of the pressure.
Key chart
| State | Risk-free reference | Risk/development premium | Growth quality | Illustrative cap-rate direction |
|---|---|---|---|---|
| Higher-for-longer | High | Wider | Uncertain | Higher |
| Balanced | Stable | Stable | Contracted | Stable |
| Scarcity plus strong credit | Moderate | Narrower | Durable | Lower |
The argument
The most durable AI-infrastructure asset is not necessarily the one with the lowest quoted cap rate. Durability comes from the quality and timing of cash flow, tenant credit, lease structure, power economics, maintenance obligations, financing, and residual utility.
A development-stage project has more duration than an operating leased facility because cash outflows occur before cash inflows. That makes it more sensitive to discount-rate changes even if the future asset could eventually trade at a core-infrastructure cap rate.
Model AssumptionA useful starting framework is long-duration risk-free yield + asset risk premium + equity and development premium − durable NOI growth.Key findings
- The relevant benchmark depends on duration, not merely the current policy rate.
- Debt pricing can reveal the market's view of project and counterparty risk before public-equity multiples adjust.
- Contracted escalators support value only if the tenant pays, the asset remains competitive, and capex does not absorb the growth.
- A rising cap rate and rising NOI can occur simultaneously.
- A lower nominal cap rate can still produce a poor equity return if the investor overpays, uses expensive leverage, or funds large future capex.
Counterarguments
Strong data-center demand, scarce power, and contractual escalators can offset higher rates. A tenant may also fund equipment or construction, reducing equity duration.
However, scarcity should be evidenced in binding economics. Pipeline demand, letters of intent, and broad market forecasts do not provide the same protection as a contract with credit support and enforceable remedies.
Risks
The framework can mislead if debt and equity claims have different collateral, duration, covenants, and residual rights. Public company overhead, taxes, development losses, and dilution also separate equity value from property value.
Catalysts
A refinancing, asset sale, debt issuance, lease signing, or disclosed project yield can provide a market-observable anchor. Changes in the long end of the yield curve should be reviewed alongside credit spreads and inflation expectations.
Scenario analysis
Valuation analysis
Suppose a stabilized asset produces $100 million of forward NOI. At a 6% cap rate, the implied asset value is approximately $1.67 billion. At a 10% cap rate, it is $1.0 billion. That simple comparison is useful, but it omits construction spending, lease-up, financing, taxes, ownership, and terminal capex.
A full equity DCF should therefore model cash flow from today through stabilization and use a residual value only after the asset reaches the assumptions embedded in the exit cap rate.
Assumptions
The linked sample framework exposes Treasury yield, debt spread, incremental equity premium, NOI, NOI growth, and cap rate. None of those should be inferred from one another without checking the contract and capital structure.
Methodology
The preferred hierarchy is:
- price current project debt or comparable secured debt;
- identify the equity risks not borne by debt;
- model construction-period cash flow and dilution;
- model contracted NOI and capex obligations; and
- select a terminal cap rate consistent with asset quality and duration.
Disconfirming evidence
The higher-rate bear case would be weakened by falling long-duration yields, tighter project spreads, higher-quality tenant guarantees, customer-funded capex, or asset transactions at lower cap rates.
The scarcity bull case would be weakened by power bottlenecks, tenant cancellations, rapid technology obsolescence, or evidence that rent growth is offset by equipment and energy costs.
What would change the conclusion
The conclusion changes when the market provides a better observable: project debt pricing, a comparable asset sale, a binding lease with transparent economics, or operating NOI with a credible growth record.
Primary sources
The source library includes a Treasury data placeholder and can be extended with project debt documents, lease disclosures, and transaction evidence.
Related spreadsheet model
Use the AI Infrastructure Cap Rate and Duration Framework to explore cap-rate, growth, debt-yield, and equity-premium sensitivities.
Article revision summary
Version 1.0.0-sample adds explicit separation between bond yield, debt yield, discount rate, and cap rate.
Audit this conclusion
The conclusion can be summarized elsewhere. The full Ephesus Research page remains the place to inspect the calculation, evidence, sensitivities, revisions, and contrary evidence behind it.
Inspect every material assumption
Review evidence type, source label, date, confidence rating, and the note attached to each model input.
Open exact sectionCompare execution paths
Move between bear, base, and bull conditions, then inspect the phase-by-phase buildout schedule.
Open exact sectionStress-test the valuation
Open the complete sensitivity matrices for discount rates, terminal values, unit economics, delays, dilution, and execution risk.
Open exact sectionReview what changed
Open the dated revision record rather than relying on an undated excerpt or an older model output.
Open exact sectionDownload the underlying model
Open the public spreadsheet or machine-readable JSON and CSV representations for independent review.
Open exact sectionTest the conclusion against contrary evidence
Read the facts, limitations, and developments that would weaken, invalidate, or materially change the stated conclusion.
Open exact sectionTrace the evidence to its sources
Follow the source map to filings, company disclosures, contracts, permits, and other cited records.
Open exact sectionEvidence guide
Evidence and judgment labels
Statements marked Fact are intended to be directly supported by cited evidence. Guidance, estimates, assumptions, inferences, and speculation remain separately named so they are not mistaken for verified facts.
4
mapped sources
Yes
primary support
Related spreadsheets
Audit the linked model
Cross-sector · AI infrastructure
AI Infrastructure Cap Rate and Duration Framework
This educational model starts with an example asset's annual net operating income and applies different cap rates and growth assumptions to show how the estimated asset value changes.
Question this model answers
How do interest rates, income growth, and cap rates change the value of a stabilized infrastructure asset?
Base example asset value
US$1,333m
Outcomes shown
3
Evidence map
Mapped public sources
U.S. Treasury market data — source placeholder
Jul 31, 2026 · United States
Relevant finding
Primary reference for nominal Treasury yields used in duration comparisons.
Review notes
Capture exact observation date and maturity.
Version control
Article change log
Research status
Research status
Preliminary
Conclusion
Mixed
Version
1.0.0-sample
Last reviewed
Aug 2, 2026
Access
Public and free
Continue reading
Related public research
IREN · NUAI · WULF
The Hidden Credit Link in the AI Power Boom
This report examines the credit channel linking frontier-model economics to data-center and power-development finance. Version 1.1 adds two explicit circuit breakers: proprietary providers can retain disproportionate monetization after losing token share, and stronger technology balance sheets can substitute for weaker standalone lab credit. The likely near-term result is financing bifurcation rather than uniform contraction.
IREN
IREN: Five-Year DCF and Buildout Valuation
A current preliminary cohort-cash DCF mapped to the live Google Drive workbook. The analysis distinguishes IREN's official capacity stages, cohort financing, five-year primary GPU life, non-core post-contract tail economics, residual support, revenue sharing, and public community-model reference outputs.
NUAI
NUAI: TCDC Site Economics and Tenant Scenarios
A current preliminary TCDC model, mapped to the live Google Drive workbook, showing why nameplate capacity, tenant obligations, project financing, ownership, promote economics, and dilution must be modeled separately.
Challenge a fact, formula, or interpretation.
Submissions are reviewed before publication. Credited contributors are listed only with permission; private email addresses are never displayed publicly.