Quick Summary
Covering the latest research from top Wall Street investment banks

Comparison of returns in data centers and AI infrastructure: EQIX has a structural premium, while emerging AI infrastructure is significantly affected by capital expenditure structure

Institution
Bernstein
Date
2026-07-15
Authors
Madison Rezaei, Gautam Chhugani, Nancy Wu, Mahika Sapra, Sanskar Chindalia, Harsh Misra
Company
EQUINIX INC; DIGITAL REALTY TRUST INC
Ticker
US.EQIX; US.DLR
Industry
REIT - Specialty; REITs; REIT - Retail
Rating
Outperform for DLR, EQIX, WULF, CIFR, CORZ, CLSK and RIOT; Market-Perform for MARA
BullishLow confidenceThe report says Bernstein remains Outperform on the entire sector, supported by higher stabilized returns at EQIX, improving AI colocation economics, and credible contract demand from hyperscalers and neoclouds.
AuthorsMadison Rezaei, Gautam Chhugani, Nancy Wu, Mahika Sapra, Sanskar Chindalia, Harsh Misra
Target priceDLR $232; EQIX $1,222; WULF $36; CIFR $32; CORZ $32; CLSK $24; RIOT $30; MARA $17
CoverageOther
Asset classesEquity
Business segmentsdata centers、colocation、retail colocation、wholesale colocation、interconnection、managed infrastructure services、emerging AI infrastructure、bitcoin miners
Research firm divisions/subsidiariesBernstein(Other)

AI summary card

Comparison of returns in data centers and AI infrastructure: EQIX has a structural premium, while emerging AI infrastructure is significantly affected by capital expenditure structure

Bernstein compares traditional data center REITs and AI infrastructure platforms transformed from bitcoin miners using stabilized asset ROA, yield on cost, and unlevered IRR, concluding that EQIX delivers better returns than DLR, while the high returns of CORZ and RIOT are more attributable to low incremental capital expenditure arrangements.

Bernstein maintains Outperform on DLR, EQIX, WULF, CIFR, CORZ, CLSK, and RIOT, and Market-Perform on MARA.
Data center REITsAI infrastructureColocationEQIXDLRBitcoin minersYield on costROAUnlevered IRR
  • EQIX consistently outperforms DLR in stabilized asset returns because it has a higher mix of retail colocation, interconnection, and managed infrastructure services.
  • The report prefers yield on cost as a measure of project-level unlevered returns because ROA is more easily distorted by depreciation policy, the definition of stabilized assets, and capital structure.
  • The higher 5-year average ROA and yield on cost of CORZ and RIOT mainly stem from tenant-shared capital expenditures or the conversion of existing bitcoin facilities, and do not necessarily represent the industry's long-term norm.
  • The return levels of WULF, CIFR, and CLSK may be more representative of the emerging AI infrastructure sector, with 5-year average ROA of about 4%-5% and yield on cost of about 17%-19%.
  • Emerging AI colocation transactions have unlevered IRRs of about 8%-13%, which can still create equity value against a 6%-7% financing cost backdrop.

Report interpretation

Overview

This report compares traditional data center REITs and emerging AI infrastructure companies under the same return framework. On traditional colocation REITs, it focuses on comparing the stabilized asset ROA and yield on cost of Digital Realty and Equinix; on emerging AI infrastructure, it compares the contract economics, capital expenditure structure, and unlevered IRR of AI colocation transitions by bitcoin miners such as WULF, CIFR, CORZ, RIOT, and CLSK.

Core views

The core view is that EQIX's retail colocation and interconnection services make its stabilized asset productivity materially higher than DLR's; within emerging AI infrastructure, some high-return cases are mainly driven by special capital expenditure arrangements and should not be simply extrapolated as the industry norm; however, as tenants expand from neoclouds to hyperscalers, enterprises, and chipmakers, and as follow-on contract economics improve, the sector still has positive investment implications.

Analysis framework

The report uses a stabilized asset return framework, combining ROA, yield on cost, and unlevered IRR. For traditional REITs, ROA is measured as stabilized NOI divided by total PP&E, while yield on cost is estimated stabilized net income divided by average net PP&E. For emerging AI infrastructure, based on announced contract terms, steady-state margins, and capital expenditure guidance, the report estimates 5-year average ROA, yield on cost, and project unlevered IRR.

Methodology notes

  • Asset return analysisStabilized asset ROA

    Return on the asset base from stabilized NOI or net income

    Used to measure how efficiently data center assets convert the asset base into earnings, but it is affected by depreciation policy, the definition of stabilized assets, and capital expenditure structure.

  • Project economics analysisYield on cost

    Return on stabilized income relative to initial or total project cost

    The report considers this a better metric for measuring unlevered returns on capital-intensive data center projects, because economic useful life is usually longer than accounting depreciation life.

  • Project valuation analysisUnlevered IRR

    Internal rate of return of data center colocation projects without leverage financing

    Used to compare contract economics in emerging AI infrastructure, with the report disclosing IRRs of about 8%-13% for relevant AI colocation transactions.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

  • EQIX
    Core beneficiary
    Strengths
    High share of retail colocation, with interconnection and managed infrastructure services accounting for about 24% of FY2025 revenue; service margins are closer to software-type businesses, and stabilized asset returns consistently exceed DLR's.
    Weaknesses
    Valuation depends on AFFO multiples and the persistence of long-term high returns, while ROA may still be affected by accounting policy.
    Comparison
    Relative to DLR, EQIX has smaller customer deployment scale, higher unit pricing, and a higher share of interconnection services, resulting in stronger asset productivity.
    Risks
    If interconnection service growth slows, retail colocation pricing comes under pressure, or capital expenditure returns decline, the return premium may narrow.
  • DLR
    Mature data center REIT target
    Strengths
    Owns a mature and large-scale data center portfolio, with wholesale colocation and powered shell businesses providing longer contract duration and revenue stability.
    Weaknesses
    Wholesale colocation has lower unit margins, interconnection revenue accounts for about 8%, and some colocation services rely on third-party partnerships, which may limit profit capture.
    Comparison
    Relative to EQIX, DLR is more focused on wholesale and large-scale deployments, with lower stabilized asset returns but stronger contract stability.
    Risks
    Depreciation policy, the definition of stabilized assets, and development asset classification may affect return comparability; if wholesale colocation pricing weakens, AFFO valuation may come under pressure.
  • CORZ
    High-return emerging AI infrastructure case
    Strengths
    Its capital expenditure sharing arrangement with CoreWeave significantly reduces its own investment, with 5-year average ROA and yield on cost reaching 75% and 79%, respectively.
    Weaknesses
    The high returns mainly come from special financing and capital expenditure structures and may not represent reproducible long-term economics.
    Comparison
    Returns are significantly higher than WULF, CIFR, and CLSK, but sustainability and representativeness are lower than normalized projects.
    Risks
    Tenant concentration, contract execution, non-replicability of capital expenditure sharing arrangements, and construction delivery risk.
  • RIOT
    Emerging AI infrastructure conversion beneficiary
    Strengths
    By converting existing bitcoin facilities, it reduces incremental capital expenditure to about $3.5Mn/IT MW, with 5-year average ROA of 23% and yield on cost of 29%.
    Weaknesses
    High returns depend on existing facility conversion opportunities, and project scale and replicability may be limited.
    Comparison
    Returns are close to EQIX levels, but risk and asset maturity differ from traditional REITs.
    Risks
    Conversion cost overruns, customer acquisition, execution of bitcoin mining asset transformation, and long-term power resource constraints.
  • WULF
    Representative emerging AI infrastructure target
    Strengths
    Uses existing power and transmission infrastructure at brownfield industrial sites, with capital expenditure of about $8-10Mn/IT MW, while the long-term Anthropic contract improves revenue visibility.
    Weaknesses
    Its 5-year average ROA is about 5%, lower than CORZ and RIOT, which have clearer capital expenditure advantages.
    Comparison
    The report believes WULF is more representative of the sector's normal returns than special low-capex cases.
    Risks
    Project delivery, tenant concentration, financing costs, and volatility in AI data center demand.
  • CIFR
    Representative emerging AI infrastructure target
    Strengths
    The AWS triple-net lease structure means most operating expenses are borne by the tenant, with blended average EBITDA margin of about 94% and recurring contract revenue yield improving by about 13%.
    Weaknesses
    Capital expenditure is about $9-11Mn/IT MW, higher than WULF.
    Comparison
    Offsets part of its power asset disadvantage through operating efficiency and a triple-net lease structure, with return levels closer to WULF and CLSK.
    Risks
    Concentration in major clients such as AWS, capital expenditure pressure, contract renewal, and construction execution risk.
  • CLSK
    New entrant in emerging AI colocation
    Strengths
    Signed a data center lease of about $6.6Bn, 175 IT MW, and 20 years, with the triple-net structure implying NOI margin close to 100%.
    Weaknesses
    This is CLSK's first AI colocation transaction, and its execution track record still needs validation.
    Comparison
    Its 5-year average ROA is about 4% and yield on cost about 17%, closer to WULF and CIFR.
    Risks
    Execution of the first AI colocation project, customer concentration, construction costs, and long-term contract performance risk.

Key data

  • EQIX target price$1,222Based on a 25x valuation of 2027E AFFO per share of $48.63.
  • DLR target price$232Based on a 27x valuation of 2027E AFFO per share of $8.52.
  • CORZ 5-year average ROA / yield on cost75% / 79%Mainly affected by the capital expenditure sharing arrangement with tenant CoreWeave, with effective unit capex of about $1.5Mn/IT MW.
  • RIOT 5-year average ROA / yield on cost23% / 29%Derived from converting existing bitcoin facilities, with incremental capex of about $3.5Mn/IT MW.
  • WULF 5-year average ROA / yield on cost5% / 19%The report believes this is more representative of one of the normal return profiles of the emerging AI infrastructure sector.
  • CIFR 5-year average ROA / yield on cost4% / 17%CIFR partly offsets its capital expenditure disadvantage through an AWS triple-net lease structure.
  • Miner power pipeline30GW planned power portfolio; 7GW contractedOver the past two years, miners have contracted 7GW with hyperscalers, neoclouds, and AI chip manufacturers, with contract value exceeding $135Bn.
  • AI colocation unlevered IRR8%-13%The report believes this can generate equity value creation under a 6%-7% financing cost.

Impact & implications

For investment, among traditional data center REITs, EQIX's business mix and service value-add support higher stabilized returns; DLR has larger exposure to wholesale colocation and powered shell, with steadier returns but lower unit margins. The emerging AI infrastructure sector has potential for contract growth and valuation re-rating, but investors need to distinguish true contract economics from one-off capital expenditure advantages, and should not simply treat the high-return cases of CORZ and RIOT as the sector average.

Risks

  • ROA is affected by depreciation policy, the definition of stabilized assets, and capitalization policy, making comparability limited.
  • The high returns of CORZ and RIOT may mainly come from special capital expenditure arrangements, creating high risk in extrapolating to the broader industry.
  • Emerging AI infrastructure projects are still under construction, with risks of delivery delays, cost overruns, and rising financing costs.
  • Tenant concentration is relatively high, and changes in contracts with a single hyperscaler, neocloud, or AI chip customer could affect cash flow.
  • Changes in AI compute demand, data center power access, and the regulatory environment may alter long-term project economics.
  • Valuations of traditional data center REITs depend on AFFO multiples, capital expenditure returns, and stable renewal performance.

What to watch

  • Whether EQIX's share of revenue from interconnection and managed infrastructure services continues to rise.
  • Whether DLR can improve unit margins through wholesale colocation, powered shell, and third-party service partnerships.
  • The revenue yields, lease terms, triple-net structures, and capital expenditure levels of follow-on AI colocation contracts for WULF, CIFR, and CLSK.
  • Whether the low-capex projects of CORZ and RIOT are replicable.
  • The scale of new contract signings within miners' 30GW power pipeline and the delivery progress of the already contracted 7GW projects.
  • Whether unlevered IRRs on AI colocation transactions can remain steadily above the 6%-7% financing cost.
Zhejiang ICP No. 2022035445-5
Disclaimer: Market data, charts, indicators, research views, and other information provided on this website are intended solely for information display, research communication, and educational reference. They should not be regarded as personalized investment advice, securities recommendations, trading instructions, solicitations, or guarantees of return. While we strive to improve the reliability of our data and content, such information may still be subject to delays, errors, incompleteness, or untimely updates due to source differences, methodological limitations, system processing, or market volatility. Users should exercise independent judgment based on their own circumstances and bear all risks and responsibilities arising from the use of this website.

Settings

Sign in to view recent logins