AnalysisAI InfrastructureData CentersROIC
Part of the AI Infrastructure Investing Hub series

AI Data Center Risk Screen: 4 Signals Wall Street Misses (DLR, EQIX, VRT, PWR)

EQIX's capex intensity jumped +8.7pp in 8 quarters. CCI's ROIC collapsed to -12%. ANET's customer concentration is HIGH (Microsoft + Meta). Standard screeners show none of this. Here's our 4-signal framework for screening AI data center infrastructure stocks.

15 min read

AI Data Center Risk Screen: 4 Signals Wall Street Misses

Last Updated: December 19, 2025 Data Currency: Q3 2025 10-Q filings. DLR, EQIX, AMT, CCI, VRT, PWR, EME, ANET

Standard screens miss the risks hiding in SEC filings. We analyzed 8 AI infrastructure companies across 24 quarterly filings (Q1-Q3 2025) using Filing Intelligence extraction. Our 4-signal framework surfaces what Wall Street overlooks: $2.3B hidden commitments at DLR, -12% ROIC at CCI, and HIGH hyperscaler concentration at ANET.

Bottom Line: AVOID CCI (structural decline). CAUTION on AMT (5.1x leverage). STRONG profiles at EQIX, VRT, EME, PWR.

  • Digital Realty (DLR): $2.3B off-balance sheet construction commitments; active SEC cybersecurity investigation
  • Crown Castle (CCI): ROIC collapsed from 10.5% to -12.0% over 8 quarters; HIGH customer concentration
  • Equinix (EQIX): Largest customer = 3% of revenue; capex intensity increased +8.7pp to 37.1%
  • Arista Networks (ANET): 161.6% ROIC but HIGH customer concentration (Microsoft + Meta)
  • Vertiv (VRT): ROIC accelerated from 17.5% to 29.1% (+11.6pp); asset-light AI infrastructure beneficiary
  • American Tower (AMT): 5.1x Net Debt/EBITDA (highest in group); 0.76x capex/depreciation (underinvesting)

Sources: Company 10-Q filings, SEC EDGAR

MetricDuck automatically tracks these 4 signals across 1,000+ companies. Compare All 8 AI Infrastructure Stocks →

Introduction

AI infrastructure spending is accelerating as hyperscalers race to build data center capacity for generative AI workloads. But not all "AI plays" are created equal. This analysis examines 8 companies spanning data center REITs (DLR, EQIX, AMT, CCI) and industrial suppliers (VRT, PWR, EME, ANET) using metrics that standard screeners ignore.

What you'll learn:

  • How to use 4 signals to separate quality from risk in AI infrastructure stocks
  • Why CCI's ROIC collapsed to -12% while VRT's accelerated to 29%
  • Which companies have hidden off-balance sheet liabilities
  • Where customer concentration creates binary risk

Data source: This analysis covers 24 quarterly filings (Q1-Q3 2025) processed through MetricDuck's Filing Intelligence extraction, plus 8-quarter trend analysis for key metrics.

Why Do Standard AI Infrastructure Screens Fail?

Everyone's screening for AI infrastructure stocks using the same metrics: P/E ratios, revenue growth, and maybe FFO for REITs. Here's what these screens miss entirely:

  • Digital Realty (DLR) has $2.3 billion in off-balance sheet construction commitments plus an active SEC cybersecurity investigation
  • Crown Castle (CCI) shows decent FFO but ROIC collapsed to -12.0%—invested capital is destroying value
  • Arista Networks (ANET) posts 161.6% ROIC but Microsoft and Meta concentration creates binary risk
  • Equinix (EQIX) increased capex intensity by +8.7 percentage points in 8 quarters—aggressive AI buildout most don't track

Standard screeners show none of this. They'll tell you all these stocks are "AI plays" without distinguishing quality from risk.

This framework uses 4 signals derived from SEC filing analysis to separate AI infrastructure winners from potential value traps.

What Is the 4-Signal Framework?

SignalWhat It MeasuresThresholdWhy It Matters
Off-Balance Sheet ExposureHidden liabilities / Market Cap>15% = ELEVATEDFuture FCF drag not in earnings
Customer ConcentrationTop 3 customers / Revenue>50% = BINARY RISKSingle customer loss = catastrophic
Capex/Depreciation RatioCapEx TTM / Depreciation<0.8 = UNDERINVEST, >2.5 = AGGRESSIVEAsset base aging vs. overbuilding
Net Debt/EBITDALeverage ratio>5x = HIGH RISKRefinancing risk in rising rates

Source: Thresholds derived from analysis of 8 AI infrastructure companies, Q3 2025 10-Q filings.


What Are Off-Balance Sheet Liabilities and How Do They Hide Risk?

Off-balance sheet commitments—construction obligations, purchase agreements, surety bonds—represent contractual cash outflows that don't appear on standard screeners.

TickerMarket CapOff-Balance Exposure% of Mkt CapRisk Level
DLR$59.0B$2.3B construction commitments3.9%MODERATE
AMT$90.0B~$6.2B purchase obligations6.9%MODERATE
ANET$183.4B$4.8B purchase commitments2.6%LOW
EME$29.4B$3.3B (surety + insurance)11.2%ELEVATED
VRT$57.4BN/A (asset-light)-LOW

Key Finding: DLR's $2.3 billion in construction commitments doesn't appear in FFO calculations. Combined with an active SEC investigation into cybersecurity disclosures, this creates compounding risk that FFO/AFFO payout ratios won't capture.

"We are subject to an ongoing SEC investigation...regarding our disclosures relating to a previously disclosed cybersecurity incident." — Digital Realty Q3 2025 10-Q

EME has the highest percentage exposure at 11.2%, though surety bonds and insurance liabilities differ from construction commitments in timing and conditionality.

Product Link: View DLR Filing Intelligence | View EME Filing Intelligence


Which Companies Have Dangerous Customer Concentration?

Customer concentration creates binary risk—when 2-3 customers represent the majority of revenue, losing one isn't a headwind, it's existential.

TickerConcentration LevelTop CustomersRisk Assessment
CCIHIGHT-Mobile, AT&TCarrier consolidation risk
ANETHIGHMicrosoft, MetaHyperscaler in-housing risk
AMTELEVATEDAT&T MexicoSingle-market exposure
EQIXLOWLargest customer = 3%Best diversification
DLRLOWGlobal enterprise mixWell diversified

Key Finding: ANET's exceptional 161.6% ROIC masks significant concentration risk. Per their Q3 2025 10-Q, Microsoft and Meta are dominant customers. If hyperscaler AI capex slows or customers in-house networking infrastructure, ANET's revenue could drop materially.

"Our largest customer accounted for approximately 3% of our recurring revenues for both the three and nine months ended September 30, 2025 and 2024." — Equinix Q3 2025 10-Q

EQIX's 3% largest-customer concentration represents best-in-class diversification among AI infrastructure plays.

Product Link: Compare EQIX vs ANET


What Does the Capex/Depreciation Ratio Reveal?

The capex/depreciation ratio reveals whether companies are investing for growth or letting their asset base age.

TickerCapex/Deprec8Q ChangeInterpretation
EQIX1.93x+0.42xGROWTH MODE - heavy AI buildout
EME1.53x+0.02xHEALTHY - maintaining growth
PWR1.41x+0.07xHEALTHY - steady investment
ANET1.37x-HEALTHY - asset-light model
AMT0.76x+0.16xUNDERINVESTING - asset aging
CCI0.72x-0.09xUNDERINVESTING - declining trend
VRT0.64x-1.08xTRANSITIONAL - was 1.72x

Key Finding: EQIX at 1.93x is investing nearly twice depreciation into new capacity—consistent with AI-driven demand for interconnection. AMT and CCI below 0.8x are letting their asset bases age, a concerning signal for infrastructure businesses.

Red Flag: CCI's capex/depreciation declined from 0.81x to 0.72x while ROIC collapsed to -12.0%. This combination signals structural decline, not temporary headwinds.


How Does Leverage Create Refinancing Risk?

Net Debt/EBITDA above 5x creates refinancing risk, especially as interest rates remain elevated.

TickerNet Debt/EBITDAInterest CoverageRisk Level
AMT5.1x3.6xHIGH - refinancing pressure
PWR2.0x6.5xMODERATE - manageable
VRT0.8x17.1xLOW - deleveraging
EQIX-0.8x3.0xLOW - net cash
ANET-0.7x-LOW - net cash
EME-0.5x-LOW - net cash
CCI-19.5x3.3xRED FLAG - negative EBITDA

Key Finding: AMT's 5.1x leverage is the highest in the group and creates real refinancing risk. Combined with underinvestment (0.76x capex/depreciation) and elevated customer concentration, AMT's profile warrants caution.

CCI's negative ratio reflects negative EBITDA—a structural profitability problem, not a leverage issue.


8-Quarter Trend Analysis: Who's Improving?

Point-in-time metrics miss trajectory. Here's what 8 quarters of trend data reveals:

TickerQ4 2023Q3 2025ChangeInterpretation
EQIX28.5%37.1%+8.7ppAggressive AI buildout
VRT3.0%4.4%+1.4ppModerate increase
PWR4.6%5.0%+0.3ppStable
EME4.2%4.5%+0.3ppStable
AMT4.3%4.0%-0.3ppDeclining
CCI6.4%5.1%-1.3ppDeclining
TickerQ4 2023Q3 2025ChangeVerdict
ANET95.2%161.6%+66.4ppExceptional
EME34.4%51.3%+16.9ppExceptional
VRT17.5%29.1%+11.6ppStrong improvement
AMT7.0%11.1%+4.1ppImproving
PWR9.3%10.4%+1.1ppStable
EQIX12.9%13.2%+0.3ppStable
CCI10.5%-12.0%-22.5ppCollapse

Key Finding: VRT and EME show the strongest ROIC improvement trajectories among AI infrastructure plays. CCI's ROIC collapse from 10.5% to -12.0% is a structural red flag that standard screens completely miss.


What Is the Complete Scorecard for AI Data Center Stocks?

TickerMkt CapSignal 1Signal 2Signal 3Signal 4Verdict
EQIX$76.4B-LOW1.93x-0.8xSTRONG
VRT$57.4BLOW-0.64x0.8xSTRONG
EME$29.4B11.2%-1.53x-0.5xSTRONG
PWR$61.5B--1.41x2.0xSTRONG
DLR$59.0B3.9%LOWN/AN/AMODERATE
ANET$183.4B2.6%HIGH1.37x-0.7xMODERATE
AMT$90.0B6.9%ELEVATED0.76x5.1xCAUTION
CCI$41.9B-HIGH0.72xNEGAVOID

Source: Q3 2025 10-Q filings via MetricDuck Filing Intelligence extraction, December 2025.


What Should Investors Do With This Analysis?

Avoid: Crown Castle (CCI)

CCI exhibits multiple red flags:

  • ROIC collapsed from 10.5% to -12.0% over 8 quarters
  • HIGH customer concentration (T-Mobile, AT&T)
  • Capex/depreciation ratio declining (0.72x)
  • Structural decline signals, not temporary headwinds

Caution: American Tower (AMT)

AMT has the highest leverage in the group:

  • 5.1x Net Debt/EBITDA creates refinancing pressure
  • Underinvesting at 0.76x capex/depreciation
  • Elevated customer concentration (AT&T Mexico)
  • ROIC improving but from low base

Watch: Arista Networks (ANET) & Digital Realty (DLR)

ANET: Exceptional 161.6% ROIC but HIGH customer concentration (Microsoft + Meta) creates binary risk. Monitor hyperscaler capex announcements closely.

DLR: $2.3B hidden liabilities (3.9% of market cap) plus SEC investigation. Low customer concentration is positive. Watch for resolution of SEC matter.

Strong Profiles

EQIX: Best diversification (3% largest customer), 1.93x capex/depreciation (growth mode), net cash position, stable ROIC. Premium valuation reflects quality.

VRT: ROIC accelerating from 17.5% to 29.1%, asset-light AI infrastructure beneficiary, low leverage. Pure play on data center cooling and power infrastructure.

EME: Exceptional 51.3% ROIC, healthy capex/depreciation (1.53x), net cash. Exposed to construction cycle but fundamentals are strong.

PWR: Balanced profile with improving ROIC, healthy investment levels, moderate leverage. Electrical infrastructure for data center buildout.


Screen These Stocks Yourself

Use MetricDuck's peer comparison tool to track these 4 signals in real-time:

Compare All 8 Companies

Filing Intelligence for Hidden Liabilities:

Screen for AI Infrastructure:


Get Notified When Q4 2025 Data Arrives

This analysis uses Q3 2025 10-Q filings. When Q4 2025 quarterly reports are filed (late January 2026), our Filing Intelligence will automatically extract updated signals for all 8 companies.

Subscribe to filing alerts to receive updated analysis:

CompanyAlert Link
DLR - Digital RealtySubscribe to DLR alerts
EQIX - EquinixSubscribe to EQIX alerts
AMT - American TowerSubscribe to AMT alerts
CCI - Crown CastleSubscribe to CCI alerts
VRT - VertivSubscribe to VRT alerts
PWR - Quanta ServicesSubscribe to PWR alerts
EME - EMCORSubscribe to EME alerts
ANET - Arista NetworksSubscribe to ANET alerts

Or manage all your filing alerts from your dashboard.


Methodology Notes

Data Sources:

  • All metrics calculated from SEC EDGAR XBRL filings
  • Filing Intelligence extracted using LLM analysis of 10-Q/10-K text sections
  • 8-quarter trends use TTM (trailing twelve months) calculations

Metric Definitions:

  • Capex Intensity: Capital Expenditures / Invested Capital
  • Capex/Depreciation Ratio: CapEx TTM / Depreciation & Amortization TTM
  • ROIC: Net Operating Profit After Tax / Invested Capital
  • Net Debt/EBITDA: (Total Debt - Cash) / EBITDA

Limitations:

  • DLR capex metrics unavailable due to REIT accounting differences; FFO/AFFO used instead
  • Customer concentration derived from 10-Q disclosures; some companies don't quantify percentages
  • Off-balance sheet exposure categories differ (construction vs. surety bonds vs. purchase obligations)
  • REIT metrics (FFO/AFFO) not directly comparable to industrial supplier ROIC
  • VRT's capex/depreciation ratio reflects transition from high-investment period; trajectory matters more than point-in-time
  • Market cap used for off-balance sheet % calculation; enterprise value may be more appropriate for some analyses
  • Hyperscaler AI capex is cyclical; current strength may not persist

External References:

Next Update: After Q4 2025 earnings (late January 2026)


Explore More AI Infrastructure Research

This article is part of our comprehensive AI Infrastructure Hub, which covers capex efficiency metrics, depreciation manipulation detection, and supply chain analysis.

Related AI infrastructure research:

MetricDuck Research

SEC filing analysis and XBRL data extraction