GuidesAI InfrastructureCapital ExpendituresData Centers

AI Infrastructure Investing Hub: Capex Analysis, Risk Screening, and Beneficiary Research

AI infrastructure spending is approaching $500B annually across hyperscalers alone. Our research library covers capex efficiency metrics, depreciation manipulation detection, concentration risk screening, nuclear power for data centers, and supply chain analysis—everything you need to evaluate AI infrastructure investments.

min read

AI Infrastructure Investing Hub

Your central resource for navigating the $500B+ AI infrastructure buildout

AI infrastructure spending is unprecedented in scale and pace. Hyperscalers are investing $370B+ annually, with combined projections approaching $500B by 2026. This creates opportunities—and hidden risks—across the supply chain.


What You'll Find Here

Our AI infrastructure research library covers:

  • Capex Efficiency Framework — Track $370B in hyperscaler spending with 3 key metrics
  • Depreciation Manipulation Detection — Identify when companies inflate earnings through accounting
  • Concentration Risk Screening — Spot hidden customer and supply chain dependencies
  • Supply Chain Analysis — Power generation, equipment makers, data licensing
  • Nuclear Power for AI — Utility deal structures, uranium supply chain, and hidden liabilities
  • Company Deep Dives — From CAT to Reddit to the Magnificent 7

Start here to understand how hyperscalers deploy capital.

Our 3-metric framework monitors AI infrastructure spending quarterly, updated after each earnings report.

How to Track AI Capex Efficiency: 3-Metric Framework

The 3 metrics:

  1. Capex/Revenue Trend — Is spending growing faster than revenue?
  2. Capex/Depreciation Ratio — Above 2.5x may signal depreciation manipulation
  3. Revenue Growth Alignment — Is capex driving proportional growth?

Key findings:

  • Amazon has best earnings quality (1.7x ratio, shortened depreciation)
  • Google and Meta show concerning ratios (4.3x and 3.5x respectively)
  • Michael Burry's $176B depreciation thesis explained

Risk Screening Framework

Data Center Risk Signals

AI Data Center Risk Screen: 4 Signals to Watch

Not all AI beneficiary stocks are equal. This 4-signal framework identifies hidden risks:

  • Customer concentration (>10% revenue from single hyperscaler)
  • Margin sustainability concerns
  • Power availability constraints
  • Technology obsolescence risk

Concentration vs Margin Stress Test

AI Beneficiary Stress Test: Concentration vs Margin

When hyperscalers account for a large share of your revenue, you're exposed to:

  • Contract renegotiation pressure
  • Insourcing risk (hyperscalers building their own)
  • Capex cycle dependency

This analysis stress-tests AI beneficiaries for these vulnerabilities.


Supply Chain Analysis

Power Generation

Caterpillar AI Infrastructure: Data Center Power Demand

  • CAT's data center segment growing faster than traditional construction
  • Diesel generators for backup and peaking power at 1GW+ facilities
  • How power constraints are reshaping data center location decisions

ETN vs GEV: The Smarter AI Power Play

  • Why Eaton delivers 2x the margin at similar growth with half the execution risk
  • GEV trades at 98.9x P/E with 3.57% ROIC vs ETN at 37.3x P/E with 19.09% ROIC
  • Multi-company comparison: GEV, ETN, and CAT data center exposure

Fuel Cells

Fuel Cell Stock Quality Divide: BE vs FCEL vs PLUG

  • 130-point ROIC spread reveals fundamental quality divide: BE +4.5% vs PLUG -104%
  • BE's $288M quarterly Brookfield revenue proves execution; FCEL has $0 quantified data center contracts
  • PLUG's "aggressive" accounting rating and $28.1M customer dispute impairments signal distress
  • Why 90-day fuel cell deployment beats SMRs' 2029+ timeline for data center operators

Data Licensing

Reddit AI Data Licensing: Revenue Opportunity vs Legal Risk

  • Reddit's data licensing revenue from AI training deals
  • Legal risks as content creators challenge scraping
  • Business model sustainability analysis

Nuclear Power for AI

The nuclear renaissance is being driven by AI data center power demand. Hyperscalers need 24/7 baseload power that's carbon-free for ESG commitments—nuclear is the only scalable solution.

Nuclear Utility Risk Analysis

Nuclear Utility Risk Screen: 3 Filing Red Flags in VST, CEG, and TLN

  • Deal structures compared: CEG's 20-year Meta PPA vs TLN's Amazon co-location vs VST's optionality premium
  • $510M Moss Landing battery disaster at Vistra—hidden liability in Q3 filing
  • 0.9x interest coverage at Talen Energy signals debt stress
  • 970bps margin dispersion at Constellation across regions

Uranium Supply Chain

LEU vs CCJ: Why Filing Data Shows Margin Collapse vs Quality Expansion

  • LEU's 69% SWU price collapse despite uranium bull market
  • CCJ's 530bps margin expansion on contract repricing
  • HALEU monopoly positioning: LEU is only US producer of next-gen reactor fuel
  • Russia TENEX risk: LEU's license expires 2028—filing disclosure analysis

Small Modular Reactors

OKLO vs SMR: Why the $495M Filing Clue Matters More Than NRC Approval

  • $495M paradox: NuScale paid $6.9M per reactor to customers to trigger a non-binding agreement
  • NRC moat that wasn't: 18 months post-approval, SMR has one paying customer (100% concentration)
  • Cash runway reality: OKLO's $1.18B is uncommitted; SMR has committed future payments
  • Investment framework: OKLO = regulatory call option; SMR = execution bet

Company Analysis

Hyperscaler Valuation

Magnificent 7 Valuation Scorecard: December 2025

  • Side-by-side valuation metrics for AAPL, MSFT, GOOGL, AMZN, NVDA, META, TSLA
  • AI capex intensity comparison
  • Which Mag 7 stocks are fairly valued vs stretched

Enterprise AI Platforms

Enterprise AI Earnings Quality: Palantir vs Snowflake

  • PLTR's government vs commercial segment analysis
  • SNOW's consumption model and cash flow dynamics
  • SBC intensity and path to profitability comparison

AI Infrastructure Spending Context

Company2024 Capex2025 Capex (Est)2026 GuidanceFocus
Amazon$70B+$85-90BAcceleratingAWS, custom silicon
Microsoft$40B+$60B+AcceleratingAzure, OpenAI partnership
Google$50B+$60B+AcceleratingCloud, TPUs, Gemini
Meta$37B$71B$110B (2026)AI training, Prometheus

Total hyperscaler AI capex: $366-370B in 2025, approaching $500B in 2026

Key risk: When capex grows faster than revenue for extended periods, either revenue acceleration must follow or returns on invested capital will decline. Track capex/revenue trends quarterly.


How to Use This Research

  1. Start with the capex framework — Understand how to track hyperscaler efficiency using our 3-metric system

  2. Screen for risks — Apply the 4-signal framework to AI beneficiary stocks

  3. Evaluate the supply chain — Understand which companies genuinely benefit vs ride temporary tailwinds

  4. Monitor quarterly — AI capex is lumpy; trend matters more than any single quarter




Keep Learning

This hub is updated as new AI infrastructure research is published. Current coverage: 12 posts across capex, risk screening, nuclear power, fuel cells, and supply chain themes.

Coming soon: Semiconductor supply chain analysis and AI chip maker comparisons.

In This Series (13 articles)

How to Track AI Capex Efficiency: 3-Metric Framework

Learn how to monitor $370B in AI infrastructure spending quarterly with a 3-metric framework. Track capex/revenue trends, depreciation manipulation signals, and growth alignment across Google, Microsoft, Amazon & Meta. Updated December 2025 with Meta's $600B commitment and Michael Burry's depreciation thesis.

14 min read
Read article

Why ETN Beats GEV: The Smarter AI Power Play (2026)

GE Vernova's Electrification segment grew +32% YoY, but the company trades at 98.9x earnings with a 3.57% ROIC. Eaton's Electrical Americas grew +15% with 30.3% operating margins—twice GEV's margin at half the P/E. MetricDuck data shows why quality beats hype in AI infrastructure.

14 min read
Read article

Reddit's AI Data Licensing: Hidden Revenue and Legal Risk

Reddit is playing both sides of the AI data war: licensing to Google and OpenAI for $130M/year while suing Anthropic and Perplexity. SEC filings reveal 'content licensing agreements' driving 'Other revenue' growth. We analyze the dual strategy, quantify the revenue, and assess the legal risk.

15 min read
Read article

Enterprise AI Earnings Quality: Palantir (6/10) vs Snowflake (4/10)

Two enterprise AI giants, two very different earnings quality profiles. Palantir (6/10) wins on cash conversion, accounting practices, and litigation risk. Snowflake (4/10) struggles with negative cash conversion (-0.47x) and aggressive software capitalization ($228M). Our 5-pass SEC filing analysis reveals what standard screeners miss.

20 min read
Read article

Nuclear Utility Risk Screen: 3 Filing Red Flags in VST, CEG, and TLN

VST's +321% appreciation prices in nuclear optionality, but ignores a $510M battery fire disaster and antitrust lawsuit. CEG's 'capacity price boom' masks 970bps margin spread between regions. TLN's 0.9x interest coverage means debt service consumes nearly ALL operating income. We analyzed 9 filings to surface what the market overlooks.

18 min read
Read article

MetricDuck Research

CFA charterholders and former institutional equity analysts building financial intelligence you can trust.