AI Ledger Hepburn Advisory Hepburn Advisory

Capital Ledger

Data refreshed:

$766B of AI infrastructure investment, 2023–2025. Every dollar of cumulative AI capital expenditure — from spender → silicon → current workload — mapped as a balance-sheet view. Anchored from NVIDIA DC revenue: $356B cumulative (Tier 1A). Cross-checked against Microsoft, Google, Meta, Amazon 10-K CapEx disclosures. The balance-sheet mirror of the Revenue Ledger.

Cumulative 2023–25 AI CapEx · Balance-sheet view · Current state of each asset

Scroll horizontally to explore the Sankey, or rotate to landscape.
Current asset state: Inference (Paid) Inference (Free Tier) Inference (Ad Platform) Model Training Idle In build / in transit
Click any node to isolate its flows. Click again or the background to reset.
Meta, Google, and Microsoft allocate approximately $170B of AI CapEx to ad ranking, search, recommendations, and cloud workloads. These investments are funded by existing business models and do not depend on new AI revenue. Excluding them reduces the infrastructure-to-revenue ratio from $34:$1 to ~$13:$1.
Capital allocation analysis
3. Revenue vs. depreciation Total AI CapEx reached approximately $330B in 2025. Current AI customer revenue is approximately $19B/yr.
QuestionWhat it requiresCurrent status
Revenue growth rate vs. depreciation3–4x annual growth to cover depreciationGap: $24B $17.5B rev vs $41B dep
Ad/cloud workloads covering their shareExisting business models justify the CapExCovered ~$170B self-funding
CapEx growth trajectoryNew purchases stabilise so depreciation levelsAccelerating FY27 $368B guided
NVIDIA $1T target: cumulative AI-chip revenue through CY2027 = ~$1.07T at projected rates. That represents $1T of silicon entering depreciation over a 3–4yr window.
The Sankey above shows the stock (cumulative CapEx). The chart below shows the flow (annual) — and where it is heading.
The Three-Stage Lag
How capital, depreciation, and revenue flow through time
4.0 yr
CapEx is when the money leaves the bank account — NVIDIA gets paid first. Depreciation is when that money hits the P&L, spread over 3–4 years of useful life. AI revenue is when the investment generates returns. Three conveyor belts, each trailing the last.
The gap between depreciation and revenue represents the difference between annual P&L cost and annual AI revenue. Revenue is compounding but the depreciation wave has not yet crested. Convergence requires CapEx growth to flatten and revenue to keep compounding — projected around 2030 if current growth rates hold.

Depreciation watch

Quarterly hyperscaler depreciation already exceeds AI customer revenue by an order of magnitude. WSJ's 2030 model — using each company's own capex guidance and useful-life policy — projects depreciation eats a material share of net income at the largest hyperscalers if revenue does not compound through.
Loading depreciation data…
Why this sits below the Three-Stage Lag. The lag chart is annual aggregate flow; this block is quarterly per-name detail and a forward 2030 framing. Q1 2026 hyperscaler depreciation alone is approximately the same magnitude as the entire 2025 AI customer revenue line — that is the timing problem the lag chart compresses into yearly bars. Sourced from each issuer's quarterly PP&E and depreciation disclosures.

Three paths to convergence

For cumulative AI revenue to approach cumulative depreciation, one or more of these conditions must hold.
1
Revenue compounds at 3–4x annually
AI customer revenue ($19B in 2025) must grow at 80–100% CAGR for several years. Enterprise adoption, pricing power, and new product categories all contribute.
Signal to watch: quarterly revenue growth rate across OpenAI, Anthropic, Google Cloud AI, AWS Bedrock
2
CapEx growth decelerates
If annual CapEx stabilises at $700–750B, the depreciation wave crests by 2029–2030. Current guidance suggests 2026–2027 is still accelerating.
Signal to watch: hyperscaler CapEx guidance in quarterly earnings calls
3
GPU useful life extends
If GPU depreciation is extended from 4 to 5–6 years (as Microsoft and Google have done for servers), annual depreciation charges decrease by 25–35%.
Signal to watch: depreciation policy changes in 10-K filings

Structure notes

  • LHS = 10 source buckets, cumulative 2023–25 CapEx ($766B). Anchored from NVIDIA DC $356B (Tier 1A) + cross-checked against MSFT/GOOG/META/AMZN 10-K disclosures.
  • Middle = 5 nodes — what the money physically bought. NVIDIA GPU $303B cross-checks against calendarised NVIDIA DC revenue (Tier 1A).
  • RHS = 6 nodes = current physical state of each asset (balance-sheet view, full purchase price):
    • Inference (Paid) ($51B) — fleet currently serving paid API + subscription queries
    • Inference (Free Tier) ($33B) — fleet serving free-tier ChatGPT, Gemini in search, Meta AI
    • Inference (Ad Platform) ($207B) — fleet running ads, feed ranking, search, recommendations
    • Model Training ($87B) — fleet currently dedicated to training runs
    • Idle ($68B) — commissioned, powered, no current workload
    • In build / in transit ($310B) — CapEx committed, not yet commissioned. Mostly 2025 DC shell + substations.
  • Bridge to Revenue Ledger: Paid + Consumer fleet ($82B) generates ~$14B/yr COGS via depreciation + hosting + electricity. Ratio ~5.9x consistent with ~3.5yr asset life + operating overhead.
  • Ad Platform context: Meta/Google/MSFT use this fleet for ads/search/cloud workloads. These $201B of assets are funded through existing business models and do not require new AI revenue to justify the investment.
Key Assumptions
AssumptionValue usedSource / rationaleTier
NVIDIA DC revenue (cumulative 2023–25) $356B FY24 $47.5B + FY25 $115.2B + FY26 $193.7B — quarterly earnings 1A
NVIDIA revenue split (GPU vs networking) ~85% / ~15% NVIDIA segment reporting; networking = InfiniBand + NVLink 1B
Silicon as % of total AI CapEx ~55% Industry rule of thumb; cross-checked against hyperscaler 10-K CapEx vs known GPU purchases 2A
GPU useful life (depreciation period) 3.5–4 years MSFT/GOOG extended from 4→6yr for servers but GPU-specific life shorter; Meta uses 5yr blended 2B
Commissioning lag (purchase → production) 6–18 months DC construction timelines; substation permitting is the critical path 3A
AI-attributable CapEx method Growth above 2022 baseline Hyperscalers do not cleanly split AI vs non-AI CapEx; using pre-AI-boom baseline as proxy 3A
Inference fleet → annual COGS $14B/yr From the Revenue Ledger 2025 inference spend; cross-checks at ~4x ratio to fleet value 2A
Ad Platform fleet allocation ~$170B Meta ($55B GPU + ads infra), Google (TPU fleet for search/ads), MSFT (Bing/Copilot). Largest single workload category. 3B
China NVIDIA GPUs (estimated) 474K H100e Epoch AI tentative estimate; export controls make this inherently uncertain 3C
Idle compute (utilisation gap) ~$50B Residual after allocating to known workloads. Public utilisation data is limited — CoreWeave S-1, earnings commentary 3C
Tier key: 1A/1B = directly sourced from filings or earnings. 2A/2B = derived from sourced data with clear methodology. 3A/3B/3C = modeled estimates with stated assumptions.