AI Ledger Hepburn Advisory Hepburn Advisory
Overview/Capital Ledger
The questionHow much has been committed before revenue catches up?

$1.3T committed through 2026. $176B of customer revenue against it.

The Capital Ledger treats AI infrastructure as a balance-sheet stock. It traces the spend from buyer to silicon to current workload, then compares that stock with the revenue flow now visible in the Revenue Ledger.

Cumulative infrastructure vs cumulative customer-paid revenue, 2023–26
Infrastructure : Revenue 2026 projected
$7.5 : $1
$1.3T of infrastructure committed across 2023–26, against $176B of customer-paid AI revenue over the same years. Both sides carry a projected 2026 leg: $550B of infrastructure and $145B of revenue.
On closed years only, the ratio is $24 : $1, $766B against $31B through 2025. The line’s open final point is the projection. Revenue is counted from 2024; 2023 has no published aggregate, which makes both readings marginally high rather than low.
Infrastructure capex
$766B

Infrastructure is now in the capital stack across hyperscalers, neoclouds, sovereign builds and enterprise estates.

Shared with other workloads
~$170B

A material share supports ad, search and cloud workloads funded by existing business models.

Revenue passes annual cost
~2028

Revenue passes the annual cost of owning the infrastructure around 2028, on the useful life its buyers book. Covering everything else it costs to run takes longer.

Why this matters — far more has been spent building AI than it earns back today. The whole bet rests on revenue growing fast enough to pay that spending off before the hardware wears out and needs replacing. If growth slows first, much of it never earns its money back. The Revenue Ledger combines what that revenue is today with the forward projections, so you can see how these numbers line up over time.
Where this is heading — what the revenue has to cover
Committed through 2026
$1.3T
Annual cost to own it
$236B
Covered by revenue
61%
Written off over the 5.6 years its buyers assign it, the infrastructure committed through 2026 costs about $236B a year simply to own, before power, staff, or the next generation of chips. Customer-paid revenue covers 61% of that today and 107% on the 2027 forecast. That is the whole bet in one line: the spending is already done, and the revenue has to arrive.

What the AI build-out is doing today

Modelled estimate

Infrastructure by the current physical state of each asset. The interactive spender → silicon → workload Sankey below renders the full model; this bar gives the first-read summary.

What actually earns — the hardware serving paid and free users costs real money to run as it wears out and draws power and hosting. Only the paid share is clearly bringing in revenue today.

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

$1.3T committed through 2026 → $766B spent through 2025 → $447B switched on · $319B still being built
Site rebuilt · the rebuild date, not the date the capex series moved · balance-sheet mirror of the Revenue Ledger
Scroll horizontally to explore the Sankey, or rotate to landscape.
Current asset state: Inference (Paid) Inference (Free Tier) Inference (Ad Platform) — Meta, Google, Microsoft 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, on the 2025 collected-revenue basis, from $33:$1 to ~$25:$1.
Capital allocation analysisrevenue vs depreciation
3. Revenue vs. depreciation Total AI CapEx reached approximately $330B in 2025. Current AI customer revenue is approximately $24B/yr.
QuestionWhat it requiresCurrent status
Revenue growth rate vs. depreciation3–4x annual growth to cover depreciationGap: $23.5B ($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.
Capex, depreciation and revenue over time
Each stage trails the one before it. Annual flow, in US dollars.
5.6 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 the 5.6 years of useful life the largest buyers book. AI revenue is when the investment generates returns.
The gap between depreciation and revenue is the difference between what the infrastructure costs on the P&L each year and what AI earns that year. On the 5.6-year life the largest buyers now book, revenue passes depreciation in 2026, falls behind again in 2027 as the spending peaks, and stays ahead from 2028. That is earlier than this chart used to show, for two reasons: it was drawn on a four-year life none of them uses, and on a forward revenue line the Ledger has since revised upward. Clearing depreciation is not the same as paying back. Compute, power, research and sales come out of the same revenue, so only the gross profit is available to cover the buildings and the chips. The Revenue Ledger measures that fuller test.

What to watch

Convergence means AI revenue catching the depreciation it has to cover. Three levers can close the gap. Each is shown against where it stands now.
1Revenue growth
NeedsSustained
→
Now5.1×/yr
Customer revenue grew about five times over between 2025 and 2026, faster than any layer it has to pay for. That is what took the infrastructure-to-revenue ratio from $24 : $1 to $7.5 : $1 in a single year. Holding this rate is what closes the rest of it.
Watch: OpenAI · Anthropic · Google Cloud AI · AWS Bedrock quarterly growth
2CapEx trajectory
NeedsFlat
→
NowRising sharply
CapEx must stop accelerating for depreciation to crest (~$700–750B). Guidance still climbing through FY27.
Watch: hyperscaler CapEx guidance each earnings call
3Useful life
Assumed5.6yr
→
Chip cycle18–24mo
The four largest buyers now write servers off over five to six years, and this page follows them. A new chip generation arrives every 18 to 24 months. Whether the hardware still earns at the end of that schedule is the open question.
Watch: depreciation-policy changes in 10-K filings
Detail & references
Depreciation watchquarterly per-name detail & 2030 framing
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.
Q1 2026 depreciation across Microsoft, Google, Amazon and Meta totalled ~$41B — against ~$17.5B of 2025 AI customer revenue. Per-hyperscaler detail loads below.
Quarterly, and by name. The chart above carries annual aggregate flow. These are the quarterly figures for each buyer, 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.
Structure noteshow the Sankey is built
  • 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 ($70B) — 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 $207B of assets are funded through existing business models and do not require new AI revenue to justify the investment. This is the fleet's balance-sheet asset stock (full purchase price), a different basis from the ~$170B of self-funded ad/cloud CapEx excluded in the counterfactual above, which is a capex allocation rather than the whole asset stock.
Key assumptions & tier ratingssources and confidence per input
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
Useful life (depreciation period) 5.6 years Capex-weighted from what each buyer discloses for servers and network equipment: Amazon 5.0 (shortened from 6.0 in Feb 2025, citing the pace of AI), Alphabet 6.0, Microsoft 6.0, Meta 5.5. None discloses a GPU-specific life. 1A
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.
Methodology caveat

Projections are directional arithmetic, not forecasts. Benchmarks are dated and sourced. Depreciation defaults to the 5.6-year useful life the four largest buyers disclose; every assumption is editable. Where a number appears on a Ledger page, that page is the canonical source. Full method on the Methodology page.