AI Ledger Hepburn Advisory Hepburn Advisory
Overview/Revenue Ledger
The questionWhat money has actually arrived from customers?

What customers are paying for AI.

Every other number on this site is measured against this one. It counts money that has arrived. It leaves out list prices, free users, money that has only been pledged, and one company's promise to spend with another.

Customer-paid revenue, 2026 gross basis
Customer revenue, 2026 2026 projected
$145B
What customers pay for AI across 2026, from every buyer, gross of the margin resellers and cloud channels keep on the way through.
A narrower measure sits underneath it: the companies with a sourced revenue figure are running at $138B a year today, across 91 of them. That counts what AI companies book, not what buyers hand over.
2026 · the 1× baseline
$145B

Customer-paid AI revenue across 2026, gross of channel margin. This is the 1× denominator every multiple on this site is measured against. $24B was collected in 2025.

2027 forecast
$251B

Where that run-rate carries booked revenue by 2027. Modelled company-by-company from each provider's ARR and growth, not a single flat multiple.

The margin
~40%

Gross margin: what's left after the compute it costs to serve the models. The rest has to cover training, R&D and sales, which today it doesn't. The gap is cash burn. This one is an editorial estimate. No provider discloses it cleanly and the credible range is wide; the Ledger holds this figure as a stated position and shows what the alternatives do.

Why this matters — revenue is the only layer in the Ledger that does not require a model to exist. It is the denominator for the system: the cash line that capital, compute, usage and power must ultimately justify. The Capital Ledger shows the infrastructure spend this revenue has to pay off over time.
Where this is heading — how big does the revenue have to get?
AI revenue by 2030
$919B
Needed to pay back the build-out
$1.5T
Revenue reaches a trillion
2031
Likely AI revenue Revenue needed for the build-out to pay for itself Log scale · shaded band is the forecast range
Revenue has exploded. It still has further to go than it has come. Customer-paid AI revenue reaches roughly $919B by 2030 on the Ledger’s own forward path — about 39 times what it was in 2025. But the infrastructure being built to serve it has to be paid for. Written off over the 5.6 years its owners actually book, and covered out of gross profit rather than revenue, the build-out needs revenue of about $1.5T a year by 2030 — still 1.7× what it is likely to be.

One number in that calculation is a judgement, not a measurement. Nobody discloses a clean gross margin for serving frontier models. The published figures treat training costs, free usage and volume discounts differently at every provider, and the credible range runs from about 30% to 50%. The Ledger holds it at 40% as an editorial position and publishes the swing: at the low end the bar is $2.0T, at the high end $1.2T. Revenue falls short of all of them.

The gap is closing, and quickly: it was 14.6× in 2025. Revenue compounds faster than the bar rises. But the bar rises every year the spending continues, and revenue does not reach a trillion dollars until 2031, which is the part that takes longer than the growth rate alone suggests.

Revenue Flow

Site rebuilt · how customer money routes from the buyers who pay — Consumer, AI Natives (Cursor, Glean, Perplexity), Enterprises & Govs — through the channels to the model providers earning it (OpenAI, Anthropic, Google/Gemini and smaller labs). AI-specific revenue by buyer segment, channel, provider, and cost outcome.

Scroll horizontally to explore the flow chart, or rotate to landscape.

Provenance tier: 1 Sourced (A = first-party disclosure, B = corroborating sources) 2 Derived (A = deterministic calc from Tier 1, B = triangulated) 3 Projected (A = anchored extrapolation, B = interpolated, C = scenario assumption) 4 Editorial (round-number estimate, carried only until better data arrives) · hover any node for source citation
! Scope. This flow chart shows model providers earning AI-attributable customer revenue in the year. Labs that don't directly monetise their models are excluded; Meta (Llama, open-weight, ad-funded indirectly) is the canonical example. Their AI spend appears on the Capital Ledger (capex) and Compute Ledger (chip purchases) instead.
! On comparing frontier revenue. Anthropic reports gross (principal basis, including cloud-channel resale). OpenAI states it reports net of its Microsoft revenue share. The two headline figures are not directly comparable. OpenAI estimates the gap at ~$8B; that estimate is unaudited and contested.
Click or hover a node to see flow details and tier provenance
Detail & references
How this chart is builtchannel margins & per-archetype routing
? Why do the middle columns shrink? Each column shows the dollars that flow through it. Channels (Hyperscalers, and traditional SaaS vendors reselling AI features) retain a margin before passing revenue to Model Providers. Hyperscalers clip ~20%, Trad. SaaS ~60%. That retained margin flows directly to Generated Cashflow (bottom right), bypassing Model Providers entirely. So each column is smaller than the one before it — the drop is the channel margin.
i Per-archetype channel routing. Each provider's revenue routes to channels per its entity archetype. Frontier labs send ~95% of API revenue direct (5% via hyperscaler resale); AI Natives ~90% direct; Enterprise SaaS ~50/50. The Who Pays column splits AI buyers into AI Natives (heavy API consumers like Cursor / Glean / Perplexity) and Enterprises & Govs (Fortune 500, regulated, sovereign). See methodology § entity archetype taxonomy.
Where each buyer's money landsnet-to-provider routing table
ARR by companythe sourced-ARR companies behind the headline
91 companies with sourced ARR summing to ~$138B (modelled 2027 recognised revenue ~$230B). Full company-by-company table loads below.

What to watch

Three forces decide whether revenue pays the build-out off: how fast it grows, what it costs to produce, and whether it can fund itself. Each is shown against where it stands now.
1Revenue growth
NeedsSustained
→
NowRising fast
Revenue roughly tripled in a year. It has to keep compounding like this for several years to grow into the cost base.
Watch: OpenAI · Anthropic · Google quarterly revenue growth
2Gross margin
NeedsWidening
→
NowThin
Gross margin is positive but thin. Serving compute eats most of each revenue dollar. It has to widen as inference costs fall to cover training, R&D and sales.
Watch: inference price per token · provider gross-margin disclosures
3Self-funding
NeedsSelf-funded
→
NowVC-funded
Cash burn is still covering the gap between revenue and cost. Convergence needs revenue to carry the cost base on its own.
Watch: frontier-lab burn rate and path to breakeven
Methodology caveat

Projections are directional arithmetic, not forecasts. Benchmarks are dated and sourced. Channel and buyer routing defaults to the per-archetype weights set out on the Methodology page; every assumption is editable. Where a number appears on a Ledger page, that page is the canonical source. Full method on the Methodology page.