Anthropic’s revenue run-rate hit $65B in July, up 7× in a year. That’s roughly five years from founding. Google and Meta took fourteen. Apple and Microsoft took more than thirty. We charted the whole club — with the honest asterisk built in.
Legacy figures are the first full fiscal year revenue crossed ~$50B (company filings / Macrotrends). AI-lab figures are reported annualized run-rates, not audited fiscal-year revenue — see the caveat below. Salesforce ($37.9B, FY2025) hasn’t crossed yet and is shown for scale.
One company breaks the frame. Nvidia took ~31 years to reach $50B (FY2024) — slow, by this chart. But it is the sharpest proof that the AI boom compresses time at every scale: on its August 2026 blowout, Nvidia guided to a $108B quarter — a ~$430B annualized run-rate — with 70%+ growth still ahead (CNBC). The labs are the fastest to $50B; Nvidia is the fastest above it — scaling from $50B toward half a trillion in a handful of years. Compute and models are the same demand curve, read from opposite ends.
The old record holders were the fastest-scaling companies humanity had ever built — and they still needed the better part of two decades. Google and Meta got to $50B in 14 years by wiring the entire consumer internet to an ad auction. Amazon took 18 by pouring every dollar back into warehouses and cloud. So what compressed 14 years into 5? Four things, all structural — which is why this may repeat for the whole frontier cohort, not just Anthropic:
A frontier model ships to the entire planet through an API the day it’s ready. No stores, no supply chain, no hiring a sales force ahead of demand. Adoption is a code change on the customer’s side, not a rollout on yours.
There is no subscription ramp or seat-by-seat expansion lag. The moment a customer runs more tokens, the revenue is booked. Consumption pricing means revenue scales at the speed of usage — and usage is compounding.
Enterprises aren’t waiting to be convinced a category exists. They are moving existing software, services and headcount budget onto AI — a fast substitution into pools that were already funded, not a slow greenfield build.
The old guard was paced by how fast they could self-fund the next factory or data centre. Frontier labs raise tens of billions at a time, so compute — the one hard limit — is bought years ahead of the revenue it serves.
Every legacy bar on that chart is a full fiscal year of audited GAAP revenue. Anthropic’s $65B is an annualized run-rate — a recent month multiplied by twelve. Google’s 2012 $50B came with twelve months of known retention; a run-rate comes with none. The two open questions the number can’t answer: churn (does the usage stick, or is it experimentation that lapses?) and concentration (is this a broad base, or a handful of mega-deals that can renegotiate?). Run-rates flatter fast-growing businesses precisely when growth is steepest. Hold the velocity as real and the durability as unproven — and note that even discounting the run-rate by half, Anthropic would still be the fastest ascent on this chart by years.
The chart is a stat; the consequence is a procurement decision. When the #1 and #2 fastest ascents to $50B in history are both foundation-model vendors — and they’re pulling away from everyone else — the enterprise-AI wallet has effectively become a two-horse race, and the horses are compounding. Three implications:
If your AI strategy quietly standardized on one vendor 18 months ago, the ground has moved. A vendor growing 7× a year is not a hedge you bolt on later — it’s a platform decision to make deliberately now.
Revenue this steep funds the next model and the next price cut. That’s good for buyers, but it also means lock-in gets more expensive as the frontier moves. Keep a multi-model path open (see our Killer App Watch on the middleware that makes switching cheap).
The revenue trajectory is real and sourced. The valuations riding on it price in years of continued 7×-ish growth that no company has sustained at this scale. Underwrite the capability; stress-test the multiple.
The most valuable companies of the last generation took 14 to 47 years to reach $50B. The AI labs are claiming the same scale in 5 to 10 — on a run-rate basis, with real questions about durability, but a lead so large the caveats can’t close it.
The scaffold got built in record time. The open question was never whether AI would generate revenue — it’s how fast, and to whom. Now we have the first honest answer: faster than anything before it, and concentrating.