The AI Data Treadmill

A small group of companies is currently generating enormous, non-recurring revenue by selling training data, expert judgment, simulated environments, evaluations, and physical capture to frontier AI labs. Some are closing tens of millions in purchase orders per month with teams of fewer than a dozen people. The cash is real. The contracts are not forever.

What is actually being sold falls into six categories: hours of labeling (already dying), high-end expert judgment, simulated worlds for agents to practice in, benchmarks and red-teaming (the referees), physical sensor data from real workers, and licensed rights to existing corpora. Almost everything arrives as a one-time purchase order against a deliverable. Nothing renews by default.

The deeper dynamic is a treadmill. Once models solve a task reliably (roughly 70% success), that task is discarded. The product depreciates by succeeding. Labs keep paying because no one holds a durable capability lead — every price tier collapses and open models trail the frontier by only months. Data vendors sell the fuel for a race no one can permanently win.

The companies that scaled fastest did not win because they understood data. They won because they already had systems for verifying strangers at scale (recruiting and talent platforms). Verified judgment, not raw data, was the scarce resource. Looking ahead, durable value is likely to concentrate in domains where verification remains hard, in provenance and trust infrastructure, in physical capture, and especially in private “gyms” — high-fidelity environments that enterprises can run inside their own walls so their proprietary judgment never leaves the building.

This Ledger Entry expands how readers think about the economics of frontier AI by showing that the data market is a treadmill powered by the scarcity of human verification — products depreciate by succeeding, so durable value concentrates in hard-to-verify judgment, provenance, trust, and private enterprise environments rather than generic labeling.