Data from the US Census Bureau on 583,000 businesses in 2018 shows AI adoption is slow: 8.9% use AI in any form, 2.8% use ML, and 2.5% use voice recognition
A big study by the US Census Bureau finds that only about 9 percent of firms employ tools like machine learning or voice recognition—for now.
Context & Ripple Effects
This Census Bureau survey of 583,000 firms was the first large-scale baseline for business AI use in the US, and its numbers were strikingly low: 8.9% using AI in any form, 2.8% machine learning, 2.5% voice recognition. It set the yardstick everything since has been measured against.
Two years later, McKinsey counted 50% of businesses using AI in 2022 — but found adoption had plateaued between 50% and 60%. The gap between that firm-level plateau and employee-level surveys like Gallup's rise from 21% to 40% of workers using AI between 2023 and 2025 is the story this baseline makes legible.
First-order effects
- Vendors selling enterprise AI in 2018 faced a market where fewer than one in ten potential customers were buyers at all — the addressable base for ML and voice tools was a sliver of the 583,000-firm economy the Census measured.
- Policymakers and analysts lost the assumption that AI was already widespread in business; the Census data forced forecasts to start from single-digit penetration rather than hype-driven baselines.
Second-order effects
- As firm-level counts climbed toward McKinsey's 50%-and-stuck range, measurement shifted downstream from 'does the firm use AI' to which employees use it daily — Gallup's Q4 2025 figure of 12% daily users versus 49% never-users shows adoption is uneven inside firms even where they officially count as adopters.
- The plateau in firm adoption pushed attention toward productivity proof: with half of firms nominally using AI, the question became whether it pays, and Bloomberg's coverage found the productivity evidence thin despite adoption outpacing PCs and the internet.
Third-order effects
- If the pattern holds — rapid climb to roughly half of firms, then stalling — AI diffusion looks less like a firm-by-firm transformation and more like a tool adopted selectively inside organizations, meaning headline adoption rates will keep overstating day-to-day usage.
- A durable two-tier structure emerges: an adopter majority that uses AI narrowly (information lookup, per Gallup's 60% of adults) and a near-half that never touches it, keeping demand transmission from vendor capex to broad business outcomes slower than the infrastructure buildout implies.
The trend: Business AI adoption follows a fast-then-plateau curve — single digits in the Census's 2018 baseline, roughly half of firms by McKinsey's 2022 count, and stuck there — while actual employee usage diffuses far more slowly inside the firms that count themselves as adopters.