이미지: X — 벤치마크·평가 화면 갈무리
Summary
- Epoch AI and Ipsos surveyed 1,103 US workers on how they pay for AI tools at work
- About 47% of respondents used free plans, while only about 20% had employer-paid subscriptions
- Only computer, engineering, and science occupations stood out with a 67% employer-support rate
- 조사기관·시점
- 엡록AI(Epoch AI)·입소스(Ipsos), 2026년 7월
- 전체 표본
- 미국 성인 근로자 1,103명
- 업무용 AI 최근 7일 사용자
- 469명
- 전체 무료 플랜 비중
- 약 47%
- 전체 회사 제공 구독 비중(단독)
- 약 20%
- 컴퓨터·엔지니어링·과학직군 회사 지원(겸용 포함)
- 67%
- 경영·비즈니스·예술직군 회사 지원(겸용 포함)
- 41%
- 무역·생산·운송직군 무료 플랜 비중
- 약 62%
Nearly Half Work with the 'Free Version'
In a survey conducted by Epoch AI together with Ipsos in July, targeting 1,103 US adult workers, just under half of the 469 respondents who had used AI for work in the past 7 days handled their tasks entirely on free plans. Among all respondents, roughly 47% used free plans, while only about 20% had their subscription fully covered by their employer. Even adding personal-paid subscriptions (about 10%) and mixed employer/personal payments (about 6%), subscriptions involving the company in any form still fall short of a third of the total.
Epoch AI summarized: "For most US workers, the default for AI at work is what the free tier provides." The gap becomes even clearer when broken down by occupation.
Science and Tech Occupations Live in a Different World
Among computer, engineering, and science occupations (sample of 81), about 52% had their subscription fully paid by their employer, rising to 67% when combined with mixed employer/personal payments. Free plan usage, meanwhile, was only about 20% — less than half the overall average. In roles like developers and researchers, who use AI tools directly for writing code or running experiments, employers paying for subscriptions is effectively the standard.
By contrast, in management, business, and arts occupations (sample of 207, excluding computer, engineering, and science), that share dropped sharply to 41%. Free plan usage, at about 43%, roughly matched the employer-support rate. This suggests that even among similarly classified "office workers," how much employers are willing to pay depends heavily on whether development or research work is involved.
Reliance on free plans grew further down the list: sales and clerical occupations (sample of 72), service occupations (sample of 53), and trade, production, and transportation occupations (sample of 49). In trade, production, and transportation, free plan usage reached about 62%, while employer support was only about 5%.
Subscription Patterns by Occupation
| Occupation | Sample (n) | Free Plan | Employer-Supported (incl. mixed) |
|---|---|---|---|
| Computer, Engineering, Science | 81 | 20% | 67% |
| Management, Business, Arts | 207 | 43% | 41% |
| Service | 53 | 55% | 21% |
| Sales, Clerical | 72 | 58% | 18% |
| Trade, Production, Transportation | 49 | 62% | 8% |
The "employer-supported" figures in the table combine responses for "fully employer-paid" and "mixed employer/personal." The confidence interval is at the 90% level, and the sample was weighted to be representative of the US population of wage workers aged 18 and older.
Context: Coinciding with an AI Price War
These findings coincide with a period when OpenAI and Anthropic have sharply cut prices on their flagship models. As previously reported by METAL LAB, OpenAI cut both input and output token prices for GPT-5.6 Luna by 80% each, while Anthropic released Claude Opus 5 at half the price of its top-tier Fable 5 model, responding to price competition from Chinese models. While falling token prices could in theory lower the bar for companies to offer paid subscriptions, this survey shows the opposite: in most occupations, employers still aren't opening their wallets, and workers are largely finding their own ways to make do within free tiers.
Epoch AI noted that the survey followed the BLS's top-level categories under the 2018 Standard Occupational Classification (SOC), and that given the relatively small sample sizes per occupation (49 to 207), more weight should be given to the direction of the gaps between occupations rather than the individual figures themselves.
Editor's Take
What this data shows is simple: the perception that AI subscriptions are "developer tools" companies pay for still runs strong. The 67% figure for computer, engineering, and science occupations means AI subscriptions have already settled into IT budgets much like software licenses. The roughly 20% figure for other occupations means AI is still classified as a "personal productivity tool" from the employer's perspective. In effect, within the same company, the engineering team may be billing AI to the corporate card while the marketing team pays out of pocket.
Anyone who has felt this gap firsthand won't find these numbers surprising. Whether at startups or large corporations, approval for AI subscriptions tied to coding tends to pass relatively easily for engineering teams, while marketing, planning, or HR teams requesting the same budget often hear "just use it on your own." The reason is simple: it's easier to prove the value of developer tools through productivity metrics like code output or bug-fix speed, whereas other occupations struggle to quantify the difference AI makes.
For HR and IT managers at Korean companies, this survey is worth using as a benchmark. It's worth checking whether the gap in AI subscription support between development and non-development teams at your own company is this wide. If a department relies on free plans more than half the time, it's often not because employees aren't using AI — it's because the company isn't paying for it. Given that token prices keep falling, the cost of rolling out paid subscriptions company-wide is now far lower than it was a year or two ago — leaving less reason to keep postponing a budget review.
What happens next seems predictable. As the price war continues, companies are likely to move toward expanding company-wide licenses regardless of department, at which point the very question this survey raises — who pays for AI — may become moot. That shift, however, won't happen within this quarter — budgeting cycles simply don't move that fast.
