New Federal Survey Data Tie Faster AI Adoption to Weaker Job Growth in the Industries Embracing It Most

Goldman Sachs economists, in a research note published Monday updating their monthly tracker of U.S. AI diffusion, report that 24.6% of American business establishments now say they use artificial intelligence to produce goods or services, according to the Census Bureau's Business Trends and Outlook Survey. That is up 0.8 percentage points from September, and establishments expect the figure to reach 28.6% within six months, based on their own stated plans. The survey, run by the Census Bureau as a high-frequency, nationally representative check on business conditions, has tracked AI use climbing steadily across firm sizes and sectors since late 2025.

The breakdown shows adoption nearing 50% in the information, professional-services and finance sectors. More than half of establishments with over 250 employees now expect to be using AI within six months, a far higher share than among the smallest employers — the same large-firm-first pattern that has defined AI diffusion since the technology moved into production use.

A different ruler gives a very different number

Set against Ramp's independently compiled AI Index, which tracks actual payments for AI products and services across more than 70,000 U.S. businesses on its corporate cards, the picture looks different in scale, if not in direction. Ramp puts overall AI adoption at 55.9% as of September, up 12.1 points from a year earlier — more than double the Census figure — largely because it counts any paid AI subscription rather than self-reported use in production. Read side by side, the two measures suggest that the oft-cited "share of companies using AI" depends heavily on where the line is drawn between trying a tool and running it in production; the latter, which is what Census asks about, is climbing far more slowly than card-spend data implies.

Ramp's sector ordering largely mirrors Census's even where the levels diverge: information leads at 79.9%, finance and insurance at 73.7%, manufacturing at 60.9%, while accommodation and food service trails at 33.0%. Intensity of spending varies even more than the adoption headline suggests — the median paying firm spent just $15.70 per employee per month on AI in September, compared with $813.43 for the top 10% of spenders and $9,567.53 for the top 1%, and the blended price per million tokens from OpenAI and Anthropic combined fell 25.4% over the same period even as token volume jumped more than 50%.

The employment link Goldman is now tracking monthly

The sharper finding in Goldman's note concerns jobs rather than tools. Comparing AI adoption rates against employment trends at the subsector level, the bank's economists describe what they call a statistically meaningful relationship: subsectors adopting AI faster have shown weaker non-manufacturing job growth and higher unemployment over the past three months. Separately, AI-attributed corporate layoffs — job cuts companies or trackers have explicitly tied to automation — totaled roughly 4,000 in September alone and about 207,000 cumulatively since May 2023, by Goldman's accounting. On third-quarter earnings calls, roughly 23% of Russell 3000 companies that have reported results so far paired mentions of AI with labor-related language, a sign that more executives are now willing to describe AI's role in headcount decisions directly to investors rather than folding it into generic efficiency talk.

None of this settles whether AI adoption is causing the job-growth gap or is simply concentrated in industries — tech, media, professional services — that were already restructuring for other reasons; Goldman's economists describe the relationship as correlational, and the bank does not claim otherwise. But set against academic estimates that productivity rises an average of 23% in roles where generative AI has been deployed, and company-reported efficiency gains running closer to 32%, the data argue against the idea that nothing measurable is happening on either side of the ledger. Both the productivity payoff and the employment reallocation are now showing up in the same monthly data, at the same time.

The timing matters for corporate boards still deciding how fast to move. Gartner forecast in the spring that global AI spending would jump 47% this year to $2.59 trillion, with enterprises more than doubling what they spend on generative AI models and agents — even as Gartner analyst John-David Lovelock warned that most organizations remain unenthusiastic about AI-driven disruptive change and favor tactical projects with incremental, measurable payoffs over wholesale reinvention, an approach he said "persists despite AI hype." The Census and Ramp numbers released this fall suggest that caution is itself becoming measurable: adoption keeps broadening into the median firm, prices per token keep falling, and the employment effects once dismissed as speculative are starting to show up as a line item that economists, regulators and investors can now track from one month to the next.