How Many Businesses Are Using AI?
And how?
How many businesses use AI?
That seems like a simple enough question. Yet finding the right answer is deceptively tricky.
If workers at one business need AI to make the goods and services that customers pay for, while at another business AI has been implemented only to automate customer service, do we say that both businesses “use” AI? Perhaps, but clearly not in the same way.
For several years, the Census Bureau’s Business Trends and Outlook Survey (BTOS) has been asking businesses whether they use AI in the production of goods or services. The percent of firms using AI, by this definition, has crept up from 4.6 percent at the beginning of 2024 to 10 percent in September 2025.
This in production measure is restrictive, which helps explain why AI use rates from BTOS tend to be lower than in other surveys of businesses or individuals. There are many ways a business might use AI that fall outside the production of goods and services — to generate marketing materials, organize accounting or tax information, or brainstorm ideas for new products or services, for example.
Early in November 2025, however, the BTOS re-worded its question about AI to ask businesses if they use it for any business function. By this wider, less restrictive measure, the national AI use rate among businesses is 17.3 percent.
Figure 1 compares the restrictive September survey and the looser November survey by sector. It shows the percent of businesses using AI in production in green, while the percent using AI for any business function is orange. The gap between the two measures can be quite large and varies by industry.
Even with the broader definition, however, diffusion is still quite low in many sectors. Even in the Information sector, where use rates are the highest, more than six in 10 businesses are not using AI for any business function.
What’s next?
An underappreciated feature of the BTOS is that it allows us to peer into the future. The survey asks not only about current AI use but also about the expectation that a business will use AI in the next six months.
Figure 2 shows that these predictions do a good job of capturing the eventual path of AI use rates. It charts the percent of businesses currently using AI (in production) along with the percent of businesses that, six months earlier, had expected to use AI in the next six months. Essentially, we are shifting predictions so that they line up with contemporaneous measures of current use.
The six-months-ahead question correctly anticipated the change in the slope of AI diffusion that happened in early 2025.1
So what does the new, broader measure of AI diffusion portend for the future? Figure 3 shows current and expected AI use by sector for the new “any business function” question.
Use rates are predicted to climb across the economy, but some sectors are expecting bigger increases than others.2 Educational services, wholesale trade, and manufacturing all expect use rates to be about 5 to 6 percentage points higher in six months than they are today.
Other sectors, with already higher use rates, expect more modest increases. Professional, scientific, and technical services and finance and insurance expect use rates to be about 3 percentage points higher in six months.
It’s worth noting that these expected increases, to some eyes, may appear small. Not a single sector is expecting a double digit increase in AI usage. AI diffusion among business, it seems, will keep chugging along, with an expected increase nationally of about 4 percentage points over the next six months.
Measuring the diffusion of nascent general purpose technologies is a challenging task. Federal statistical agencies have done a commendable job tracking the use of AI by businesses. That data tells a story about the real-world challenges of integrating new technologies. Incorporating AI into business processes, it turns out, is not as simple as flipping a switch.
PS: Diffusion is “quite low” compared to what?!
Incorporating new technologies into a firm’s production process typically involves experimentation by both innovators and businesses. Adoption of nascent, quickly evolving technologies is often a non-trivial task.
The history of technological change provides many examples of long periods of diffusion:
It took 45 years after their first use in a factory for electric motors to reach 78 percent of capacity (Devine, 1983).
Electricity rose from 9 percent of manufacturing horsepower in 1909 to 53 percent in 1929 (Goldin & Katz, 1998).
Unimate, the first industrial robot, was installed in 1961. Sixty years later, in 2021, 12 percent of manufacturing establishments used robots and 31 percent in Transportation Equipment manufacturing (U.S. Census Bureau, 2021).
Work use of computers among employed adults rose, in the thirteen years between 1984 and 1997, from 24.6 to 49.8 percent (U.S. Census Bureau, 1999).
In contrast to hardware, adoption of digital technologies can sometimes occur much quicker. Facebook reached 1 billion monthly users 8 years after its founding in 2004. In just five years the percent of adults who use social media rose from 5% to over 40% between 2005 and 2010.
When compared with digital technologies, perhaps the speed of AI diffusion among consumers and workers isn’t completely jarring. But compared with prior waves of physical technologies, it certainly is.
Regardless, we can appreciate the remarkable speed of AI adoption while at the same time acknowledging that it is not ubiquitous among businesses, at least not yet.
View the Github with code for replicating this analysis here.
REFERENCES
Devine Jr, W.D., 1983. From shafts to wires: Historical perspective on electrification. The Journal of Economic History, 43(2), pp.347-372.
Goldin, C. and Katz, L.F., 1998. The origins of technology-skill complementarity. The Quarterly Journal of Economics, 113(3), pp.693-732.
U.S. Census Bureau, 2021. Annual Survey of Manufactures: 2021. Washington, DC: U.S. Department of Commerce.
U.S. Census Bureau, 1999. Computer Use in the United States: October 1997. Current Population Reports, P23-207. Washington, DC: U.S. Government Printing Office.
In August 2024, the percent of businesses that expected to use AI in the next six months started to increase while the trend in the current AI use rate remained on its previous trend. At the beginning of 2025, six months later, current use rates began to rise more quickly.
Sector-level estimates of AI usage from the BTOS are somewhat noisy. The standard errors for current and expected use rates are about 1 to 3 percentage points depending on the sector.