Photo by Steve Johnson on Unsplash
 

A question I’m getting asked a lot recently is whether all this fuss about AI is just a bubble. My standard answer is that two things are happening here, and it’s wise not to confuse them. One is the sky-high stock prices for companies closely involved in AI. The other is how much AI itself will actually end up mattering a great deal to the US economy.

I have no particular insight into the short- or medium-term dynamics of the stock market, much less individual stocks. But I wouldn’t be much surprised if stock prices for some of the currently leading AI companies drop substantially at some point in the next few years, nor would I be surprised if some AI companies I’ve never heard of see their stock price boom. It’s also perfectly possible that the stock price of some leading AI companies falls, but the reality of AI in the production of goods and services rises.

The stock market isn’t actually part of the real economy of goods and services. This is literally true: the stock market isn’t part of the gross domestic product, because when stock is bought and sold, there is just an exchange of an asset, but nothing is actually produced. For the same reason, building a new house is counted as part of GDP (that is, something is produced), but selling an existing house is not part of GDP (it’s just an exchange of an asset). There’s a saying that helps to clarify the separation between stock market prices and the real economy: “The dot-com stock market boom of the 1990s was a bubble, but the Internet was not a bubble.”

Of course, although the stock market and the real economy are not the same, there are feedbacks between them. If the stock market drops, I’ll see it in my retirement account, and the signal of a declining stock market will make it more expensive for firms to raise capital. Conversely, the stock price for AI companies will depend on the extent to which their products create value for firms and individuals. Here, I’ll focus on some recent survey evidence on how broadly AI tools are being adopted in corporate America–and for all the talk about AI, the adoption rate so far is lower than you might expect.

As one example, the US Census Bureau does a Business Trends and Outlook Survey based on responses from 1.2 million US businesses. Here are some results from the BTOS as updated on November 20, 2025. Here’s a figure showing businesses that are using AI tools: the blue line is those that have used such tools in the last two weeks; this has risen from 5% of firms at the start of 2024 to about 10% near the end of 2025. The orange line shows firms that expect to use AI tools in the next six months. The lines are trending up, but not skyrocketing. A more detailed breakdown by industry shows higher levels of use in information industries, finance, and professional/scientific/technical services, and near-zero levels in manufacturing and retail.
 

Here’s a breakdown of the Census data by size of firm. As you can see, the use of AI by the biggest firms (top line) seems to have levelled off or even dropped a bit in the last six months or so. The category of firms with the biggest ongoing rise in AI use is the small firms (lightest blue line) with 1-4 employees.
 

Here are some other survey results, these from the National Opinion Research Center at the University of Chicago. Malihe Alikhani, Ben Harris, and Sanjay Patnaik report the results in “How are Americans using AI? Evidence from a nationwide survey” (Brookings Institution, November 25, 2025). The sample size here is about 1100 people, selected to be a nationally representative sample. They emphasise that those with higher education levels are more likely to be using AI tools.

This figure shows how AI tools are being used in the workplace, broken down by education level. The most common use is for those with a BA degree, using it for writing and editing documents. But many of these groups are reporting 10% or fewer people who use AI tools, “supported by your institution,” in the workplace. Again, the current use of these tools is lower than one might expect.

Indeed, the NORC survey asks some follow-up questions on whether workers feel that the AI tools are improving their productivity, and the results are decidedly mixed:

The impact of generative AI on worker productivity is often unclear, even to the workers themselves. Only 19% of all respondents report that AI increased their productivity in their daily tasks, and only 4% say it increased their productivity significantly. Even among respondents with a bachelor’s degree or more, just 28% say that AI increased their productivity in daily tasks. More than one in five respondents report that their daily productivity remained the same (22%) and over half of all respondents say they are either not sure about the effect of AI on their productivity or say it does not apply to them (53%).

Putting these kinds of results together, it’s perhaps no surprise to see a story in the Economist magazine (November 26 issue) headlined “Investors expect AI use to soar. That’s not happening” and subtitled “Recent surveys point to flatlining business adoption.” The article whimsically suggests that the GPT in ChatGPT might stand for “Generally Paused Technology.” Using the Census data above on the size of businesses using AI tools as well as other survey and research results, the article estimates that share of Americans using AI at work has recently declined slightly.

Of course, the uncertainties around how to design AI tools for business and personal use are very large at this stage. Sometimes, when a remarkable new technology is being adopted, there is a period of several years where many people are learning about the technology and multiple applications are being developed, but this doesn’t show up in the big-picture statistics. Then at some point, some of those applications gain traction with a broad group of users and the technology takes off. But at least so far, with the current abilities of AI and the AI applications currently available to US firms, that launching point hasn’t happened yet.

More By This Author: