The headlines make it sound like every company on earth has already rebuilt itself around AI. The actual data says something calmer: real, deep use of AI inside big companies is still the exception, not the rule.
A recent study looked at the legal filings that S&P 500 companies must submit to regulators every year, where lying about your business is against the law. Using that stricter standard instead of press releases or executive interviews, researchers found that by 2025, only about one in five of these giant companies had AI truly built into how they run the business or make their products.
That number has grown fast since 2022, roughly quadrupling. But it started from almost nothing, so the absolute level going into this year is still modest.
Split the number by industry and the picture gets more lopsided. Technology companies make up most of the group that has gone all in on AI. Outside of tech, only a small handful of household names, including a vaccine maker, a payments company, and a large bank, have reached that same level.
The other 90 percent of the American economy, the retailers, manufacturers, insurers, and service firms that most of our readers work in, are still mostly in the early stages. Here is the part that should worry any executive under pressure to show quick AI wins.
A separate study from MIT researchers looked at hundreds of company AI projects and found that the vast majority never produce a measurable financial return, even after real money gets spent on them. Only a small slice of projects, built by teams who picked one specific, painful task and stuck with it, actually paid off. The rest stayed stuck as expensive experiments.
Put these two studies together and a clear pattern shows up. Adoption is spreading, but slowly, and most of the value so far is going to companies already built around software. For everyone else, the failures are not because the technology does not work, but because most teams try to automate too much at once instead of picking a narrow task and getting it right.
There is also a useful economic lesson in this research. Getting an AI system to be roughly right is fairly cheap, but getting it to be almost perfect costs a lot more, sometimes more than paying a person to do the last stretch of the work by hand.
That means for most companies, the winning move is not to replace an entire job with AI. It is to let AI handle the bulk of a task and keep a person for the final check, which is often cheaper, more accurate, and safer than betting everything on full automation.
Companies that move first on this narrower approach are already seeing profits improve, even before their spending on computers and software shows any change. The businesses that test small, prove value, and then scale up will be the ones that turn today's AI experiments into a real advantage over the next few years.