By every measure, U.S. companies are winning on AI adoption—but a series of high-profile snafus shows they’re getting pummeled by costs

American firms are quickly outpacing their international counterparts at getting their employees to use AI. As reward for their efforts, they are also running up a monstrous bill, with lavish outlays on token costs, proprietary AI platforms, and just plain goofs, sometimes worth hundreds of millions of dollars. Corporate America has set itself a goal to comprehensively integrate AI across its employees’ workflows, and by some measures, their efforts have been a wild success. Around half of U.S. workers now use AI in their roles at least a few times a year, up from less than 40% a year ago, according to recent Gallup polling. The rate of adoption has outpaced that of rival firms abroad, according to research from the Brookings Institution published in March, which found 43% of U.S. workers now use AI on the job, compared to 32% among their European counterparts. The gap holds at the firm level, too, with 7% of U.S. companies now using AI for production of goods and services compared to just just 4% of European firms. Bosses have been under pressure to figure out how they can turn AI tools and experiments into profits. And while AI usage has in most cases yet to translate to an acceptable return on investment, the technology is starting to show up in productivity statistics—albeit modestly. The Brookings research found differences in AI use led to aggregate time savings worth 2.3% of working hours in the U.S. versus 1.4% in Europe. It’s a potentially meaningful edge—one that companies hope will accelerate once learning curves are overcome and as the technology improves. The problem for firms now, though, is they’re making a bet that could take years to pay off, and the bills are already coming due. U.S. companies are beginning to reckon with the reality that even if experimenting with AI seems free, it usually isn’t. Loosely monitored usage limits and a trial-and-error process of figuring out how AI can improve bottom lines has led to a series of high-profile snafus,