When the numbers don’t match the story, there’s a certain kind of anger that settles over a manufacturing boardroom. The story was believed by everyone. The pilot did its job. The consultants were sure of themselves. The demos really were impressive. Even so, the quarterly results look almost the same now that a year has passed than they did before any of your changes.
Actually, that’s pretty much where a lot of corporate America is right now. Generative AI has used up operating budgets, reorganized IT departments, and caused a huge amount of excitement inside companies. What it hasn’t usually done is give a measurable return.
A survey of CEOs by PwC in 2026 found that 56% said AI had not led to either more sales or meaningful cost savings. Not many people (12%) said they had both outcomes at the same time. These numbers don’t fit well with the bigger story, which is that AI is changing industries, changing how competition works, and needing urgent, large-scale investment. The news says that every company that doesn’t move quickly is falling behind. The survey results suggest that things are a lot messier on the ground.
A venture capitalist named Chamath Palihapitiya made this point more clear in a recent episode of the podcast All-In. He said that if you throw out companies like Nvidia, the big cloud providers, and chipmakers that are actually building AI infrastructure and look at what the rest of corporate America has made from spending on AI, you won’t find much. Based on his estimate, only a small part (about 2% to 0%) of the rise in the S&P 493’s earnings per share since generative AI became popular can be attributed to real AI-driven productivity. The rest comes from pricing power caused by inflation and share buybacks. Not quite the revolution that the flyers said it would be.
Pilot purgatory is even a word used in the business world to describe the trap many companies have fallen into. A company does a controlled test of AI. It works well enough by itself. Then the work to make it work for a real company stops because of old systems, bad data, and employees who weren’t asked for their opinion. The pilot just sits there, technically successful but not doing anything. At the same time, people keep spending because stopping feels like giving up when everyone else seems to be pushing forward.
The only thing that’s changed is who is paying attention now. Early investments in AI were often put in innovation budgets and looked over by tech teams that were okay with long lead times and unknown outcomes. Over time, that money has moved into the main operating budgets. That means that now the people who are asking the questions are the chief financial officers. Instead of showing off skills, CFOs are more likely to want to know if the investment is making more money than a Treasury bond would.
That’s the uncomfortable way Palihapitiya put it: capital costs money. If a company spends twice or three times as much on AI and still doesn’t know what the benefits are, it would have been better to leave that cash on the balance sheet at some point. This is a direct way to say it, but it’s the kind of thinking that tends to focus minds when margin pressure is already building from other directions, like prices for energy, labor, and the supply chain.
All of this doesn’t mean AI is weak. Companies that sell AI infrastructure are doing really well, and it’s a good point that historically, it takes years for productivity gains to show up in the overall economy. The internet didn’t instantly change how much money companies made either. But even though the comparisons are valid, they seem a bit old. The groups that spend the money need to be able to point to something real at some point.
It seems like the industry is getting close to that turning point. For companies that moved quickly and built with confidence, the first wave was a reward. It looks like the second wave will reward something much less exciting: being able to show that any of it worked.
