A study out of MIT in 2025 found that 95 percent of enterprise AI projects return nothing measurable. The number went everywhere, and the people spending the most on AI would rather it hadn’t. Like most numbers that travel that fast, it is probably too clean. The figure appears in a single sentence of the report, the data behind it was never released, and people who went looking have said so. What interests me more is that the picking-apart changed nothing. The number kept moving because it told the people repeating it something they already suspected. It matches what I see, too, and it is landing in the same year that every earnings call and half my group chats treat AI as the most important thing ever to happen to work.
I know the shape of this moment because I was paid to stand inside the last one, and the metaverse is the one that fooled the serious rooms. Facebook renamed itself Meta and has since lost more than $80 billion building it. McKinsey sized the prize at $5 trillion by 2030, and Citi went as high as $13 trillion. Gartner announced that by 2026, one in four of us would spend at least an hour a day inside it. JPMorgan opened a lounge in a virtual world called Decentraland, complete with a cartoon tiger and a portrait of Jamie Dimon. These were not speculators on a message board. They were the most credentialed institutions we have, and they were certain.
Across the table from me were consumer brands with a real story to tell, following the next generation of customers into a place those customers already spent their evenings. And then there were the banks, which had no such story and came anyway, trying to work out where to put a branch in a world with no streets. They were paying my team hundreds of thousands of dollars to map how they would show up in a place that did not exist yet, and I helped draw those maps.
Then the room went quiet, and Meta wrote down the losses and changed the subject to AI. A report went around that Decentraland, valued at more than a billion dollars, had 38 daily users. Virtual land that traded in the millions one year was nearly worthless the next. Gartner's deadline arrived in 2026 with no one spending an hour a day anywhere, and my clients stopped mentioning the metaverse the way you stop mentioning a diet that didn't take. There was no reckoning. Everyone moved on, and the maps we drew went in a drawer.
What I took from it is that the useful question was never whether the metaverse was real. Some version of it, computing that is more immersive and less flat, may still arrive. The question I should have pressed was real for what, and by when. The hype took a plausible idea, married it to the most exciting possible version of itself, and set the clock to now. That is the move every cycle makes.
The durable shifts are boring at the start, and easy to miss, because they show up in flows before they show up in headlines: money moving into some unglamorous corner, good people quitting safe jobs for a problem that does not have a name yet. Cloud spent a decade that way, as line items in IT budgets and engineers changing jobs, before anyone called it a platform shift. The money that poured into the metaverse was less talent chasing something real than fear of being late to something everyone had agreed was coming, and when excitement becomes the safe position, the one that keeps you in the room, it is worth distrusting. The metaverse was all headline and no flow.
I want to be clear that AI is not the metaverse, and I am not hedging when I say it. The metaverse asked you to believe in a world that did not exist yet. AI is already here. I use it every day, I have built things with it I could not have built alone, and it is doing real cognitive work inside real companies right now, not in a forecast. The last cycle was a bet on a place. This one is a capability you can hold in your hands today.
Which is why the hype worries me, and why the 95 percent is worth reading correctly. The mistake underneath it is the same one I made about the metaverse: a real capability, married to the most exciting possible application, with the clock set to now. The version of AI that remakes your company in eighteen months is the forecast I already wrote once, about a world with no streets. And I have seen where that ends. When the metaverse collapsed, it buried its own real kernel with it, and immersive computing got set back years by the people who oversold it. The danger for AI is the same: the loudest promises write a check the real thing cannot cash this year, and when it bounces, the backlash can take the durable part down with the loud one. What the number is pointing at, whatever its provenance, is the distance between what AI can already do and what it was promised to do by now. Precision about that distance is how you keep the real thing from being discredited by the noise around it.
I am not a skeptic. I am an optimist who has been wrong enough to be careful, reading this cycle with the memory of the last future I was certain about. The signal is in the flows, and the flows move slower than the people selling you the future will admit. I know. I used to be one of them.