A few days ago Microsoft’s CEO announced that Copilot is becoming “a new OS for work.” Every model, every form factor, every task. I read it twice to make sure I wasn’t imagining it.
This is the same Copilot I had to fight to get out of my own Microsoft subscription.
The cheaper plan without it only appeared after I’d started to cancel, buried at the bottom of the exit door. I saved twenty bucks a year by nearly quitting. And now the thing I paid extra not to have is supposed to be the operating system for all of work?
I use AI every day as a digital assistant. It’s a lot like having a virtual assistant, only digital, and I still do the writing. It’s useful, and some days it’s damn impressive.
But what I’m watching right now is a stampede, and I’ve seen this movie before. So has everybody over fifty, if they’d stop and remember.
What is the Gartner hype cycle?
The research firm Gartner has been drawing the same curve since the mid-1990s, and it keeps being right. Every hyped technology goes through five stages.
First comes the technology trigger, when something new appears and people get excited. Then the peak of inflated expectations, when the excitement outruns reality and everybody’s sure this changes everything by next quarter. Then the trough of disillusionment, when the promised results don’t show up, the money dries up and half the companies die. After that comes the slope of enlightenment, when the survivors figure out what the thing is good for. And finally the plateau of productivity, when it becomes a normal, boring, useful part of life.
Almost nothing escapes the trough. Some technologies take longer to reach it, and some fall harder when they hit it, but it comes for nearly every one of them. The people on stage never mention it. Vendors don’t put the trough on a slide. They sell the peak, because that’s where the money changes hands, and the customers who bought in get to find the trough on their own with the invoices still arriving. I think that’s the most dishonest part of every tech boom, and AI is no exception.
Watch: Gartner Hype Cycles, Explained, Gartner
Where is AI on the hype cycle right now?
By Gartner’s own chart, generative AI passed the peak of inflated expectations back in 2024 and has been sliding into the trough of disillusionment ever since. The analysts who invented the curve say the hype has already broken, and the spending hasn’t noticed.
Look at the money. Earlier this year, CNBC reported that Alphabet, Microsoft, Meta and Amazon were expected to spend nearly $700 billion combined on their AI build-outs in 2026. Later trackers put it around $725 billion, up about 77% from 2025, with analysts projecting more than a trillion dollars in 2027. That’s four companies, in one year, spending more than most countries produce.
Now look at what it’s buying. In 2025, MIT researchers studied how companies were doing with generative AI, drawing on interviews with 150 business leaders, a survey of 350 employees and an analysis of 300 public deployments. They found that 95% of corporate AI pilots stalled out without delivering measurable financial returns. Only about 5% produced real revenue growth. Some critics have questioned the study’s methods, but nobody has produced numbers showing the opposite.
Read those numbers together and tell me you’re not a little stunned. The research firm that tracks hype says we’re in the trough, and the biggest companies on earth are nearly doubling their bets anyway, close to three-quarters of a trillion dollars in a single year on technology that most businesses can’t yet make pay.
I can’t wrap my head around it. If a client came to me with a business plan like that, I’d tell them to go sleep on it for a month.
Has a hype crash like this happened before?
Over and over, going back further than you’d think.
In 33 AD, Rome had a credit crash. The Senate suddenly started enforcing an old law that required lenders to keep part of their money in Italian land. Every lender called in loans at once, land prices collapsed and credit froze. The emperor Tiberius ended the panic by pouring 100 million sesterces of interest-free loans into the banks. A bailout, two thousand years ago.
In the 1630s, the Dutch went mad for tulip bulbs, until they suddenly didn’t.
In 1720, the South Sea Bubble wiped out fortunes across Britain, reportedly including Isaac Newton’s. The line attributed to him is that he could calculate the motions of the heavenly bodies, but not the madness of people. In the 1840s, Britain went through railway mania, with investors pouring money into lines that were never built or never paid.
And then there’s the one I lived through. In the late 1990s, anything with “.com” in its name could raise money. Then the Nasdaq lost about three-quarters of its value between 2000 and 2002. Pets.com became a punchline. Even Amazon lost around 90% of its stock price on the way down, and Amazon was one of the survivors.
Sound familiar? The technology is different every time. The people are exactly the same.
Why do smart people keep falling for technology bubbles?
Because being wrong with the crowd is safer than being right alone. That’s the ugly truth underneath every bubble.
There used to be a saying in corporate IT: nobody ever got fired for buying IBM. If you picked the safe, popular option and it failed, everyone failed together and nobody blamed you. AI is the new IBM. A CEO who pours billions into AI and watches it flop will be forgiven, because every other CEO did the same thing. A CEO who holds back and turns out to be wrong gets eaten alive by the board and the analysts.
So everybody spends. The numbers don’t make sense, and not spending feels like the bigger career risk anyway. Multiply that fear across every boardroom in the country and you get a stampede that nobody individually believes in and nobody can stop. I find that hard to watch. These are people running some of the most powerful companies on earth, and they’re making the biggest bets in business history mostly so they won’t look foolish in front of each other.
Where is all the AI money coming from?
The answer makes me angrier than anything else here, because it’s so simple.
Multiply that fear across every boardroom in the country and you get a stampede that nobody individually believes in and nobody can stop.
There’s only so much investment money in the world. When everyone wants in on AI, investors sell their “boring” stocks to buy the AI ones. That means every big company has to look like an AI company, or it becomes the stock that gets sold to pay for someone else’s. So Microsoft stuffs Copilot into everything and calls it a new operating system for work. Did you ask for Copilot? Did anybody you know? Nobody begged for it. Wall Street needed to see it.
Inside companies, the money comes out of everything else.
Fewer staff, smaller support teams, shelved projects. Critics have started pointing out that some of the tech layoffs blamed on AI look more like companies cutting people to free up money for data centers. You’ve felt this yourself, every time you called a support line and hit an AI gate that couldn’t help you. Somebody’s AI budget got paid for with the support team.
And a big share of the data center buildout runs on borrowed money. Borrowing for big projects is normal. Borrowing at this scale, on the bet that revenue nobody has seen yet will show up in time to pay it back, is not. When that bet goes wrong, the executives who made it will keep their bonuses. The people who pay are the support staff already cut, the smaller vendors waiting to get paid, and the ordinary retirement accounts that hold these stocks. I think that’s obscene.
Can the AI hype cause the crash it’s trying to avoid?
Yes, and it’s the most damning thing about this whole business.
The crash doesn’t have to be as bad as the ones before it. AI is useful. It could slide gently into the trough, get sorted out and settle into everyday work. But the way the industry is behaving is practically guaranteeing a hard landing.
Think about how it plays out. Companies pour borrowed billions into data centers based on projected demand. To make the demand look real, they force AI into products whether customers want it or not, and they hide the way out.
They fire the people who knew how the work got done and replace them with tools that 95% of the time can’t show a return.
You get frustrated, and so does every other customer, and you leave. Quality drops. The revenue that was supposed to pay for all of it doesn’t arrive on schedule, and the debt comes due anyway.
That’s the same sequence as every bubble before it, and the people causing it are the ones who swear they’re building the future. I’ve watched it happen once already. I wrote about how Klarna replaced its customer service people with AI and ended up bringing humans back. Now imagine that mistake made by a hundred companies at once, on borrowed money.
How can people this smart not see it? Some of them do. They’re just betting they’ll be out before the music stops. The rest of us will be the ones left holding the bill.
What survives the trough of disillusionment?
The dot-com crash came right after the peak, while AI spending keeps climbing after Gartner's peak has passed.
There’s some hope in all this. The technology usually survives the crash. The hype and the money don’t.
When the dot-com bubble burst, the internet didn’t go away. It went on to change almost everything, just slower and differently than the 1999 pitch decks promised. The companies with real products and real customers came out the other side. The ones built on hype were gone.
AI will probably follow the same path. A lot of these data centers will turn out to be overbuilt. A lot of AI startups will vanish. Some of the giants will take a beating they’ll spend years explaining. But the tools that really help people, like a digital assistant that helps a writer research and organize, will stick around, get cheaper and settle into ordinary work. That’s the plateau of productivity, and it’s a fine place to end up. It’s just not worth three-quarters of a trillion dollars a year to get there faster.
I’ve written more about this middle ground in why the doomers and the hypers are both wrong, because the people screaming that AI will destroy everything are as far off as the ones selling it as a miracle.
My take
Crashes don’t kill good technology. They kill the companies that bet borrowed money on a schedule the revenue never met. AI is headed for the same kind of crash, and the loudest cheerleaders are speeding it up. I’d ignore both the doomers and the miracle sellers. Keep using the tools that save you time. Don’t sign long contracts on promises, and don’t build your business on a vendor with no plan for the trough.
What should writers and business owners do during the AI hype cycle?
Stay calm, and don’t bet your business on somebody else’s hype. When a business owner brings up AI with me, my first question is how they’re using it. I’ll bet the answer is a pile of apps. Do you know what they do? Do you know how to use them, or did you just buy a bunch of shit?
Use AI where it helps you today, not where a keynote says it will help you someday. I use it for research, organizing and grunt work, and I keep the writing for myself. I’ve laid out how I split the work on a book if you want a concrete model.
Don’t fire the people who know how your work gets done. When the crash comes, the businesses that kept their experienced people will be the ones still standing.
And if you’re a writer, stop panicking. When everyone’s finished pretending a machine can write their memoir, the people who want a real book will still want one, and they’ll want a human who’s done it for years. More of my thinking on all of this is in the AI and Writing Hub, and if you want help using these tools without losing your voice, that’s what my AI services are for.
The hype cycle has never failed. Not in Rome, not in Amsterdam, not in London, not in Silicon Valley. The only question is how hard this one comes down, and right now the people with the most money are doing everything they can to make it come down hard. I can’t believe we’re doing this again. I really can’t.
After the crash comes the useful part. I wrote The Birth of the Augmented Human about that other path, where people use AI properly and become more capable across almost every domain.
