It's not the end of AI, but there is a reckoning coming.

NOTE: I have simplified A LOT of the details here. This is a complicated marketplace worthy of a full economic paper. But these are the high points.

TLDR: OpenAI and Oracle are not looking great right now. There's an 85% chance that the AI bubble pops in the next 8-12 months. There's a 46% chance the bubble pops violently with a crash of OpenAI AND Oracle. If the violent crash happens, Microsoft, Amazon, and Google all have a VERY BAD day. Survivable, but very bad. Details on the first indicators below. Frontier AI is in trouble. It's not obvious trouble at first blush. There's no question that ChatGPT and Claude are beloved by many. I mean, how can a billion users be wrong? And yet...

The AI slowdown isn't a prediction. It's already showing up in company reports. On July 14, IBM had the worst single day in its 110-year stock market history after admitting that a bunch of big deals it was counting on "failed to close due to AI pressures." A week later, on a call with investors, IBM revealed that a third of those deals had signed after all, just late. Another software company, Pegasystems, dropped 14% after saying customers were putting off buying decisions because AI has them confused about what to buy. Uber blew through its entire 2026 AI budget by April, and companies everywhere are now putting hard limits on AI spending. Even the people who lend these companies money are getting nervous: in February, investors were lining up to buy tech giants' bonds—offering five dollars for every one dollar of bonds available. By July, when Amazon tried to borrow $25 billion, barely more than a dollar and a half showed up for every dollar it wanted, and Amazon had to pay extra interest to close the deal.

Why are things slowing? Mostly because, other than plagiarizing art and books, most company's have no idea how best to use it.

95% of AI projects in the last year were failures of one kind or another.

I'll just let that sit for a second: a 95% failure rate.

Don't get the wrong impression. Demand for AI is real. People use it more every month. Google just signed $50 billion in new cloud contracts in three months. The problem isn't that customers are disappearing. It's that things are slowing as company's try to figure out if what it is actually useful for. And slow is deadly when you've borrowed big.

OpenAI has promised to spend about $1.4 trillion on computing power, while making about $13 billion a year. That only works if the money shows up exactly on schedule. Oracle spent $23.7 billion more than it took in last year and wants to borrow $40 billion more—while rating agencies have its debt one step above "junk" status.

Think of it like a landlord who bought ten buildings with borrowed money, counting on every tenant paying rent right on time on the first of the month. If the tenants pay a month late, they haven't stolen anything, but the landlord still misses his bank payment, and the bank doesn't accept "the money is coming eventually."

What is shape of things to come?

So how will you know the peak has hit and we're on the slide down? Ignore the big headlines and watch three small things that will show the slide has started (hint, some of these have already happened).

One: buried in the fine print of their financial reports, five tech giants have promised $1.65 trillion in future payments for data centers, chips, and leases that don't show up as normal debt. The quarter they start talking about "discipline in capital expenditures" while they keep bragging about their spending plans, the retreat has started. (Note: Google committed to over $200B to datacenter capacity last week and their stock was hammered for it.)

Two: watch for companies announcing "record quarters" while also mentioning that deals are taking longer to close that combination means trouble, and IBM just showed us what it looks like.

Three: watch whether lenders keep showing up when these companies want to borrow. The last time this happened was the telecom boom of the late '90s: internet usage kept growing the whole time, but the companies that borrowed billions expecting even faster growth still went bankrupt right on schedule. The slowdown won't announce itself. It'll just show up a quarter late — which, if you owe money, is the same as not showing up at all. (We've already seen this in small bits and bites lately. It will likely accelerate.)

What does it mean for us?

The era of Freemium AI services will end. AI Art, AI writing, all that will become more expensive. Likely too expensive to run at the rate we're seeing. Eventually, the effort of making AI Slop books to sell on Amazon will stop being profitable. We'll see that industry die off, or at least scale back considerably. AI art and AI narration will start to feel the pinch and will pass that on to their users.

The AI that has destroyed our discoverability on social media will likely also thin out, giving us back some measure of discoverability on Social Media, but honestly, not much. Now the actual bad news.

Private Equity lending in this space is now infecting everything: life insurance policies, pension funds, index funds that are tied to your 401K. Private equity lending also picked up the slack from the banks after the 2008 crash. If any of this is starting to sound familiar, then you were alive and able to understand the news in 2008. Will it be that bad? Probably not. Will it be painful for every single person out there? Absolutely. This will be a real recession, not the hidden one we're going through today.

Watch this space for updates.