Okay – I just went back through my notes app – and hooooboy. I started to research and write sooooo many things this week that I just abandoned because there’s too much to cover – even if you spend 12 to 14 hours a day, everyday trying to keep up with it all.
But I posted a video of AOC talking about the Ai bubble earlier this week – and I want to expand on what she said.
As you know, Trump keeps celebrating economic growth like he personally invented prosperity. The numbers look good on paper. The stock market’s up. GDP’s climbing. But here’s what those headlines aren’t telling you: this entire economy is being propped up by seven companies betting everything on technology that hasn’t figured out how to make money yet.
The Magnificent Seven – Amazon, Apple, Alphabet, Meta, Microsoft, Nvidia, and Tesla – account for about 35% of the S&P 500. Between 16% and 20% of real GDP growth in late 2024 came from AI infrastructure spending alone. Four of those seven – Meta, Amazon, Alphabet and Microsoft – spent about $230 billion on buildouts in 2024, heading toward more than $380 billion this year. That’s roughly 0.8% of the entire US economy going toward data centers, chips, and computing power.
So, let’s examine exactly what that investment is actually buying.
OpenAI, the poster child of the AI revolution, lost $5 billion in 2024 on $3.7 billion in revenue. They’re projecting a loss of about $9 billion this year. By 2028, they expect operating losses of $74 billion. Their business model is spending about $1.69 to make $1. Their valuation – now at $500 billion – assumes they’ll somehow reach $200 billion in annual revenue by 2030, up from about $13 billion this year.
The basic formula is simple: promise investors that AI will transform everything, raise billions based on that promise, spend it all on infrastructure, then raise more billions to cover the losses. Rinse and repeat. Only about 5% of ChatGPT’s 800 million users actually pay for the service. Back in January, Sam Altman said the $200-per-month Pro tier was losing money because people use it more than expected.
And here’s what makes this genuinely frightening: we’ve been here before.
One analyst, Julien Garran of MacroStrategy Partnership, figures the AI bubble is now 17 times larger than the dot-com bubble and four times bigger than the 2008 real estate bubble. AI-related investment has increased by less than 0.4% of GDP since 2022, but unlike the dot-com era when investment spread across hundreds of companies, this is concentrated in seven firms. When the dot-com bubble burst, it took 15 years for the NASDAQ to recover. The 2008 crisis got Congress to approve up to $700 billion in taxpayer bailouts and triggered the worst recession since the Great Depression.
Experts believe the real difference this time is exposure. Index funds and retirement accounts are loaded with these seven companies because they represent more than a third of the S&P 500’s value. About 80% of stock market gains this year came from AI companies. If they correct from these valuation levels, everyone holding an index fund or 401(k) goes down with them.
The whole structure looks disturbingly familiar. Oracle is taking on massive debt to build AI data centers – its debt is about four times its shareholders’ equity. Their Credit Default Swap spreads are surging, signaling rising default risk. Companies like Microsoft are building computing clusters requiring entire country-level amounts of electricity. By 2028, their computing requirements could reach multiple gigawatts of power.
I’m working on another essay that is specifically about AI power consumption – so put a pin in it. If you mention this in the comments – you are a bad reader. (This is known as the David Lee Roth brown M&M Easter egg.) Anyway, here’s how it works: the capital flows in circles. Nvidia invests in OpenAI. OpenAI buys Nvidia chips. Oracle borrows billions to build data centers. Everyone’s revenue numbers look impressive because they’re essentially paying each other with borrowed money. The whole game isn’t built on customer demand – it’s built on promises about future returns that keep getting pushed further out.
What looked like economic growth is actually a massive infrastructure buildout for technology that stalled in about 95% of company pilot programs, according to an MIT report this summer. Companies are planning $1.4 trillion in data center spending by 2027 for demand that doesn’t actually exist yet. So far, the promised productivity gains haven’t materialized. AI isn’t making most of the companies trying it more profitable – it’s burning through capital at unprecedented rates.
Wall Street knows this. Hedge fund manager Michael Burry, who predicted the 2008 subprime meltdown, bought put options against Nvidia and Palantir. He has since deregistered his hedge fund – I’m not sure yet why. Deutsche Bank is using Synthetic Risk Transfer mechanisms that strongly resemble the Collateralized Debt Obligations that amplified the 2008 crisis. Investment banks are scrambling to hedge their growing exposure.
The optimistic take is that this won’t be as catastrophic as 2008 because the investments aren’t built on leverage from the banking system. Optimists point out that tech companies are mostly using their own cash, not borrowed money backed by complex financial instruments that could freeze the entire credit system. The IMF suggests it’ll look more like the dot-com bust – painful for tech workers and investors, but not a systemic banking crisis.
The pessimistic take is that we’re already at a point of no return. Absent AI spending, the economy is slowing to a crawl. AOC says at least 40% of economic growth this year is attributed to just seven companies. When – not if – these valuations correct, the impact will hit every retirement account and index fund in America.
Trump can claim credit for the booming economy all he wants. What he’s actually presiding over is the largest speculative bubble in modern history, concentrated in seven companies – and in AI startups like OpenAI that are spending far more than they earn while promising that profitability is just around the corner. We’re told the music is still playing, so everyone keeps dancing. But we’ve heard that song before, and we know exactly how it ends.
Correction, September 27th, 2026: This essay first said the Magnificent Seven aren’t profitable – all seven make money. It got OpenAI’s numbers wrong – the company was valued at $500 billion, not $300 billion, it projected a loss of about $9 billion for 2025, not $14 billion, and it was spending about $1.69 for every dollar it made, not $2.25. It also gave a 93% growth figure that didn’t match OpenAI’s own numbers. The MIT report found 95% of company AI pilots stalled – not that they were abandoned. The spending figures were for four of the seven companies – about $230 billion in 2024, with more than $380 billion expected in 2025. Oracle’s filings put its debt at about four times its equity – the 500% came from a bank chart its own author later called misleading. And the seven were about 35% of the S&P 500 at the time, not 37%. It also said the M&Ms in the David Lee Roth story were green – they were brown – and that Michael Burry dissolved his company – he deregistered his hedge fund.
Update, September 27th, 2026: Lines about the bubble’s size, the Pro plan, the 2008 bailout, AI profits and productivity, and how tech companies pay for all this now say who says so or how far the evidence goes. And the apostrophes an old glitch dropped are back. And the 40% growth figure now says it’s AOC’s.
Everything I have written about AI is in one place at rachelandthecity.com/the-ai-files.