There is an old saying in finance that markets can stay irrational longer than you can stay solvent. This week, artificial intelligence gave us a live demonstration — not in one market, but in two, telling two completely different stories at the same time. The lenders are getting nervous. The buyers are getting euphoric. Both can’t be right forever, and the tension between them is the most important thing happening in finance right now.
Let me start with the nervous side, because it’s the newer and more surprising of the two. According to FT reporting from September 11, carried by Reuters, JPMorgan Chase — the biggest bank in America, not exactly a shrinking violet — halted lending to the hedge fund Situational Awareness, run by Leopold Aschenbrenner, after the fund suffered large losses tied to its AI bets. The numbers are stark: the fund lost 67% in July, a single month, and sold most of its public stock book to Citadel.
Sit with that for a moment. A 67% loss in a month. That’s not a bad quarter; that’s a life event. For a saver, imagine opening your retirement statement and finding two-thirds of it gone since the last one — the kind of number that ends marriages and starts lawsuits. And this wasn’t some fringe operation. This was a fund whose entire thesis was that AI was the future, run by someone the AI world took seriously, financed by the most powerful bank on Wall Street. When JPMorgan cuts off the credit line, it’s not making a philosophical statement. It’s doing arithmetic.
This is, as the FT reporting framed it, the first major institutional credit retrenchment linked to AI investment underperformance. Firsts matter. Credit is the cautious older sibling of finance — it gets paid back first, it asks harder questions, and it usually smells trouble before equity does. When the cautious sibling starts leaving the room, it’s worth asking what it knows.
The FT Group’s Breakingviews column, in a September 10 piece, gave that question a historical frame that I can’t stop thinking about. It drew a parallel between today’s independent AI data-centre operators — the “neoclouds” financing tens of billions of dollars in AI infrastructure — and the North American telecom buildout that crashed after 2000. Around $500 billion was poured into fibre-optic networks in the late 1990s; roughly 40 alternative carriers were listed; the capacity was real, the demand eventually arrived, but the companies that built it mostly died waiting. The piece noted that only about 10% of post-2012 de-SPACs trade above $10, per SPAC Research — a quiet statistic about how often the future arrives without rewarding the people who paid for it.
That parallel lands because the structure rhymes. Then: build the pipes, assume the traffic. Now: build the data centres, assume the workloads. Then as now, the technology itself was not the illusion — the internet really did change everything, and AI really might too. The illusion was the financing: the belief that whoever builds first gets paid first. Telecom taught us that you can be right about the future and bankrupt in the present. It’s a lesson written in other people’s money, and it keeps needing to be relearned.
But now turn the page, because the other bank account — the equity side — is having the time of its life. In Fortune’s CFO Daily editions this week, Sheryl Estrada reported that Nvidia agreed to acquire Hugging Face for $12.93 billion. Hugging Face’s platform generates roughly $150 million in annualised revenue. Do the division: Nvidia is paying about 86 times revenue. Eighty-six times. For context, a healthy software company might trade at ten or fifteen times revenue; eighty-six is not a valuation, it’s a declaration of faith. As Fortune framed it, this is a bet on Hugging Face’s strategic position at the centre of open-source AI, not on its current business. The current business is almost irrelevant to the price. What’s being bought is a position on the board.
And Nvidia isn’t alone in writing big cheques. The Hustle’s September 11 edition reported that Harvey, the legal-tech AI company, raised $550 million at a $15.5 billion valuation — following a $200 million round back in March. Half a billion dollars for software that helps lawyers do legal work. The lawyers, presumably, are still employed for now.
So here we are: JPMorgan won’t lend to an AI fund that lost two-thirds of its value in a month, while Nvidia will pay 86 times revenue for an AI platform. The credit market says: show me the cash flow. The equity market says: cash flow is a detail; destiny is the asset. My honest take — clearly labelled as my own — is that both are behaving rationally within their own rules, and that’s exactly what makes it dangerous. Credit has to be repaid; it cannot afford dreams. Equity can wait; it can afford to be wrong for years, until suddenly it can’t. The 2000 parallel isn’t that AI is the dot-com bubble reheated. It’s that the financing always cracks before the technology does, and the crack starts in credit.
There’s a third voice in this conversation worth hearing, because it comes from the person charged with keeping the whole economy stable. Fortune also reported that Fed Chair Kevin Warsh, in his Jackson Hole speech, pointed to AI token prices — the units measuring the data AI models process — as a potential clue to whether AI is delivering the productivity gains businesses are betting on. I love this detail. The most powerful central banker in the world, squinting at the unit economics of machine cognition the way his predecessors squinted at the price of steel. It’s a reminder that even at the summit of global finance, people are still searching for a dashboard light that tells them whether the bet is working.
And the early readings are mixed. A McKinsey survey of 1,719 professionals globally, reported in CFO Daily, found that nearly 9 in 10 respondents say their organisations regularly use AI for at least one business function, and 44% say they’re scaling it enterprise-wide — but the productivity gains aren’t translating into operating profit for most companies. Nine in ten using it; most not profiting from it. That’s the whole debate in two numbers. Adoption is not the same as return. Buying the treadmill is not the same as getting fit.
What does this mean for the rest of us — the savers, the small business owners, the people whose pensions ride inside index funds stuffed with AI-adjacent stocks? It means living with the split screen. Your 401(k) loves the Nvidia-Hugging Face deal; your bank’s risk department is quietly reading the JPMorgan story and nodding. Both are in your life already. The practical wisdom, I think, is old-fashioned: don’t confuse a good technology with a good investment at any price, and don’t confuse one fund’s blowup with proof the technology is fake. Aschenbrenner’s 67% July doesn’t prove AI is a mirage, and 86 times revenue doesn’t prove Hugging Face is worth every penny. They’re data points, not verdicts.
It’s not enough to just pick a side in the AI debate — we must listen to what the lenders are whispering and what the buyers are shouting, learn the difference between a technology’s promise and its financing, and contribute our own clear-eyed patience to a story that is still being written. The telecom bust gave us the modern internet; the survivors were built on the graves of the overleveraged. AI may follow the same script: the future arrives, but not everyone holding a ticket gets a seat. Our job is to make sure we’re still standing — with our savings, our businesses, and our judgment intact — when the easy-money era ends.
