November 24, 2025 | Happy Third Birthday, AI Mania

November marks the third anniversary of ChatGPT, OpenAI’s AI language model.

Since the fall of 2022, competitor models have been released, including Claude (Anthropic), Gemini (Google), Copilot (Microsoft), Llama-based systems (Meta ecosystem), Mistral (Europe) and others.

OpenAI and its partners have been pumping hundreds of billions of dollars into new data centers in the United States as part of a technology arms race they call Project Stargate. The New York Times notes that, “In today’s dollars, that is enough to fund the Manhattan Project 15 times over. It could pay for the entire Apollo moon project. Twice.”

Meanwhile, ex-AI and related spending, activity in the rest of the economy has flatlined (shown below since 2023, courtesy of Lance Roberts).

These trends are increasingly hostile to the labour market (US payroll data since June 2023 charted below).

And, to put it mildly, AI is not generating income to cover its costs. OpenAI’s losses are predicted to roughly triple to more than $40 billion by 2027, with no profit before 2030. See, Big Tech’s Soaring Profits Have an Ugly Underside: OpenAI’s Losses.

Unlike the internet infrastructure buildout of the late 1990s, which has served us for the last 25 years, today’s AI chips have a relatively short lifespan of 18 to 30 months before obsolescence. Hence, tech cos must constantly replace hardware to stay competitive, and they are now issuing record amounts of debt to fund it. At the same time, they are turning to increasingly creative accounting to prop up near-term earnings as free cash flow falls.

The five major spenders on AI — Amazon.com Inc., Alphabet Inc., Microsoft Corp., Meta Platforms Inc. and Oracle Corp. — have raised a record $108 billion in debt combined in 2025, more than three times the average over the previous nine years. Investment markets are finally starting to appreciate the increased capital risk in all of this. See Big Tech’s Debt Binge Raises Risk in Race to Build an AI World.

 

Just a few months ago, AI spending was primarily coming from a few companies with strong balance sheets and robust growth in free cash flow. That has changed, and the tech industry’s risk profile has along with it.

The new dynamic was on display Thursday, as tech stocks swung from way up after Nvidia’s strong earnings to way down as investors assessed how much capital will be required to finance an AI world compared with the profitability timelines of those investments.

“We’ve seen an expansion of the ecosystem to include companies with weaker balance sheets like Oracle and CoreWeave, more debt, and we’ve also seen more interlocking and circular revenue relationships,” Shalett said. “That interconnectivity between the players brings systemic risk.”

…Oracle’s offerings have come under particular scrutiny. The stock soared in September after the company sold $18 billion in US investment-grade bonds to ramp up its AI spending and banks launched a $38 billion debt offering to fund data centers tied to Oracle. But since hitting a record high on Sept. 10, the shares have plunged 40% as investors reassess what the company’s aggressive capex is doing to its balance sheet and how it is financing its huge capital expenditures. The stock is on track for its biggest one-month drop since August 2001.

The Fed’s most recent semi-annual Financial Stability Report noted that pension funds and life insurance companies are now heavily exposed to private credit and have also emerged as buyers of the massive volume of data-center debt being issued. See, Who will pay for the AI revolution? Retirees?

The global bet on artificial intelligence could require trillions of dollars of borrowing. Who might lend that much? How about America’s retirees?

Tech companies are facing an enormous investment need that will outstrip even their considerable existing resources. Morgan Stanley analysts in July estimated that of the roughly $3 trillion that is expected in global data-center capital expenditures through 2028, only about half of that could be funded by projected cash flows. That would leave a roughly $1.5 trillion financing gap.

To meet such a tremendous need, the companies will need to turn to the biggest funding markets. And when it comes to corporate borrowing, those include the mainstream, high-grade bond market. Issuance in the investment-grade corporate bond market represented around two-thirds of the more than $2 trillion sold this year through October across the U.S. corporate bond and asset-backed securities markets, according to figures compiled by Sifma.

Little mentioned amid all the excitement is that environmental groups, including the Center for Biological Diversity, warn that the booming AI industry’s high resource consumption threatens the world’s climate-harm mitigation goals.  Who’s going to pay for all of that?

While we wish AI well, mounting costs are massive and multifaceted, and, as always, individuals with finite time horizons need to be on guard against the risks posed by corporate aspirations.

Danielle Park November 24th, 2025

Posted In: Juggling Dynamite

Next: Trump’s 28-Point Peace Plan »