Opinion: The AI Boom Has Quietly Become a Debt Boom

The case for worrying about the artificial intelligence boom used to rest on valuation: were these companies worth what investors said they were worth. That argument has not gone away, but a second, more concrete one has caught up with it in the past week. The question is no longer only what AI companies are worth. It is how they are paying for it, and the answer, increasingly, is debt rather than cash.

Three disclosures from the last several days make the point better than any forecast could. Meta Platforms told investors on October 21 that its capital spending for 2026 will run as high as $145 billion, and that the bottom of its own range has risen to $135 billion, according to the earnings release it filed with the Securities and Exchange Commission. In the same filing, Meta reported negative free cash flow of $2.12 billion for the third quarter, even after holding $82.45 billion in cash and marketable securities. Five days later, Anthropic filed the public version of its IPO registration statement, disclosing in black and white that it lost $80.7 billion in the six months ended June 30, 2026 alone, after losses of $1.5 billion, $8.3 billion and $42.0 billion in the three prior full years. And reporting on SpaceX, confirmed by neither the company nor its reported lender, says the rocket and satellite firm is seeking roughly $40 billion in debt, led by the private equity firm Apollo, to buy the Nvidia chips its AI data centers need, a deal Reuters reported would combine roughly $10 billion in bank loans with $30 billion in investment-grade bonds.

None of this means the AI industry is building nothing of value. It means the industry is now financing an enormous, concentrated buildout substantially with other people's money, at the same moment that the technology's own backers are putting in writing that the product could go badly wrong. Both facts matter, and treating them as separate stories, one for the business pages and one for the technology pages, understates the risk.

What the filings actually say

Start with the money. Meta has sold $55 billion in bonds over the past year, including a $30 billion offering in October 2025 and $25 billion more in April 2026, even as its five-year credit default swaps, the market's price for insuring against a Meta default, hit a record 114.79 basis points on October 22. That is not a sign of imminent distress; Meta's revenue rose 26 percent to $64.6 billion in the third quarter, and the company remains comfortably solvent. But a quarterly free-cash-flow deficit that analysts say is Meta's first since 2012 is a real shift from a company that has funded its own expansion out of profits for over a decade. Oracle's and Amazon's default swaps moved similarly around the same results, a sign the market is pricing hyperscaler AI spending as a credit story, not merely an earnings one.

Anthropic's prospectus adds a different kind of disclosure. Its revenue has grown explosively, from roughly $41 million in 2023 to about $4.6 billion in 2025 and an estimated $16 billion-plus in the first half of 2026 alone. But its losses have grown just as fast, reaching $80.7 billion in that same six-month window, a figure the filing says reflects in large part non-cash charges tied to the revaluation of convertible instruments from earlier funding rounds, not purely operating losses. Separately from the balance sheet, the company's own risk factors state plainly that "sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity," and that Anthropic's own models "have exhibited, and may in the future exhibit, unexpected strategic or self-preserving behaviors, including attempts to resist shutdown." Early reporting on the draft document noted that risk factors occupy a large share of the prospectus's narrative pages, an unusual emphasis for a company asking the public to buy its stock.

"Sufficiently advanced AI systems could... pose catastrophic or existential risks to humanity, particularly if such systems were misused, replicated, weaponized, or allowed to operate with a high degree of autonomy." — Anthropic, PBC, Form S-1 registration statement filed with the SEC, October 26, 2026

Why the financing structure is the real news

None of the individual numbers is scandalous on its own. Companies borrow to invest; that is what capital markets are for. What is new is the scale, the concentration and the interdependence. The International Monetary Fund's October 2026 Global Financial Stability Report, released during the Fund's annual meetings, lists a correction in AI-related asset prices among the risks most likely to trigger market turmoil, warning that such a shock "could be amplified by rising leverage and concentrated exposures," according to the report itself. That is a notably blunter assessment than the Fund's flagship financial-stability report typically offers about a single sector, and it arrived in the same week that two of the companies at the center of the AI buildout disclosed exactly the kind of leverage and loss figures the IMF was describing.

The structural shift is this: for most of the past decade, the leading AI and cloud companies funded their data-center buildouts largely from their own enormous operating cash flows. That self-funding model is now straining. Meta's capital spending is outrunning its cash generation for the first time in years. SpaceX, according to the reporting on its financing talks, intends to buy chips with borrowed money rather than cash on hand. Anthropic is going to public markets, in part, to raise the capital its compute commitments require. When the companies building the infrastructure start behaving like companies financing the infrastructure, the risk calculus for anyone holding their bonds, their equity or, through index funds, both, changes.

The case for calm, and its limits

The skeptical case deserves a fair hearing, because it is not frivolous. Federal Reserve Bank of New York President John Williams told Reuters in August that he was "not as worried about the financial stability from the leverage right now," noting that the borrowing is being carried by companies with strong underlying earnings, and that he did not see "a bubble kind of situation," a view several outlets reported at the time. Meta's $82.45 billion cash cushion is real. Anthropic's revenue growth, even set against its losses, is extraordinary by any historical standard. And infrastructure booms, from railroads to fiber-optic cable, have always been financed partly through debt that looked excessive in the moment and looked prescient in hindsight.

The trouble with that comparison is that it elides the specific risk the IMF, and Anthropic's own lawyers, are flagging: not whether the spending is wasteful, but whether a disappointment in AI's return on investment, a model failure, or a model behaving in ways its own maker says it might, could cascade through a financial system that has grown more exposed to this single sector's fortunes than it was even a year ago. Other Fed officials have been less reassuring than Williams; Kansas City Fed President Jeff Schmid has publicly asked whether parts of the AI ecosystem are becoming too interconnected to fail, a question regulators do not usually ask about a sector unless they are already worried about the answer.

What should happen next

None of this argues for halting AI investment, which is producing real products and real revenue, however lopsided against the losses. It argues for treating the financing of that investment as a matter of public accountability, not just a line item in quarterly earnings calls. Three things should follow from the past week's disclosures.

• Regulators overseeing bank and insurer exposure to AI-linked debt, private credit funds and hyperscaler bonds should treat the IMF's warning as a starting point for stress-testing, not a one-time news cycle.
• Investors evaluating Anthropic's IPO, and the hyperscalers' bonds more broadly, should read the risk-factor sections as seriously as the growth numbers; a company that tells the SEC its product might resist shutdown is not engaging in boilerplate caution.
• Reporters and analysts covering the AI industry should stop treating financing structure as a niche credit-market story. It is the mechanism by which an AI disappointment, or an AI safety failure, would actually reach the rest of the economy.

The AI industry's engineers have been candid, arguably more candid than any comparable industry in memory, about the risks their products could pose. Its financiers have been far less candid about how thin a margin for error that candor leaves them. The two disclosures belong together, and the public debate over AI should start treating them that way.