AI Can Be Real—and the Financing Can Still Be a Bubble

The useful technology is not the same thing as a sensible price tag

As of October 31, 2026

AI is producing real revenue, real productivity experiments and real technical progress. It is also attracting extraordinary capital commitments, complex financing and expectations that may outrun what customers can pay for. Those statements are not contradictory. They are the starting point for a more useful question than “Is AI real?”: which parts of the AI economy are earning their cost of capital, and which parts are being priced on a story about the future?

That distinction matters because the current AI-bubble debate mixes together the usefulness of software, the profitability of individual companies and the value of financial claims built around future demand. A correction in one does not automatically mean the others disappear.

The headline corrections matter

S&P did not downgrade Meta to junk. On October 30, S&P affirmed Meta’s AA- rating but revised its outlook from stable to negative, citing uncertainty around capital spending and future free cash flow. That is a credit warning, not a collapse. It says the company’s spending plans deserve closer scrutiny; it does not say Meta suddenly became a distressed borrower.

Anthropic’s credit story also needs precision. The Financial Times reported that Kroll Bond Rating Agency assigned Anthropic a BB+ issuer rating in June—the highest speculative-grade tier, one notch below investment grade—based on documents it reviewed. The report also says KBRA had limited financial information and used a risk score rather than a traditional full model. That is an issued KBRA rating, not an S&P, Moody’s or Fitch decision, and it should not be turned into a claim that every major agency has already rated Anthropic junk.

Anthropic’s October 26 S-1 is serious evidence of scale and risk, but the viral version is misleading. The filing reports an approximately $80.7 billion first-half net loss. Reuters reports that most of that figure came from a non-operating item connected to convertible bonds; it also reports a first-half operating loss of $693 million and adjusted operating income of $1.62 billion. Accounting loss, operating performance, cash spending and future compute commitments are different lines on the map. They must not be collapsed into “Anthropic burned $80.7 billion.”

The real warning is leverage, not a magic crash date

Credit markets are already providing a measurable stress signal. Reuters reported on October 15 that five-year credit-default-swap spreads for CoreWeave moved above 1,100 basis points and that its 2031 bond yield reached 16% before retreating. CoreWeave is highly exposed to the question of whether future AI demand will arrive quickly enough to service infrastructure obligations. That is a legitimate risk signal. It is not proof that every AI company is insolvent or that a systemic crash is inevitable.

The same distinction appears in the wider financing picture. Goldman Sachs describes nearly $500 billion of AI-related debt issuance tracked in 2026 and estimates that about one-third of capex could be debt-financed. Those figures describe financing activity, not guaranteed losses. Debt can fund productive capacity when revenue grows; it becomes dangerous when utilization, pricing or customer concentration fails to meet the assumptions underneath it.

Rates matter too. The Federal Reserve raised the federal-funds target range by a quarter point on September 16, to 3.75%–4%. Higher-for-longer financing makes long-dated growth promises less valuable and makes thinly covered interest bills more painful. This is why the “bubble” question is ultimately a cash-flow question: who pays, how soon, and with what margin?

Strong suppliers do not prove every buyer is strong

A supplier’s results are not proof that every downstream investment will work. But the strongest results also show why blanket collapse narratives fail. Micron’s September 30 release reported fiscal Q4 revenue of $54.23 billion and an 86.8% GAAP gross margin. Gross margin is not net profit, and a strong quarter does not validate every data-center project. It does show that at least some parts of the memory and infrastructure chain are experiencing genuine demand and pricing power.

Nvidia’s August 26 fiscal Q2 release tells a similar story: $96.2 billion of quarterly revenue, including $89.0 billion from data center, up 117% year over year. These are not imaginary sales. They are also not a guarantee that all future capacity will earn an attractive return. The market can have real customers and still overbuild.

China is a warning, not a template

China’s STAR 50 AI-and-semiconductor index fell about 40% from its July high in less than four months, according to Reuters on October 26, wiping out roughly $480 billion in market value. Reuters also reported a roughly 20% fall in the CSI 300 from its June high and noted that valuations remained elevated even after the decline.

That is a useful case study in concentration, leverage and theme-driven speculation. It is not a mechanical forecast for Wall Street. China’s policy environment, investor base, export restrictions and index composition differ from those in the United States. The lesson is not “the US is next.” The lesson is “a popular theme can remain expensive after a large fall.”

Local AI may change the economics—but not erase the cloud

Open-weight models that run on local machines can reduce latency, keep some data on-premises and make certain inference workloads less dependent on rented cloud capacity. Meta’s October 15 local-model announcement, reported by TechCrunch, is evidence that this direction is real.

But open-weight is not the same as open-source: weights do not necessarily include training data or the full recipe. Local inference does not eliminate training, frontier-scale experimentation, storage, networking or enterprise workloads. It may redistribute demand rather than abolish it. The future could be hybrid: local models for privacy and routine work, cloud systems for the largest or most current tasks.

What I would watch next

I would watch five boring indicators instead of trying to predict a dramatic pop: whether AI customers renew compute contracts at profitable prices; whether data-center utilization matches construction plans; whether credit spreads normalize or widen; whether capex begins producing durable free cash flow; and whether model usage creates measurable value for customers rather than only more tokens.

I would also separate companies by balance-sheet role. A profitable chip supplier, a heavily leveraged compute landlord, a frontier lab with enormous commitments and a software customer adopting an assistant are not the same financial exposure. “AI” is a technology category, not a single asset.

Conclusion

The most defensible conclusion is neither “AI is a scam” nor “AI can only go up.” AI can be useful, transformative and commercially real while parts of its financing structure are speculative. A bubble is not disproved by finding one profitable company, and it is not proved by finding one stressed borrower.

For ordinary readers, the practical rule is simple: distrust certainty, inspect the denominator and ask what cash flow supports the claim. The technology deserves serious attention. The financing deserves even more careful reading.

This is an educational discussion, not investment advice or a recommendation to buy or sell anything.

Sources

1. Meta Platforms Inc. Outlook Revised To Negative — S&P Global Ratings, October 30, 2026.
2. Anthropic, PBC S-1 — SEC filing, October 26, 2026.
3. Anthropic assigned highest junk credit rating in June — Financial Times, October 2026.
4. Anthropic’s soaring sales, steep losses put AI frenzy to IPO test — Reuters, October 26, 2026.
5. AI credit worries send default insurance to record high — Reuters, October 15, 2026.
6. In China, the AI bubble has already burst — Reuters, October 26, 2026.
7. Federal Reserve issues FOMC statement — Federal Reserve, September 16, 2026.
8. Micron Reports Record Fiscal Fourth-Quarter and Full-Year 2026 Results — Micron, September 30, 2026.
9. NVIDIA Announces Financial Results for Second Quarter Fiscal 2027 — NVIDIA, August 26, 2026.
10. Meta launches an open-weight model that runs on Macs — TechCrunch, October 15, 2026.
11. How AI Debt Is Reshaping Credit Markets — Goldman Sachs, accessed October 31, 2026.