Ruchir Sharma: US Debt Raises Pressure on AI
Ruchir Sharma, Chairman of Rockefeller International, warned that rising public borrowing costs in the US will squeeze other borrowers faster than the market expects and will have a greater impact on the "bubble-like AI market." This statement appeared in a report released by Fortune on Sunday (September 13) summarizing bearish signals from multiple analysts.
Sharma's judgment is based on the fact that US debt has exceeded 100% of GDP—he pointed out that rising public sector borrowing costs will start to squeeze other types of borrowers earlier, and AI-related companies, due to their significantly higher debt service costs, will be hit harder than other industries.
In the same report, James Reilly, Senior Market Economist at Capital Economics, provided specific forecasts: the S&P 500 index will rise by 7.7% to 8250 points by the end of 2026, then drop by 21% to 6500 points by the end of 2027.
Wall Street veteran strategist Ed Yardeni simultaneously lowered the probability of a bullish scenario—reducing the probability of his previously proposed "Roaring 2020s" bull market scenario from 80% to 70%, while raising the probability of a bearish scenario from 20% to 30%.
Specific bubble indicators listed in the report include: cyclically adjusted price-to-earnings ratios are nearing the peak levels of the internet bubble; earnings growth over the next 12 months is comparable to the peak of the internet bubble; the total free cash flow of large-scale AI cloud vendors is expected to turn negative by 2027; market capitalization concentration has reached extreme levels; the queue for IPOs and follow-up offerings indicates that the end of the bubble is measured in months rather than years.
Mechanically, the direct trigger for this series of warnings is the yield on the 10-year US Treasury bond—which rose to 4.97% last Friday (September 12), approaching the 5% threshold seen as the watershed of a "new era of tightening monetary policy"; once public borrowing costs continue to rise, government bond issuance will compete with corporations (especially high-leverage AI infrastructure builders) for the same pool of credit funds, raising overall financing costs. The AI industry is currently in a phase that relies on massive debt financing to build data centers and computing power, making it far more sensitive to rising interest rates than other sectors. The pressured parties are the highly leveraged AI infrastructure and cloud vendors, while the beneficiaries are the traditional defensive sectors with more stable cash flows and lower leverage during a rising interest rate cycle.
Source: Public Information
ABAB AI Insight
Ruchir Sharma is known for his bearish stance on consensus assets and has previously warned about the high concentration in US stocks and the bubble in a few tech stocks driven by passive funds. His firm, Rockefeller International, has issued similar warnings in past market cycles; Ed Yardeni is known for creating the optimistic narrative of the "Roaring 2020s," and his decision to lower the probability of that scenario indicates a divergence even within the bullish camp.
The core capital path of this warning is the "debt-driven AI infrastructure race"—large-scale AI cloud vendors are generally financing the construction of data centers and computing clusters through debt rather than relying solely on their cash flows, which is the direct cause of the mentioned expectation that the total free cash flow of large-scale AI cloud vendors will turn negative by 2027. Meanwhile, the US federal government's debt has exceeded 100% of GDP, leading to competition for funds in the same credit market between the government and corporations. If Treasury yields continue to rise, the interest costs of new debt for AI companies will also increase, creating a chain reaction of "government borrowing pushing up rates—AI companies' financing costs rising—cash flows further deteriorating."
This logic is highly similar to the signals before the bursting of the internet bubble in 2000—the report explicitly mentions that the current cyclically adjusted price-to-earnings ratios and forward earnings growth are already comparable to the peak of the internet bubble. The difference is that telecom operators back then caused overcapacity by issuing debt to lay fiber optic networks, while this time cloud vendors are doing so by issuing debt to build data centers and GPU clusters. Currently, AI infrastructure construction is in a sensitive phase where "capital expenditures have not yet peaked, but cash flows have already been questioned by the market."
Essentially, this is a capital concentration risk exposure triggered by a shift in pricing power—over the past few years, the AI narrative has allowed a few large-scale cloud vendors to achieve valuations far exceeding their cash flows in the stock market. This premium essentially reflects the market's infinite extrapolation of the low-interest-rate environment of sovereign credit to corporate credit; once the cost of sovereign borrowing begins to rise (with the 10-year Treasury nearing 5%), the foundation for this shift in pricing power is removed, manifesting as: switching from "who can tell the best AI story can get the cheapest money" to "who has the cleanest balance sheet and the most genuine cash flow can survive the rate hike cycle."
ABAB News · Law of Cognition
- The government borrows first, and only then do companies realize how expensive borrowing is.
- Valuation is supported by stories, but cash flow determines whether one can survive rate hikes.
- The bursting of a bubble is never calculated by years, but by months.