Can AI change the world? That’s a technical issue. Can investors make money from it? That’s a business issue. Can financing continue to pay interest before returns are realized? That’s an assets and liabilities issue.
The biggest mistake the market easily makes is replacing the optimism of the first issue with the calculations of the latter two issues.
Today’s US consumer confidence data, US bond yields, and a recent large AI-related financing announcement have brought this distinction to the forefront.
On September 29, the US Treasury revealed that the 10-year Treasury bond yield was 5.26%, and the 30-year yield was 5.59%, up by 47 and 32 basis points respectively compared to September 1.
This is not just a matter for bond traders to care about.
The Treasury bond yield is a vital benchmark for US dollar financing. Corporate bonds, mortgages, etc., also factor in their respective credit, term, and liquidity factors: an increase in the benchmark price impacts new borrowing and refinancing conditions, but not every loan type increases in sync and to the same extent.
In terms of valuation, the higher the discount rate for the same future cash flow, the lower the present value. For borrowers, the future may be uncertain, but the scheduled interest payments must be made on time.
Therefore, I am more concerned about not whether “AI has a future”, but rather: if there is a delay of two years in the commercial returns, who has the cash to sustain those two years?
This is the most practical screening for high financing costs in new industries.
In SoftBank’s debt issuance plan announced on September 24, totaling $10 billion in USD-denominated bonds divided into three tranches with annual coupon rates of 8.625%, 9.25%, and 9.75%, respectively, with issuance set on September 29. The proceeds will be used for payments towards investments in OpenAI and for general corporate purposes.
Based on the calculation of each principal multiplied by the coupon rate, the annualized coupon for just this $10 billion bond issuance is about $941 million: this does not include euro-denominated bonds, fees, and existing debts. This is not the actual interest payment amount for the year 2026.
It is not important to declare “the AI bubble has burst” just by observing the high coupon. SoftBank has its own credit structure, and this interest is SoftBank’s payment obligation, not directly equivalent to OpenAI’s debt, let alone representing the financing costs of all tech companies.
What is truly worth dissecting is this structure: the timing and amount of investment returns are uncertain, while the payment timing and calculation methods for debts have already been written into the contracts.
An increase in equity valuation does not automatically mean there is cash available for debt repayment at the same time. A group can rely on other business cash flows, asset sales, or refinancing to pay interest, but these sources must be verified item by item; they cannot be replaced by the valuation of the invested companies.
Even if the technical judgment is correct, different investment results can still be obtained due to differences in financing terms, purchase prices, and cash flow arrangements.
The US consumer confidence index announced on September 29 dropped from 88.6 in August to 81.9. The present situation index fell to 109.3, and the future expectations index dropped to 63.6, the latter having dropped for three consecutive months.
This survey does not signal a recession, but it does show that consumers are simultaneously becoming less confident in current conditions and future prospects. The survey responses also more frequently mention prices and oil and gas costs.
To translate financial market dynamics into a household language, consider a mortgage as a straightforward example.
Freddie Mac’s survey as of September 24 shows that the average interest rate for a 30-year fixed-rate mortgage in the US is 7.03%, compared to 6.30% for the same period a year ago.
Assuming a $500,000 loan, a term of 30 years, and equal principal and interest payments, a 6.30% rate corresponds to a monthly payment of about $3,095, while a rate of 7.03% corresponds to about $3,337. An increase of about $242 per month, translating to roughly $2,901 per year.
The house hasn’t changed, the loan amount hasn’t increased; only the financing price change has raised the cash flow threshold for new buyers.
This is an example comparison for new loans, not to say that existing fixed-rate loans will automatically increase monthly payments; calculations also do not include property taxes, insurance, and other fees.
There is no need to fabricate a causal relationship of “AI taking away people’s money”. The real difference lies in the fact that the capital market can price future growth in advance, while households must wait until income actually arrives to improve their lives.
Active technological financing and decreasing household confidence can happen simultaneously. They do not face the same balance sheet, nor do they adhere to the same schedule for liquidation.
On September 16, the Federal Reserve raised interest rates by 25 basis points, raising the target federal funds rate to 3.75%–4%. On September 10, the European Central Bank also raised rates by 25 basis points, explicitly mentioning the inflationary pressure from the ongoing Middle East conflict.
This signifies that understanding international situations cannot solely rely on military actions or diplomatic statements but must also consider how impacts pass through energy, inflation expectations, and interest rates into the accounts of businesses and households.
For an economy that relies on energy imports and needs USD financing, energy shocks and financing tightening may cause a dual squeeze: inputs required for production become more expensive, and the money borrowed to purchase these inputs also becomes pricier. The specific impacts are still contingent on energy structure, debt maturity, and exchange rate hedging.
There is also a time lag in technology investments. Take data center construction, for example; equipment, electricity, and site investments happen first, and productivity improvements and new income follow to be gradually validated.
Therefore, a transitional phase may occur wherein long-term technological advances help improve supply, while short-term construction demand increases the need for funds and resources. In places where supply is constrained, these two are not contradictory.
From this, it can be deduced that the next stage of industrial competition may increasingly depend on “who can wait”: companies with existing cash flows, long-term financing, and stable energy arrangements versus those needing repeated refinancing to sustain investments; their ability to withstand the same market fluctuations differ.
I will focus on three questions.
First, can AI-related investments gradually form operational cash flows rather than just increasing financing amounts and book valuations?
Second, can long-term financing costs stabilize to provide manageable financing conditions for new investments and maturing debts?
Third, can household actual income, employment expectations, and significant consumption capacity keep up with the expansion on the investment side?
If productivity and cash flows improve fast enough, financing pressure can be absorbed; if returns are delayed, high leverage and reliance on short-term refinancing may become weak spots. These are conditional pathways, not a declaration of an imminent crisis.
AI doesn’t have to fail; a certain AI financing structure may encounter issues first. House prices don’t have to rise; new buyers’ monthly payments may increase. Technological progress doesn’t have to halt; ordinary households may still bear the cost of transition. These three assessments are closer to reality that need to be examined rather than simply debating “prosperity or collapse”.
AI is buying the future, while the bond market is pricing the waiting period.
The real test is not just how much value the future can create but who has the cash, time, and risk tolerance to wait for the value to be realized.