Author: Wall Street News
On September 24, U.S. long-term bonds collapsed across the board. The 30-year Treasury bond yield rose to 5.48%, a new high in 20 years; the 10-year Treasury bond yield exceeded 5.2%, a new high since the financial crisis. Long-term yields have risen almost in a straight line in the past two months. Finance Minister Bessant stepped up repurchase intervention twice, each time only getting a day's respite, and the market returned to its original shape the next day.
The deficit is too big, oil prices are too high, and the Federal Reserve is raising interest rates again—these are the real driving forces behind it, but they are not new variables in this round. What really makes this round different from the past is that the demand side has an opponent who doesn't care about the price at all.
AI is crowding out the Treasury Department from the market
To rush to build data centers and computing power, technology giants are issuing ultra-long-term bonds at a record pace, with almost no regard for price.
As of August, the debt issuance volume of the five largest ultra-large-scale cloud vendors this year has approached US$230 billion, more than twice that of last year, with a large number of them concentrated in 20-, 30-, and even 40-year terms.
These deals would make headlines in any year.
Meta was only established 22 years ago, but last October it issued a bond that was locked for 40 years—by the time it expires, Zuckerberg will be 80 years old. This amount alone was US$30 billion, setting a new record for a single issuance of corporate bonds. In April this year, Meta added an additional $25 billion.
Alphabet even issued a 100-year pound bond. Amazon raised $54 billion in a single round in March. Oracle is currently only two notches above junk status and has negative free cash flow, but it is still issuing debt one after another.
Why don't they care about interest rates? In an interview in August, Bessent himself lamented that these companies borrow money without looking at interest rates at all, because they firmly believe that the returns from AI will be so high that the current interest rates are not worth mentioning.
In other words, rising interest rates, which under normal circumstances would automatically put the brakes on the urge to borrow, have caused a failure in demand for this AI.
There are only so many funds willing to lock in "long-term money": pension funds, insurance companies, long-term bond funds - the long-term end is supported by these buyers, and the money in their hands is limited. Money flows into the long-term debt of technology companies, but cannot flow into national debt.
Alphabet's 30-year corporate bonds have a yield of 6.4%. If government bonds want to get their money back, they can only raise their yields.
This is actually a "reverse crowding out": in the past, it was the government's bond issuance that squeezed out private investment. Now it is the AI giant's turn to squeeze the government out of the bond market.
Nomura estimates that technology giants alone have borrowed about US$200 billion in long-term funds, accounting for a quarter of the Ministry of Finance’s annual medium- and long-term treasury bond issuance. Investment-grade bond funds are also reducing their holdings of treasury bonds and increasing their holdings of corporate bonds this year. Foreign private funds have also purchased more U.S. corporate bonds than U.S. treasury bonds - something that was hard to imagine before.
The mere fact that a large, price-insensitive private borrower is on the long end rewrites the rules of the game.
Bessent’s repurchases simply can’t reach here—it’s only a few billion dollars each time, which is a drop in the bucket in the national debt market of more than 30 trillion US dollars. Moreover, it can only adjust the term structure of the national debt itself and cannot control the pace of debt issuance by technology giants.
There is also a self-reinforcing cycle hidden here: the issuance of bonds by technology giants pushes up long-term interest rates, and higher interest rates in turn raise the cost of funding for AI projects themselves - the additional premium paid by AI giants when they issue new bonds has soared from more than 2 basis points last year to 12 basis points.
AI will only make the U.S. government poorer and poorer
If it is just about grabbing money from the Ministry of Finance, there is still a solution to the problem, because if AI really succeeds in making the economy develop rapidly, tax revenue will catch up sooner or later, and the government will not have so much difficulty repaying debt, and the yield will naturally fall.
But the problem is that the AI boom probably won't expand the government's tax base. Productivity and GDP skyrocket, and the government's debt solvency may actually worsen.
This deterioration mainly manifests itself in both the long-term and near-term aspects.
In the long run, U.S. personal taxes and social security taxes together contribute more than three-quarters of federal revenue, which means that the financial lifeline of the U.S. government is tied to the payroll, and the economic effect of AI is precisely to compress the value and income of labor - companies use fewer people to do more things, profits rise, but total wages decline, and the government gets less and less money from personal taxes and social security taxes. If AI reduces the share of labor income in GDP by 10 percentage points, most developed countries will have major financial problems.
In the near future, the "Big and Beautiful Act" passed in 2025 has sharply reduced corporate taxes in the United States. This bill restores the policy that capital expenditures can be fully tax deducted in the same year. The original intention of the bill is to stimulate investment, but for those technology giants who have already spent money on data centers at any cost, this is not an incentive, but a windfall. Microsoft disclosed in July that the tax payable for the current period was only US$2.5 billion, compared with US$14.1 billion in the same period last year - a decrease of 80%, but profits are skyrocketing. That's the magic of depreciation. Data from the U.S. Congressional Budget Office in September confirm this: In the first 11 months of this fiscal year, corporate income tax revenue plummeted by a quarter year-on-year, a decrease of nearly $100 billion.
Revenues are shrinking, spending is swelling, and tax revenues are off a cliff. The deficit can only be covered by debt issuance. The federal deficit has piled up to US$2 trillion in the first 11 months of this fiscal year. The total national debt just crossed the US$40 trillion mark in August. Interest alone costs about US$1 trillion every year.
Under the current system, it is simply unrealistic to expect AI to eventually come to the rescue by expanding the existing tax base.
The more successful AI becomes, the less it will be able to escape heavy taxes.
The more successful AI is, the more completely it will replace labor, and the less taxes the U.S. government will receive from its payroll.
At the same time, government spending will not shrink as a result, but will expand—because more people need retraining and welfare to support their lives.
As the government's financial gap becomes wider and wider, money can only be found from the only place that is still expanding - profits created by AI.