Long-dated U.S. Treasuries fell across the curve on Sept. 24, with the 30-year yield climbing to 5.48%, a 20-year high, and the 10-year yield moving above 5.2%, the highest level since the global financial crisis. The article says Treasury Secretary Bessent stepped in twice with buyback operations, but each move brought only a brief pause before yields pushed higher again.
Its core argument is that large fiscal deficits, elevated oil prices and renewed Federal Reserve rate hikes are all part of the move, but the sharper change this time sits on the demand side: AI giants are entering the long-end funding market at scale and, in the author’s telling, they are doing so with little regard for price.
AI borrowing is competing directly with the Treasury for long-term capital
To build data centers and secure computing capacity, large technology companies have been issuing ultra-long bonds at a record pace. The article says that, as of August, the five biggest hyperscale cloud companies had sold nearly $230 billion in debt this year, more than twice last year’s total, with a heavy concentration in 20-, 30- and 40-year maturities.
It points to several transactions. Meta, a company only 22 years old, sold a 40-year bond in October last year. That deal alone raised $30 billion, which the article describes as a record for a single corporate bond issue. In April this year, Meta added another $25 billion. Alphabet issued a 100-year sterling bond. Amazon raised $54 billion in a single transaction in March. Oracle, whose rating is described as only two notches above junk and whose free cash flow is negative, has also kept returning to the bond market.
The piece cites comments from Bessent in an August interview, saying these companies are not really looking at interest rates when they borrow because they believe AI returns will be high enough to make today’s funding costs look small. In that framework, rising rates are no longer functioning as a normal brake on borrowing demand tied to AI expansion.
At the same time, the pool of investors willing to lock up long-duration capital remains limited. Pension funds, insurers and long-duration bond funds are the natural buyers at the long end, and their capital is finite. If that money goes into long-dated debt issued by technology companies, it does not go into Treasuries. The article uses Alphabet’s 30-year corporate bond yield of 6.4% as an example, arguing that Treasuries must offer higher yields to win that capital back.
It describes this as a form of “reverse crowding out.” Instead of government borrowing pushing out private investment, the article argues that AI giants are now pushing the government out of the bond market.
Nomura estimate puts tech borrowing at roughly $200 billion of long-term capital
Citing Nomura, the article says technology giants have absorbed about $200 billion in long-term capital, equal to one quarter of the Treasury’s annual issuance of medium- and long-term debt. It also says investment-grade bond funds have been cutting Treasury holdings this year while adding corporate bonds, and that foreign private capital has been buying more U.S. corporate debt than U.S. government debt.
In the article’s view, a borrower of this size, operating with limited sensitivity to price and staying active at the long end, changes the market’s basic rules. Against that, Bessent’s buyback operations are portrayed as too small to matter. Each one was only worth tens of billions of dollars, the article says, a limited amount in a Treasury market worth more than $30 trillion. It adds that buybacks can alter the maturity structure of Treasury issuance, but cannot slow the pace of bond sales from technology companies.
The piece also points to a reinforcing loop. Heavy issuance by AI companies pushes up long-term yields, and those higher yields then raise the companies’ own financing costs. It says the extra premium AI giants have had to pay on new bond sales has risen from a little more than 2 basis points last year to 12 basis points.
The article says AI growth may not expand the tax base, and could weaken it
If the issue were only competition for capital, the article says the pressure might still be manageable. A successful AI boom could, in theory, lift economic growth enough for tax revenue to catch up. But the article argues that this is not the more likely outcome. In its telling, faster productivity growth and stronger GDP do not necessarily improve the government’s ability to service debt and may even make that ability worse.
It breaks the problem into two parts: a long-term tax-base issue and a near-term tax-receipt issue.
Long term: falling labor share could erode income and payroll taxes
The article says personal income taxes and Social Security taxes together account for more than three quarters of U.S. federal revenue. In that sense, the federal government remains heavily tied to wages and payrolls. But one expected economic effect of AI is that companies can produce more with fewer workers, lifting profits while reducing the aggregate wage base. If that happens, the government collects less from labor-linked taxes.
The article adds that if AI reduces labor income as a share of GDP by 10 percentage points, public finances across most advanced economies would come under serious strain.
Near term: full expensing for capital investment could cut corporate tax receipts
The article says the 2025 "One Big Beautiful Bill Act" restored full same-year deductibility for capital spending. The policy was designed to encourage investment, but the article argues it acts more like a windfall for technology groups that were already prepared to spend heavily on data centers.
It cites Microsoft’s July disclosure, which showed tax payable of just $2.5 billion for the period, down from $14.1 billion a year earlier, even as profit was rising. The article attributes that move to depreciation-related tax effects.
It also cites September data from the Congressional Budget Office, saying corporate income tax receipts in the first 11 months of the fiscal year fell 25% year over year, a drop of nearly $100 billion.
With deficits and interest costs rising, the article argues AI profits could eventually face heavier taxation
The article says shrinking revenue and expanding expenditure leave the government relying even more on debt issuance. In the first 11 months of the fiscal year, the federal deficit had reached $2 trillion. Total U.S. debt moved above $40 trillion in August, while annual interest costs were running at roughly $1 trillion.
Within the current system, the article argues, expecting AI to rescue public finances by broadening the existing tax base is unrealistic. The more successful AI becomes, the more fully it may replace labor, reducing the taxes the government collects from payrolls and wages. At the same time, public spending would not necessarily fall. The article says retraining needs and welfare costs could rise instead.
Its conclusion is that, as the fiscal gap widens, the government may eventually have to look to the one area still expanding: profits generated by AI. In that view, even a successful AI sector may not escape materially higher taxes.
The article is presented by PANews as the view of a Wallstreetcn columnist and not the position of PANews. PANews also says the piece does not constitute investment advice.