Fed Minutes: Why AI Infrastructure Borrowing Is Stirring the Long-End Bond Market

In February this year, Oracle issued $25 billion of bonds in a single offering. Of that total, $2.75 billion matures in 2066 and carries a 6.85% coupon. Alphabet issued $20 billion of bonds in the same month, also with maturities extending to 2066.

Both offerings were classified as funding for general corporate purposes, so the proceeds cannot all be treated as AI investment. Nevertheless, borrowing on this scale and over such long maturities sends a signal: technology giants are entering the long end of the corporate bond market in force. This provides a real-world backdrop to the discussion in the Fed's September minutes about corporate issuance affecting the Treasury market.

Is AI investment already borrowing against the future?

Oracle is a prominent exception: in fiscal 2026, its capital expenditure was $55.7 billion, compared with operating cash flow of just $32 billion. Its current operating cash generation is already insufficient to cover its expansion.

The other companies can still fund their investment from their own operating cash flow. According to LSEG's estimates for the next twelve months, capital expenditure amounts to 58.5% of operating cash flow at Microsoft and 63.2% at Alphabet, while Oracle's ratio reaches 184.9%.

Where does the money come from, and how should long-term debt be priced?

As early as June, the Fed's minutes noted that stronger corporate bond issuance partly reflected AI investment. In July, a few officials went further, pointing out that AI capital expenditure was increasingly being financed through borrowing, with funding supplied by both regional banks and nonbank investors.

Debt carries a cost, and bonds with very long maturities are harder to price. Oracle's 2066 bond, for example, was issued at a spread of 195 basis points over the benchmark 2055 Treasury. Investors must assess both the path of interest rates over several decades and the company's ability to repay its debt over the long term.

The September minutes added that leading cloud providers had issued large volumes of debt with relatively long maturities, and that their credit spreads remained relatively wide overall. After widening noticeably early in the year, those spreads had narrowed slightly between the two meetings.

Is AI pushing Treasury yields higher?

The Fed's September meeting minutes therefore recorded a market view: the competition for capital created by technology giants borrowing heavily for AI infrastructure could be one factor lifting the Treasury term premium. The minutes also noted other reasons for the rise in long-term rates, including an upward revision to the expected policy rate path and persistently strong economic data. We cannot therefore conclude that AI has already directly driven a substantial increase in Treasury yields.

What is clear is that these very long-term debts connect today's AI construction to repayment obligations stretching decades into the future. How much cash flow AI investment will ultimately generate, and when that cash flow will materialize, remain uncertain.

Our focus from here should extend beyond AI capital expenditure itself to companies' financing structures, interest burdens, and the cash flows and returns that AI eventually delivers. This may be one of the most valuable threads to follow throughout the current AI capital cycle.