U.S. Treasuries: $240B AI Debt Could Lift Borrowing Costs
America pays about $1 trillion a year to service its debt. Now $240 billion of AI borrowing could push Treasury yields—and your borrowing costs—higher.
America pays about $1 trillion a year to service its roughly $40 trillion of debt. Now it has a nasty new competitor for investors’ money: the companies building AI.
That is not a typo. Alphabet, Amazon, Meta, Microsoft and Oracle are borrowing so aggressively to build data centres, buy chips and secure power that they are helping push up the yield America must pay to borrow. If you own a business, a home, equities or a super fund, congratulations: you are in this trade whether you asked to be or not.
Big Tech has stopped funding AI out of petty cash
For years, the conventional fear was simple. Governments borrow too much, soak up available capital and crowd out private business. Companies then pay more for debt, invest less and the economy gets slower and poorer.
That old model has not disappeared. But the order of attack has changed.
The AI arms race is so capital-hungry that the world’s richest technology companies are no longer relying only on mountains of operating cash flow. They are tapping public bond markets, private credit, structured finance and project financing to build infrastructure at industrial scale.
The Bank of England says Barclays expects AI hyperscalers to finance $240 billion of their 2026 investment needs through investment-grade credit issuance. Put that beside the usual annual senior-debt issuance of America’s six biggest banks—about $150 billion to $170 billion—and the scale becomes clearer. This is not a few tech companies issuing the odd bond. It is a new buyer of capital turning up with a wheelbarrow.
By early May, the five big hyperscalers—Meta, Alphabet, Amazon, Microsoft and Oracle—accounted for more than 15% of year-to-date US investment-grade bond issuance, despite representing only 3% of the outstanding market at the end of 2025.
That is a hell of a shift in a very short time.
Fortune reported US investment-grade corporate bond issuance reached about $1.7 trillion through July, around 27% ahead of the prior year’s pace and on track to pass $2 trillion for the first time. The immediate point is not that Big Tech is broke. Most of these businesses remain enormously profitable and highly rated. The point is that even they cannot self-fund an unlimited infrastructure binge without making capital markets part of the business model.
Why Treasury yields are the price you should watch
Here is the bit most people miss: when investors can buy debt from a cash-rich technology giant or lend to the US government, those assets compete for the same pool of money.
Normally, a flood of new corporate bonds would force corporate borrowers to pay much more above Treasury yields to entice buyers. That extra gap is called the credit spread. But demand for AI-linked debt has been strong enough that spreads have stayed relatively compressed.
So where does the adjustment happen?
In Treasury yields.
Ed Yardeni’s argument, reported by Fortune, is that capital flowing into corporate bonds is capital not flowing into Treasuries. To persuade investors to buy the enormous quantity of government paper still coming, Treasury yields have to rise. That is the “reverse crowding out” story: instead of the government crushing private borrowers, private AI borrowing is adding pressure to the government’s cost of capital.
Don’t turn that into a cartoon. Treasury yields do not move because of one thing. America’s fiscal deficit, inflation expectations, oil prices, foreign demand, central-bank policy and the term premium all matter. Anyone claiming a single neat explanation for long-term rates is selling certainty because they have none.
But this AI-financing boom is a genuine new input into the machine. Ignore it because the borrowers have famous logos and you will miss the plot.
The expensive loop nobody wants to discuss
Higher Treasury yields are not just a finance-page curiosity. They are the benchmark price of money.
When long-term government yields rise, mortgage rates tend to stay higher. Commercial-property financing gets uglier. Small and mid-sized businesses pay more to refinance. Private-equity deals get harder to justify. Governments spend more servicing old debt rather than building useful things. Equity valuations also come under pressure because future profits are discounted at a higher rate.
And the US government has a particularly brutal feedback loop sitting in plain sight.
More Treasury issuance can mean higher yields. Higher yields mean higher interest costs. Higher interest costs widen future deficits. Wider deficits require more borrowing. Then the market asks for more compensation again.
That is the loop.
Now add a cohort of AI companies willing to be relatively indifferent to borrowing costs because they believe the payoff from winning the compute race will be enormous. Treasury Secretary Scott Bessent has noted that these companies appear close to yield-agnostic: they believe returns on the AI build-out will justify what they pay to finance it.
Maybe they are right. Some probably are.
But “we can afford it” is not the same as “the rest of the economy is unaffected.” If a handful of giant companies can outbid everyone for chips, electricity, construction capacity, engineers and debt capital, then everybody else operates on a more expensive pitch.
That is not innovation magically making life cheaper. At least not yet.
The contrarian angle: this is not automatically bearish for Big Tech
The lazy take is that more debt equals danger and therefore AI stocks must be doomed. That is rubbish.
Debt is not inherently bad. Debt used to finance a high-return asset can be brilliant. If an AI data centre produces contracted, durable cash flows and a company locks in long-term financing at sensible rates, borrowing can be rational. In fact, it may be the only rational move if management believes the alternative is losing strategic ground to a rival.
The better question is whether the spending creates returns before the financing bill grows teeth.
The Bank of England’s warning is useful here. It says the rapid growth of AI financing is building risks, particularly around medium-term debt servicing. It also flags the rise of more complicated off-balance-sheet structures: special-purpose vehicles, securitised data-centre assets and bespoke arrangements. That matters because complexity is often how risk gets moved around until nobody is quite sure who owns it.
I have seen this movie in different costumes. When capital is abundant and a story is hot, people convince themselves that the financing structure is a detail. It is never a detail. It is the thing that decides who gets wiped out when revenue disappoints, refinancing dries up or the asset takes longer to earn its keep.
The first-order winners may still be the hyperscalers. The second-order winners could be power suppliers, data-centre operators, network infrastructure, specialist construction firms and lenders with discipline. The losers will be businesses that need cheap capital but lack the scale, ratings or glamour to get it.
In other words, the AI boom is becoming less like a software story and more like a heavy-industry story. Heavy industry needs steel, energy, concrete, debt and patience. Plenty of investors have not adjusted their thinking.
What this means for you
If you are a founder or operator, stop building plans that assume money will become cheap because someone on television said rate cuts are coming.
Use these rules instead:
1. Stress-test your debt now. Model refinancing costs at least 1 to 2 percentage points above your current assumptions. If the business breaks, you do not have a business model. You have a rates bet.
2. Match funding to asset life. Do not fund a long-lived asset with short-term, fragile money. Data centres, plant, equipment and expansion projects need financing that can survive a bad quarter.
3. Treat AI spend like capex, not magic. Every AI project should have an owner, a measurable productivity target, a payback period and a kill date. “We need an AI strategy” is how adults burn shareholder money while pretending to be modern.
4. Protect cash conversion. Revenue is lovely. Cash arriving before bills fall due is lovelier. In a higher-yield world, strong working-capital discipline is not boring—it is a competitive advantage.
5. Do not confuse big-company access to capital with yours. Alphabet can issue bonds at a scale most businesses cannot dream of. Your financing conditions are your financing conditions. Price risk accordingly.
For investors, the practical lesson is equally blunt: do not own an index, a bond fund or a business without understanding its exposure to long-term yields. The AI boom may create extraordinary companies. It may also keep the benchmark cost of money higher than people expect.
That is the inconvenient part of progress. Somebody always pays for the shovels. This time, the bill may be showing up in the yield on the safest bond in the world.