Uncle Sam Has Competition

One mildly inconvenient feature of borrowing money is that the borrower doesn't really get to set how much they pay for the money, and the people who have it get a say in the price. This applies even when the borrower is the US government.

As Treasury yields have risen in 2026, most of the discussion has been about inflation, the Federal Reserve, and how much money the government needs to borrow over the coming months and years. And yes, each of these topics deserves that attention. But while we're watching Washington, some of the biggest technology companies are turning up at the bond market with very large borrowing plans of their own.

They need money to train models and build data centers. The cloud, as it turns out, has quite a financing bill. And some of the investors lending to those companies would otherwise be buying Treasuries, which gives the government more competition for their money.

That competition helps explain higher Treasury yields. It's not the entire increase, but it deserves a place in the conversation.

A much bigger borrowing habit

Between January 2 and September 3, 2026, the 10-year Treasury yield rose from 4.19% to 4.77%. Over the same period, the 30-year yield rose from 4.86% to 5.25%.

This wasn't a steady climb: both were close to their levels at the end of July 2026, when they had reached their highest levels of the year so far.¹

Meanwhile, US corporate bond issuance reached $1.90 trillion during the first eight months of 2026, almost 30% more than in the same period of 2025. That includes both investment-grade and high-yield debt.²

The change is particularly noticeable among the large technology companies. Alphabet, Amazon, Meta, Microsoft and Oracle issued approximately $132 billion in US bonds during the first seven months of 2026, compared with $93 billion in all of 2025 and an annual average of about $35 billion between 2020 and 2024.³ They had beaten the previous full year's total by the end of July.

Competing for debt issuance

Let's say you're managing a pension fund. You have money to invest and pensions to pay twenty or thirty years from now. You can buy government bonds, or you can lend to a company you're comfortable with and earn some additional interest. Treasuries are considered "risk-free", at least in terms of the probability of you getting your money back - but that's not true for money you lend to companies. The company might not repay you, and you need to be compensated for taking that risk.

If you decide the extra return with corporate bonds makes them worth it, you may put less money into Treasuries. The government still needs to borrow, so attracting another buyer may require a higher yield.

This upward pressure on yields doesn't even require selling Treasuries. A pension fund grows, or a bond matures, and the investor needs to put the money somewhere. No dramatic loss of faith in Washington is required in order to shift bond purchases elsewhere; sometimes another borrower is simply making a more attractive offer.

Being confident you'll get your money back isn't the only consideration, though. There is also interest-rate risk. Even when a borrower makes every payment, a fixed-rate bond can lose value if market interest rates rise. New bonds offer more income, making the existing bond less attractive. All else being equal, longer bonds are generally more sensitive to this effect. That sensitivity is called duration.

An investor buying a long-term corporate bond is taking on some of the same interest-rate risk they would get from a long-term Treasury. There is only so much of that risk they may want in their portfolio, regardless of whose name is on the bond.

This is why the maturity of all that new debt matters. For example, $1 billion of thirty-year bonds generally exposes investors to more interest-rate risk than $1 billion of five-year bonds. On August 18, 2026, BMO estimated that corporate issuance so far that year was adding interest-rate exposure equal to about 42% of the exposure supplied by Treasury issuance, compared with roughly 30% in previous years.⁴ BMO's comparison accounts for differences in interest-rate sensitivity, measuring how much risk the two markets are asking investors to absorb.

Some of this exposure also comes through less obvious routes. Dallas Fed researchers describe how an AI developer can take out a floating-rate private loan and use a derivative called a swap to effectively fix its payments. That arrangement adds exposure to fixed interest rates elsewhere in the market, even though the developer hasn't sold a conventional bond.⁵

What the data says investors are doing

BNY's August 28, 2026 analysis found that banks in its data had reduced their holdings of long-term Treasuries by 12% in 2026 up to that point while increasing investment-grade corporate holdings by 9%.⁶

That looks like the shift we would expect. The same report supplies a counterpoint, though: other investment managers increased their Treasury holdings by more than 50% over the same period. Investors weren't all marching in the same direction. These are also observations from BNY's data, which may not accurately represent the entire market.

The attraction of corporate bonds is easy enough to understand. On September 3, 2026, the broad ICE BofA US Corporate Index yielded 5.51%. Its option-adjusted spread over the Treasury curve was 0.81 percentage point.⁷

The companies don't get unlimited, low-cost credit either. MSCI's September 3, 2026 analysis found that many major AI borrowers were losing some of their borrowing advantage. Their previously unusually narrow spreads had widened toward those of more typical investment-grade companies, making their bonds less similar to Treasuries.⁸

Corporate borrowing costs and Treasury yields can both increase as the market absorbs more debt.

How much does this explain rising yields in the Treasury market?

We can explain the mechanism much more confidently than we can measure its contribution.

The Fed's July 2026 report documented a substantial increase in expected policy rates after the Middle East conflict increased inflation risks.⁹ Changes in inflation, growth expectations and monetary policy would affect Treasury yields regardless of what companies were borrowing.

Treasuries also have uses that corporate bonds don't fully replace, including as collateral and reserve assets. Some investors have mandates that limit how much corporate debt they can buy. The competition matters where investors have flexibility to choose.

The evidence supports corporate borrowing as a contributor, but doesn't tell us exactly how much of the recent increase in Treasury yields it caused. A plausible explanation isn't proof or measurement of the effect.

Uncle Sam still needs money

On August 3, 2026, Treasury estimated that it would need $739 billion in privately held net marketable borrowing during the July–September 2026 quarter.¹⁰ That need is still there even as companies have become more interested in borrowing from the same investors.

There isn't a fixed pot of money that empties out every time somebody buys a corporate bond. Investors can save more, bring money from abroad, or sell other assets. But they still have to decide whether the return is worth committing their money, and the pot isn't unlimited. Higher yields can persuade them where to allocate.

My concern is that persistent deficits leave the government exposed to this competition even when private investment is going well. The AI buildout may eventually increase productivity and tax revenue, but the financing bill seems to be arriving before the benefits. If the Treasury has to offer higher yields to attract buyers, taxpayers end up taking on higher interest costs as new debt is issued and existing debt is refinanced.

The AI investments could work out perfectly well and still make the government's financing problem harder in the meantime. There doesn't have to be a bubble for this debt to have a significant impact, companies can simply have projects they want to fund and investors willing to lend to them.

Uncle Sam can still borrow. But increased competition can make it more expensive to do so.

Sources

1. Daily Treasury Par Yield Curve Rates (US Treasury). The January 2, July 31 and September 3, 2026 yield comparisons.
2. US Corporate Bonds Statistics (SIFMA, September 3, 2026). Gross issuance of $1,899.8 billion during the first eight months of 2026, up 29.8% from the same period in 2025.
3. The AI buildout comes to the bond market (Vanguard, August 19, 2026). Gross US bond issuance by Alphabet, Amazon, Meta, Microsoft and Oracle, with 2026 data through July 31.
4. Macro Notes — The Problems in the Long End (Bipan Rai, BMO, August 18, 2026). The comparison of corporate and Treasury issuance after adjusting for interest-rate sensitivity.
5. How AI debt financing impacts duration supply and interest rates (De Vere, Ramaswamy and Searls, Dallas Fed, February 10, 2026). Duration supplied through direct bond issuance, swaps and changes in the mix of corporate borrowers.
6. Hyperscaler issuance tests long-end UST demand (David Tam, BNY, August 28, 2026). Holdings changes within BNY's data and the limits of the evidence for substitution.
7. ICE BofA US Corporate Index Effective Yield and Option-Adjusted Spread (FRED). September 3, 2026 observations. The spread is measured against the Treasury curve, not calculated by subtracting the 10-year yield from the corporate index yield. Valuation and Analysis of Bonds with Embedded Options (CFA Institute, 2026 curriculum) explains call options and the adjustment for their value.
8. The Limits of AI Debt as a Driver of Treasury Yields (Michael Hayes, MSCI, September 3, 2026). Widening hyperscaler spreads and the limits on their substitutability for Treasuries.
9. Monetary Policy Report (Federal Reserve, July 10, 2026). Changes in inflation risks, expected policy rates and financial conditions.
10. Treasury Announces Marketable Borrowing Estimates (US Treasury, August 3, 2026). The July–September 2026 estimate is net borrowing, unlike the gross corporate issuance figures above.