AI Is Becoming a Fixed-Income Story

1. The capital model is changing

Large technology platforms remain exceptionally cash generative, so describing them as abandoning an “asset-light” model entirely would be misleading. They can still fund a large share of investment internally.

What has changed is the size of the infrastructure requirement.

Morgan Stanley estimates nearly $3 trillion of AI-related infrastructure investment through 2028, with more than 80% of that spending still ahead. Its credit strategists now expect AI debt financing to broaden beyond conventional investment-grade corporate bonds into project finance, chip financing and other credit structures.

Reuters also reported that Alphabet, Amazon and Microsoft had already issued around $220 billion of debt in 2026 by early October, more than double the prior year's pace.

2. A long Treasury yield has two broad components

A useful decomposition of a long-term government bond yield is:

The first component reflects expectations about future monetary policy and inflation. The second is the additional compensation investors require for holding duration and bearing uncertainty over a long horizon.

Current yields show how demanding that environment has become. Federal Reserve H.15 data for 6 October placed the 10-year Treasury near 5.31% and the 30-year near 5.66%. Reuters reported the 10-year around 5.31% again on 7 October ahead of major Treasury auctions.

These yields cannot be attributed to AI financing alone. Inflation risk, energy prices, expectations of tighter Federal Reserve policy, fiscal deficits and heavy sovereign issuance all matter.

That qualification is important because the strongest version of the AI-crowding-out argument confuses a plausible marginal force with a complete theory of long rates.

3. What the Vayanos–Vila framework actually says

The preferred-habitat framework developed by Dimitri Vayanos and Jean-Luc Vila provides a useful way to think about bond-market supply and demand.

In the model, some investors prefer particular maturities, while risk-averse arbitrageurs connect different parts of the yield curve. When the market must absorb additional duration risk, expected returns and risk premia can adjust because arbitrageur balance sheets are not unlimited.

Later empirical and theoretical work by Greenwood, Hanson and Vayanos develops the same intuition: changes in the quantity and maturity structure of bonds can affect term premia when specialized investors must absorb the supply.

4. The institutional substitution effect is real — but not one-for-one

For a long-duration investor, a highly rated 20- or 30-year corporate bond can compete with a Treasury for portfolio space.

The corporate bond offers additional spread, while the Treasury offers superior liquidity, lower credit risk and different regulatory treatment. The two assets are therefore close substitutes in duration exposure, but not perfect substitutes economically.

The basic mechanism is straightforward:

That can reduce marginal demand for Treasuries and require a slightly higher sovereign yield to clear the market.

But the common “$10 billion into AI bonds means $10 billion less Treasury demand” formulation is too strong. Investors can receive new inflows, alter duration through derivatives, change sector weights, resize credit exposure or increase total fixed-income allocation. Balance sheets are constrained, but they are not mechanically fixed.

5. The collision matters because Treasury supply is already heavy

The corporate-supply story becomes more relevant because it is arriving alongside very large sovereign financing needs.

On 6 October, Reuters highlighted renewed scrutiny of Treasury auction demand after weak bidding in parts of the curve. This week's Treasury calendar includes large 3-year, 10-year and 30-year auctions, while long yields remain near multi-decade highs.

In that environment, new AI-related corporate issuance does not need to dominate Treasury demand to matter. It only needs to affect the marginal price-setting investor.

This is the key distinction:

Reuters has included heavy technology-sector issuance among the factors contributing to the recent global bond selloff. Yet the Financial Times has also noted that evidence for technology borrowing as a primary driver of sovereign yields remains limited.

Both observations can be true. AI debt can be a genuine new source of duration supply without being the dominant explanation for the level of Treasury yields.

6. Term premium is not the same thing as the neutral rate

The weakest claim in the original argument is that AI borrowing permanently raises the neutral interest rate, (R^*), simply because more capital is being demanded.

That conclusion does not follow mechanically.

(R^*) is a macroeconomic equilibrium concept: the real short-term interest rate consistent with the economy operating near potential with stable inflation over time. It depends on structural forces such as productivity, demographics, saving behavior, fiscal policy and desired investment.

A sustained AI investment boom could raise equilibrium real rates if it increases productive investment demand faster than desired saving. It could also raise productivity, incomes and eventually saving. The sign and persistence are therefore empirical questions.

7. Where the fixed-income spillovers could appear

If AI infrastructure borrowing contributes to persistently higher long-duration yields, the effects would extend beyond the technology sector.

The risk is therefore less “AI debt crowds out everything else” than AI becomes a common financing factor across multiple asset classes.

That matters for diversification. A portfolio containing hyperscaler equity, technology credit, utilities, data-center infrastructure and long-duration growth assets can appear diversified by security type while remaining economically dependent on the same AI capital-expenditure cycle.

What would prove or disprove the thesis?

The crowding-out thesis should be treated as testable, not rhetorical.

Evidence in its favor would include:

1. persistent widening of concessions on large AI-related bond deals despite strong issuer credit quality;
2. measurable deterioration in Treasury auction demand or higher auction tails during periods of heavy competing corporate supply;
3. higher long-end Treasury term-premium estimates even when expected short-rate paths are relatively stable;
4. portfolio data showing duration-sensitive institutional investors increasing high-grade corporate exposure while reducing sovereign duration;
5. spillover from hyperscaler issuance into wider investment-grade spread repricing.

Evidence against it would be equally informative: strong Treasury auctions, stable term-premium estimates, rapid absorption of AI paper without broader spread pressure, or new fixed-income inflows large enough to finance both markets simultaneously.