The Crowding Out: What a $1.5 Trillion Defense Budget Means for AI Capital

The financial model underlying the AI infrastructure buildout assumed a particular capital environment: patient institutional money flowing from Gulf sovereign wealth funds, low real interest rates, and a federal government borrowing modestly relative to GDP. In the spring of 2026, all three assumptions are under simultaneous pressure. This piece is about the third one.

The White House has pursued a record defense spending package for fiscal year 2026 — but two numbers circulate and they measure different things. The $1.01 trillion figure (MeriTalk, citing the Pentagon FY2026 budget release) is the Trump administration's initial Pentagon base budget proposal. After Congressional action, the enacted Pentagon base appropriation was $838.5 billion, with an additional $152.3 billion in reconciliation funding — totaling approximately $991 billion for the Pentagon alone, or roughly $1.05 trillion for all national defense discretionary spending including atomic energy and other defense-related activities. The $1.5 trillion figure (NPR, April 17, 2026) is a projection of the full package: enacted base, reconciliation supplement, and a forthcoming Iran war supplemental that has not yet been formally requested or scored by the CBO. The Congressional Budget Office projects the fiscal year 2026 federal deficit at $1.9 trillion, or 5.8% of GDP — the third-largest in American history. Net interest payments are projected to become the third-largest item in the federal budget, surpassing defense spending itself.

When the federal government borrows at this scale, it competes for the same pool of long-duration capital that AI infrastructure projects are trying to access. The mechanism is straightforward: more borrowing means more government bonds to be absorbed by the market, which means higher yields to attract buyers, which means the cost of capital rises for everyone else. This is the classic crowding-out argument. Its applicability depends on how tight the capital market is — and in 2026, it is tighter than it was when the AI buildout was priced.

What the Buildout Actually Costs

The AI infrastructure investment cycle is not small. The five largest hyperscalers — Microsoft, Amazon, Google, Meta, and Oracle — announced combined capital expenditure commitments of roughly $600 billion for 2026, with Meta alone committing to a range of $60-65 billion. These are real debt and equity financings, not operating expenses. They compete directly with federal borrowing for the attention of the same fixed-income investors — pension funds, insurance companies, sovereign wealth funds — who need long-duration assets.

The interest rate sensitivity of these projects is material. Our previous analysis estimated that a 25-basis-point increase in the 10-year yield translates to roughly $1.25 billion in additional annual interest costs on $500 billion of long-duration capex financing. The term premium on 10-year Treasuries was approximately 0.68% as of early April — already elevated relative to the near-zero term premiums of 2020-2022. The direction of travel is up.

The defense budget is one of several factors pushing in that direction simultaneously.

The Supply-Side Shock First

To understand why the current fiscal environment is more disruptive than the headline numbers suggest, it is useful to put it against the background of what has already happened to the supply side of long-duration capital.

Gulf sovereign wealth funds — which allocate a meaningful share of assets to private infrastructure and technology equity — have been under pressure since the Iran conflict began. The Lopez/Global SWF analysis found that Q1 2026 pace held, but warned that a sustained conflict would trigger the COVID playbook: drawdown of foreign assets to fund domestic needs, reduced new commitments to long-duration international investments.

The war that the $1.5 trillion defense request is partly financing is the same war that has been squeezing the pool of patient capital that the AI buildout was counting on. This is not coincidence. It is a single causal chain. The US is borrowing to finance a conflict that has simultaneously disrupted the pool of patient capital that would normally absorb that borrowing.

The Three-Way Capital Competition

What makes the current moment analytically distinct is that the long-duration capital market faces three simultaneous large-scale demands.

Federal deficit financing: at $1.9 trillion, the fiscal year 2026 deficit requires Treasury to issue an enormous volume of long-duration debt. Each auction creates marginal upward pressure on yields.

AI infrastructure financing: $600 billion in announced hyperscaler capex for 2026 alone, the majority financed through corporate bonds, project finance, and equity raises — all competing with Treasuries for institutional capital.

Defense-adjacent private capital: venture capital in defense technology raised $11.2 billion in 2025, a tenfold increase in five years. Defense-tech has become its own asset class, competing for the same growth-capital pools that AI startups are drawing from. The government is not just a borrower competing with AI companies — it is also a procurement customer creating an alternative investment destination.

The academic debate on crowding out is inconclusive in the general case, and the research suggests defense R&D can generate crowding-in through spillover effects. But that argument applies to the long run, and to R&D spending, not to operating military expenditures. The marginal borrowing to finance current military operations is competing for capital in the current market.

The Discount Rate Problem

Here is the mechanism that connects this to the AI buildout most directly. The financial case for large-scale AI infrastructure investment depends on a discount rate — the rate at which future cash flows are valued today. A higher discount rate makes future revenue worth less in present-value terms, which makes infrastructure investments with long payback periods harder to justify.

The federal deficit drives up yields. Higher yields raise the discount rate. A higher discount rate makes 10-year revenue projections less valuable. Which makes the financial models for data center construction, power infrastructure buildout, and semiconductor facility investment look worse.

This is not a catastrophic shift. The projects are not being abandoned. But the marginal project — the one that made sense at a 4.5% yield but not at 5.2% — is the one that gets deferred or cancelled. The marginal project is the leading edge of capacity expansion. It determines whether AI infrastructure grows fast enough to support the demand projections the industry has made.

What the Crowding-Out Critics Get Right

The steelman for the opposition is not trivial. The Wharton Budget Model and others argue that crowding out is less pronounced when there is global capital mobility — that Treasury bond issuance can draw from a global pool of investors, not just domestic ones. In normal conditions, this is correct.

In 2026, global capital mobility is constrained by the same conflict that produced the defense budget. Gulf SWF capital has partially retrenched. European capital markets are navigating their own energy cost shock. Chinese institutional capital has reduced US exposure under ongoing sanctions pressure. The argument that global capital mobility dampens crowding out requires a functioning global capital market — and the events of the past year have made that assumption less reliable.

This is the point at which the seven-variable chain we have been tracking becomes analytically relevant not just as a list of pressures but as a compound effect: the variables that are causing the pressure are also reducing the mechanisms that would normally absorb it.

What to Watch

The fiscal crowding-out thesis is a structural argument about where pressures are accumulating, not a claim that the pressures have produced the predicted outcome. Three indicators: the 10-year Treasury yield and ACM term premium (the cleanest signal — if it continues rising alongside deficit expansion, the mechanism is operating); hyperscaler corporate bond spreads relative to Treasuries (if spreads widen alongside yield increases, AI infrastructure financing costs are rising on both dimensions simultaneously); and project deferrals — data center construction permits, announced project delays, and capital expenditure guidance revisions in quarterly earnings.

The independence of the Federal Reserve is also relevant here — a central bank seen as potentially yielding to political pressure cannot play its normal role as an anchor for long-run inflation expectations, which adds a further uncertainty premium to long-duration yields.

The AI buildout is expensive. The war is expensive. The interest on the debt from both is expensive. These are not three separate budget lines. They are three claims on the same pool of capital. The question of which one yields is still open.

Sources

NPR Politics. White House seeks record defense budget, separate Iran war supplemental. April 17, 2026. https://www.npr.org/2026/04/17/nx-s1-5785117/white-house-seeks-record-defense-budget-but-congress-has-questions-about-spending

Congressional Budget Office. FY2026 Budget Projections. https://www.cbo.gov/publication/62105

CBO February 2026 Baseline. https://budget.house.gov/imo/media/doc/cbo_baseline_february_2026.pdf

Pentagon FY2026 Budget. MeriTalk. https://www.meritalk.com/articles/pentagon-unveils-1-01t-fy2026-budget-with-cyber-space-ai-focus/

TD Economics. US Defense Spending Impacts. https://economics.td.com/us-defense-spending-impacts

Wharton Budget Model. Capital Crowd-Out Effects of Government Debt. https://budgetmodel.wharton.upenn.edu/p/2021-06-28-explainer-capital-crowd-out-effects-of-government-debt/