AI’s Capital Squeeze Reaches U.S. Treasuries

30-Second Signal

The cleanest confirmation of SIAIntel's capital-collision thesis came from inside the Federal Reserve. Reuters' report on Christopher Waller says the Fed governor argued that the historical safety premium attached to liquid U.S. government debt has largely disappeared, raising his estimate of the neutral interest rate. He also tied the rise in yields not only to fiscal deficits but to competition for capital from artificial-intelligence infrastructure. That is the crucial link: AI is no longer competing only for chips, electricity and land; at the margin it is competing with the U.S. Treasury for long-duration savings.

At almost the same moment, Norges Bank Investment Management's official bond-strategy analysis recommended cutting the government-bond share of its benchmark from 70% to 50%. Appendix G shows U.S. government bonds falling from 34.1% to 21.9%, while the benchmark's USD currency weight barely changes from 52.9% to 52.5%. Reuters' account of the Norwegian proposal estimates that, if approved and implemented, the change could imply roughly $80 billion less Treasury exposure from a current position of about $215 billion. This is a proposal, not a completed sale, and implementation would be gradual.

The supply side is moving in the opposite direction. The U.S. Treasury's borrowing estimate projects $739 billion of privately held net marketable borrowing in July-September and $628 billion in October-December, or $1.367 trillion across two quarters. Treasury's official daily curve put the 10-year par yield at 4.78% and the 30-year at 5.24% on September 4. Meanwhile Reuters' reporting on AI-related corporate borrowing says 2026 AI-linked issuance has surged to about $220 billion among major technology borrowers and that investor fatigue is starting to appear.

SIAIntel verdict: this is not evidence of a dollar flight. It is evidence that Treasury debt may have to compete harder for the same global pool of capital. The sharper formulation is not de-dollarization, but de-Treasurization at the margin.

*Visual disclosure: the feature image is an AI-generated editorial composite. It is illustrative and is not evidence of any Treasury transaction, NBIM trade or Federal Reserve action.*

What Changed in the Capital Market

The new information is not one isolated yield print. It is the alignment of three independent channels. First, a Fed governor is explicitly connecting AI infrastructure spending to the price of capital. Second, a giant long-horizon allocator is proposing to reduce the government-bond share of its benchmark while keeping its currency mix nearly unchanged. Third, sovereign and corporate supply are arriving together while long yields are already elevated.

That combination changes the analytical question. The old question was whether Washington could issue large volumes of debt while central banks gradually normalized policy. The new question is whether Washington can still command a special price when hyperscalers, data centers, utilities, mortgage securities and higher-yielding government-related assets all ask investors for duration at the same time.

The broad issuance backdrop confirms that this is not a small niche. Federal Reserve new-security-issues data show U.S. corporations sold about $1.892 trillion of bonds in the first seven months of 2026. That number covers all corporate borrowing, not AI alone, but it establishes the scale of private competition for term capital. The investment-grade market is still functioning; the signal is a higher clearing price, not a closed market.

Why Treasuries Are Losing Scarcity Value

Treasuries retain enormous advantages: deep liquidity, collateral utility, regulatory treatment and the dollar's central role in global finance. The thesis does not require those advantages to disappear. It requires only the marginal investor to demand a little more yield than before.

Waller's point matters because a "safety premium" is effectively a willingness to accept less return in exchange for liquidity and perceived security. If that willingness weakens, the same quantity of federal borrowing must clear at a higher yield. Higher yields then feed directly into mortgages, corporate discount rates, project-finance hurdle rates and equity valuation.

The mechanism is circular. AI infrastructure raises demand for capital. Higher demand can lift the risk-free rate and corporate spreads. The higher all-in cost of capital then makes the next wave of AI projects more expensive. The boom can therefore tighten its own financing conditions even before credit availability disappears.

Japan adds another competitive edge. Reuters' analysis of Japanese capital flows reports that higher JGB yields are beginning to make domestic bonds more attractive and have reduced Japan's role as an automatic buyer of foreign duration. The Japanese Ministry of Finance's September 1 auction result showed a 2.995% weighted-average yield on the new 10-year JGB and a 3.011% yield at the lowest accepted price. Japan does not need to dump Treasuries for the global marginal-buyer equation to change; it is enough for fewer new dollars to leave home.

Norway Is Rotation, Not Dollar Exit

The Norwegian proposal is especially useful because it prevents the story from being exaggerated. NBIM is not proposing a strategic retreat from the U.S. dollar. Its own table shows the USD weight moving only 0.4 percentage point, from 52.9% to 52.5%. The large change is inside the dollar bond universe.

The proposed benchmark cuts U.S. government bonds from 34.1% to 21.9% and lifts U.S. non-government bonds from 16.2% to 27.6%. Japanese government bonds rise from 4.6% to 7.4%, while euro-area government bonds fall from 16.8% to 14.1%. That pattern is consistent with a search for broader risk premia, especially agency mortgage-backed and government-related bonds, rather than a geopolitical currency exit.

This distinction is the article's honesty boundary. The Reuters estimate of roughly $80 billion is an implication of a benchmark change if it is approved and implemented; it is not evidence that $80 billion of Treasuries has already been sold. The proposal is subject to the Norwegian policy process and would be implemented gradually to reduce market impact.

The Supply Collision

The capital collision can be summarized as a simple transmission chain:

Channel	Verified reading	SIAIntel interpretation
Treasury borrowing	$739B Q3 + $628B Q4	Heavy sovereign demand for savings
10Y / 30Y par yield	4.78% / 5.24% on Sep. 4	Capital already clears at a high base rate
NBIM U.S. government weight	34.1% → 21.9% proposed	Less benchmark preference for Treasuries
NBIM USD weight	52.9% → 52.5%	Rotation inside USD, not dollar exit
AI-linked debt	about $220B in 2026 among major borrowers	Private demand competes for duration
U.S. corporate bonds	about $1.892T Jan-Jul	Broader term-capital supply is exceptional

A separate Reuters analysis of the long-end Treasury selloff describes multiple supply-and-demand pressures rather than a single culprit. That is exactly the correct framing. Fiscal deficits, inflation risk, central-bank policy, oil, foreign demand, hedge-fund positioning and AI borrowing can all matter simultaneously. SIAIntel is not assigning every basis point of the 10-year yield to AI. The narrower claim is that AI has become large enough to enter the same marginal-capital equation as the U.S. sovereign.

Who Feels It First

Investors: A higher risk-free rate compresses the present value of distant cash flows even if credit spreads do not widen. Long-duration technology shares, infrastructure vehicles and growth assets are therefore exposed before a conventional credit crisis appears.

AI and data-center developers: The strongest hyperscalers can still fund themselves, but the hurdle rate rises for speculative capacity, merchant-power exposure and projects without contracted tenants. Financing quality becomes a competitive advantage.

Utilities and grid borrowers: AI infrastructure cannot be separated from power infrastructure. Utilities, transmission developers and public authorities are issuing or financing long-lived assets at the same time. The projects with clear regulated cost recovery should be more resilient than projects dependent on optimistic utilization assumptions.

Households and conventional firms: The transmission runs through mortgage rates, refinancing costs, consumer credit and public budgets. A structural rise in the neutral rate can keep financial conditions restrictive even without an immediate policy-rate increase.

Counter-Case: Why This May Not Become a Treasury Shock

There are strong counter-arguments. Treasuries remain the world's deepest sovereign bond market and a core source of collateral. A benchmark change by one fund, even the world's largest sovereign wealth fund, is small relative to the total Treasury market. Other buyers can absorb the supply if yields are attractive enough.

The NBIM recommendation itself is not motivated by AI. Its formal case is diversification, liquidity and the opportunity to earn risk premia in securitized and government-related bonds. Treating the Norwegian proposal as a reaction to AI would be false causality.

AI financing can also be less debt-intensive than headline investment totals imply. The largest technology companies generate enormous cash flow and can fund part of the build-out internally. If productivity gains eventually lift real growth, the additional capital stock could also expand the economy's capacity rather than merely crowd out other borrowers.

Finally, long yields can fall quickly if inflation cools, geopolitical energy pressure fades or investors seek safety. The thesis is therefore conditional: the signal strengthens only if heavy issuance persists while the marginal buyer becomes more selective.

Confirmation Dashboard

The next confirmation window is observable rather than narrative. Watch whether 10-year and 30-year Treasury yields remain elevated during heavy auction calendars; whether auction tails, dealer take-down or real-money demand deteriorate; whether AI-linked issuers must pay wider concessions; and whether Japanese investors continue reducing foreign-duration purchases as domestic yields remain near 3%.

On the Norway leg, the key event is political approval and the implementation timetable. If the benchmark change is diluted, delayed or rejected, the direct Treasury-flow implication weakens. If it proceeds while USD exposure stays near 50%, it would reinforce the distinction between currency confidence and security selection.

The thesis is falsified, or at least materially weakened, if Treasury demand stays robust despite heavy supply, the long end falls sustainably, AI issuance slows, Japanese outward fixed-income buying recovers and the NBIM proposal does not progress. Those conditions would show that global savings remain sufficient to absorb the new capital stack without a persistent price penalty.

These are monitoring thresholds, not trading instructions.

SIAIntel Bottom Line

The important signal is not that one investor is "selling America." The Norwegian proposal keeps roughly the same dollar weight while moving away from U.S. government bonds toward other high-quality dollar assets. At the same time, the Federal Reserve's Christopher Waller is explicitly saying that AI infrastructure is part of the competition for capital that is helping erode the old Treasury safety premium.

Washington is asking investors to absorb $1.367 trillion of net marketable borrowing across two quarters while AI and the broader corporate sector are issuing at extraordinary scale. Japan's 3% sovereign yield adds another credible home for global savings. None of these facts proves a Treasury crisis.

Together, however, they change the price-setting environment.

SIAIntel final signal: AI's next bottleneck may not be chips or electricity. It may be the price of capital — and U.S. Treasuries are now inside that competition.