Risk-Free Stack

For seventy years, the US dollar and US Treasuries have been a single trade.

Owning one meant owning the other. A central bank that wanted to preserve US security bought US Treasuries. By buying Treasuries, it held US currency. In this way, the privileges of global currency and global safe assets—at a single yield—were fused into one and the same financial instrument.

That weld is about to break.

Not in the headlines of crisis news, not in default events, and not in dramatic moments when one asset replaces another. It is breaking out in the only place where such events are likely to reveal themselves early: the cost the world pays for security far beyond any return.

The premium for holding dollars remains strong. The premium for holding long-term US Treasuries has weakened markedly; for long-term Treasuries, the premium has even turned negative.

This is not a prediction. It is the conclusion of a paper published in December 2025 by Winthrop, Ritt Kilarati, and Jesse Schregg. The paper was released as Federal Reserve International Finance Discussion Paper No. 1427 and NBER Working Paper No. 35000. They distinguish currency convenience from bond convenience by deviating from covered interest parity, and they document the decoupling between the two. Dollar convenience remains robust. Bond convenience has fallen sharply, even turning negative—especially for intermediate-to-long-term bonds. The world still wants to transact in dollars, but it no longer wants to store value for decades in the duration of US Treasuries on the old terms.

That means a lot. And the framework that maps it is not the one carried by most allocators.

Risk-free assets have never been singular. They were originally a bundle of services, and it just so happened that one financial instrument could provide all of them at the same time. Today, this bundle of services is splitting into multiple layers. Each layer is provided by different financial instruments; each layer is re-priced on its own timeline; and each layer bears its own custody, jurisdiction, settlement, and political risks.

The once single question—what is a risk-free asset—now has several answers, and these answers no longer align.

First, a boundary must be drawn, because the entire argument hinges on this boundary.

Geopolitical risk has been priced in; that is nothing new. The IMF already priced it in its Global Financial Stability Report published in April 2025. The report uses news-based geopolitical risk indicators, sanctions variables, sovereign credit spreads, asset yields, and other data to show that major shocks—especially military conflicts—bring persistent and measurable premia, and it calculates a so-called geopolitical risk beta. The European Central Bank and the European Systemic Risk Board have also established a fragmented monitoring framework. Academic literature has long treated sensitivity to geopolitical risk as a pricing factor.

None of this is mine.

Anyone claiming to have found that political factors affect prices is either uninformed or simply selling commodities.

My contribution here is not to propose new factors, but to decompose. By observing the disintegration of the overall structure, pointing out the separated layers, and posing a question that existing research has not yet resolved: is the next effective measure of safety structural or reactive?

Fusion, not revelation.

Where the framework’s argument holds, it will be stated plainly. Where it becomes a research project, it will be laid out unequivocally. For readers who also have the IMF report open in another window, that sentence is what lays the foundation for the rest of the article.

So, it is this bundle.

And how it is breaking.

The meaning of safety in the past

For most of the postwar period, the tradable US Treasury debt instrument served five roles for the global system—so much so that almost no one thought to separate them.

It was reserves—an asset held by central banks to anchor their domestic currency and store national wealth in the world’s unit of account.

It was collateral—common pledge-grade collateral for the repo market, and a qualifying Level-1 liquid asset under bank liquidity rules.

It was cash—Treasuries, the place where corporate, fund, money market instruments, and dollar institutions store and manage operating liquidity.

It was a store of value—long-term bonds purchased by pension funds, insurers, and sovereign wealth funds to secure decades of safety.

And beneath these four, there was settlement—ownership of assets transferred through clearing systems, which the world believed would remain open forever.

The brilliance of this arrangement—once called the United States’ “excessive privilege” by a French finance minister—was that these five bonds were bundled together with a single price. You did not have to choose. You bought US Treasuries and, in one transaction from the same issuing institution, received reserve status, the collateral utility, cash equivalents, duration, and final settlement rights—all at a single yield.

The convenience yield measured the value of this arrangement.

For decades, its value was enormous.

Now, bundled sales are unraveling. The services are separating. Different financial instruments take on different roles. Instruments that once provided all five services are, in some respects, clearly strong and, in other respects, clearly weak—this is the typical sign of a bundled product gradually decomposing and being re-priced.

Unbundling, layer by layer.

The long end loses its exemption

Start from the place where pressure is undeniable and the reasons cannot be argued away: long-term value storage.

The reason is fiscal arithmetic—and this arithmetic is not subtle.

In its Budget and Economic Outlook covering 2026 to 2036, the Congressional Budget Office projects that the federal deficit in the current fiscal year will reach $1.9 trillion, equal to 5.8% of output, and will expand to $3.1 trillion, or 6.7%, by 2036. Public debt will rise from 101% of GDP to 120%, surpassing the 106% record set after World War II.

The deficit is also changing its composition in the worst direction.

Net interest (debt service) as a share of output will rise from 3.3% in 2026 to 4.6% in 2036, when it will account for nearly one-fifth of total federal spending. The office notes that during the forecast period, net interest as a share of GDP will exceed 3.2% every year, setting the highest record since at least 1940.

Because the Supreme Court ruled in February 2026 that the government’s emergency tariffs are invalid, a large portion of expected tariff revenue is now legally uncertain. Fiscal problems are not dependent on the specific magnitude of tariff adjustments; the debt-servicing channels are already in operation.

That is the engine.

With a government carrying structural deficits while interest expenses keep accumulating, it must keep issuing debt into a world where demand for this single instrument is limited. Rating agencies have documented this. In May 2025, Moody’s downgraded the United States’ credit rating from Aaa to Aa1, meaning none of the three major agencies now rates the US as top-tier, citing rising debt, persistent deficits, and increasing interest costs. S&P Global Ratings made adjustments in 2011, Fitch in 2023, and the fourth agency, Scope, downgraded in October 2025.

Duo, Kilarati, and Schregg are responsible for translating arithmetic into prices. The decline in US Treasury convenience is driven by supply. Relative abundance of US Treasuries—compared with other developed sovereign bonds—is eroding their premium. Dollar convenience remains strong, but bond convenience has weakened, even turning negative in intermediate-to-long-term bonds. Europe is the opposite: in its June 2026 assessment report, the ECB noted that due to strong global demand for high-rated euro area safe assets, the German bonds’ convenience yield is rising.

This distinction matters.

The world has not abandoned the dollar; it is re-assessing the duration risk of sovereign issuers that issue dollars.

This is the fault line that most portfolios cannot foresee. Dollar shortages may coexist with poor absorption of Treasury duration.

So the first layer of structure begins to separate. Long-term assets are gradually becoming an ordinary risk asset; their pricing is no longer determined by reflexive safe-haven sentiment, but by supply, term premia, and fiscal credibility. This is not default; it is re-pricing.

What has been regarded as safe for decades is no longer seen as safe once the long-run system matures.

The front end gets an exclusive quote

Now things have changed again. The simple story that the world is losing confidence in US debt is starting to unravel.

As the long end weakens, the short end strengthens.

The source is surprisingly the least likely corner of modern finance: dollar-backed stablecoins.

Privately issued tokens promising redemption at par have become the main buyers of Treasuries. In July 2025, Congress passed legislation to write this demand into law.

The GENIUS Act took effect on July 18, 2025. The act limits bond issuance to approved issuers, requires at least 1:1 in cash and specific liquid asset reserves, including short-term government bonds and qualified repos, strictly restricts the reuse of these reserves, and mandates monthly public disclosure of the reserve composition.

The result is not crypto escaping national regulation, but a regulated front-end Treasury distribution channel.

Every dollar invested into compliant payment stablecoins is pushed into a narrow collateral market. Front-end traders can find pre-set buyers, while back-end traders cannot access that opportunity.

This number is real, and it must be told as it is—not exaggerated—because it is exactly the easiest number to inflate, and then to be disproved.

In the BIS working paper by Ahmad and Aldarassoro (revised version in February 2026) on stablecoins and safe-asset pricing, the authors use daily data from 2021 to 2025 and find that inflows of roughly $3.5 billion (about two standard deviations) will lower the three-month Treasury yield by 2.5 to 3.5 basis points within ten days. The impact depends on market conditions: when Treasury supply is ample, the effect is not statistically significant; when Treasury supply is scarce, the impact rises to 5 to 8 basis points. The effect is concentrated mainly in short-term Treasuries, with limited or near-zero spillover to longer maturities. The largest issuer, Tether, contributes the most, followed by Circle. According to the issuers’ own reserve reports, by the end of 2025, major dollar-backed token holdings exceed $270 billion, including about $153 billion in Treasuries, and they bought roughly 33 billion Treasuries over the past year.

A faint force at the margin, not a flood.

But reality runs counter to the long-term trend. The cash layer of the dollar system is being deepened by private digital demand, while the duration layer is shrinking due to fiscal supply.

Treasuries and bonds—once seen as different maturities of the same asset—are now fragmenting under the influence of a variety of forces.

Splitting within the yield curve.

Here is a deeper irony that supports the argument that follows.

Stablecoins strengthen the dollar’s position, but they cannot free holders from the dollar system. It is precisely this requirement—requiring stablecoin reserves—that places issuers within the regulatory framework and requires the stablecoin payment system to comply with lawful directives, including instructions to seize, freeze, destroy, or prevent transfers of specific tokens.

Digital dollars are not anti-sovereign currencies. They are a programmable extension of sovereign currency through private channels.

This makes them a powerful tool of dollar strength, but also a weak tool for escaping that strength.

Central banks worried that dollar reserves could be frozen cannot eliminate this concern—even if they hold dollar claims that are easier to freeze.

This is the clue to the next layer.

Reserve insurance is transforming, not collapsing

The reserve layer—the assets held by official institutions to anchor currencies and store national wealth—is where safety politics becomes explicit.

In early 2022, the G7 froze about $300 billion of the Russian central bank’s reserves. This was the first time in the modern reserve era that official holders witnessed a major country’s core safe assets becoming unusable—not because of default, but because custodians and clearing systems refused to use the funds under political directives.

The asset still exists.

Its owner simply cannot touch it.

Under adversarial conditions, the distinction between credit and availability shifts from a footnote to a core issue.

First, start with an argument against exaggeration, because this argument is compelling.

The dollar’s dominance has not collapsed. The dollar remains the largest reserve currency, far ahead. According to the latest official data, the dollar accounts for nearly 57% of allocated reserves. The institutions compiling these data believe that recent exchange-rate fluctuations are mainly due to valuation effects rather than active selling. The dollar still dominates most currency transactions. A Fed study found that about three-quarters of official safe-asset holders are governments allied with US military interests, with almost no reason and almost no room to deviate.

Both are true.

The dollar can still dominate financing and trading, while official sectors can diversify at the insurance layer. Different layers require different instruments.

Watch how the insurance layer is moving.

The World Gold Council reports that in 2025, central banks worldwide bought 863 tons of gold, down from the more than 1,000 tons per year level in 2022 to 2024, but far above the 473 tons per year level in the decade prior to the freeze. In 2026, gold purchases accelerated again: 244 tons in the first quarter, up 17% quarter-over-quarter and up 3% year-over-year. Poland and Uzbekistan were the main buyers. The Council’s 2025 survey of central banks—its largest to date with 73 respondent central banks—found that 95% expect global official gold reserves to increase over the next year, and a record 43% expect their own reserves to increase.

Discipline matters more than drama.

The ECB’s report states that by the end of 2025, gold—by market value—will account for 27% of global official reserves, exceeding euro (15%) and US Treasuries (22%). This figure is striking, but misleading if taken at face value, because it mainly reflects price movements. If recalculated based on the gold price at the end of 2023, excluding the rally, the share of US Treasuries would remain far in the lead. Gold is being accumulated steadily for obvious reasons. Gold has not replaced US Treasuries as the core liquidity asset for the official sector. To think gold has replaced Treasuries is to mistake a price chart for a shift in asset structure.

Gold also has an upper limit that few enthusiasts encounter.

Gold is more effective for dealing with currency freezes than currency exchange. Russia is an example. Over the years, Russia has been reducing its dollar holdings and hoarding gold. When currency freezes arrive, those gold holdings stored in Russia’s vaults are less likely to be frozen. But outside barter with sanctioning countries, they are almost unusable for buying the sanctioned countries’ currencies that Russia needs.

Controllability is not a single attribute.

An asset may be hard to seize and difficult to dispose of, or it may be easy to dispose of but easily seized. The challenge faced by reserve managers is how to hold both types at the same time. No single instrument can meet both needs.

Deconstructing at the national level.

Behind the news about reserve shares, the group of US Treasury buyers is quietly changing.

According to the US Department of the Treasury’s International Capital data, net foreign inflows in March 2026 were $150.7 billion, but the key is the composition. Foreign private investors bought $162.1 billion, while official institutions net sold $11.4 billion. For long-term securities, foreign private investors bought $111.4 billion, while foreign official institutions sold $14.9 billion. The Treasury also notes that custody data cannot fully reflect ultimate ownership.

The Fed’s custody accounts reflect the same situation. As of the end of May 2026, the Fed held about $2.69 trillion of Treasuries for foreign official institutions and international institutions, down about $225 billion from the prior year. Its total custody amount was close to $2.97 trillion, also down nearly $290 billion.

The character of marginal buyers is changing.

From officials managing fixed exchange rates who care little about price, to private investors, money market instruments, stablecoin issuers, hedge funds, and bank balance sheets that are sensitive to price.

But this does not mean that no one is buying US products.

It means buyers’ motivations are changing. And motivation determines how people behave under pressure.

Even so, there is a key issue that we must not over-interpret, and honesty requires that we do acknowledge it. A Fed study from 2025 found that capital data severely underestimates foreign assets flowing through offshore financial centers; Cayman Islands alone is underestimated by roughly $1.4 trillion. This means that a large part of recorded private demand may be flows of leverage or intermediary funds—not slow-moving money from true reserve managers. But this does not weaken the shift in asset structure. It only makes the front end’s dependence on this demand more fragile, not less.

Channels are valuable assets.

The lowest, least noticed layer in the old bundle is settlement.

The assumption is not only that the asset can be profitable, but also that the channels through which it operates remain open.

Freezing foreign exchange reserves breaks this assumption, and this reaction is now spreading globally.

ECB officials are increasingly inclined to see payment as sovereignty rather than infrastructure. In April 2026, a member of the ECB Governing Council pointed out that Europe’s reliance on payment infrastructure outside Europe is a strategic vulnerability. He noted that a large part of euro area card transactions rely on non-European payment systems, and he sees the interoperability between the digital euro and the instant payment systems as a defense against extra-territorial influence and disconnection. The same logic is reflected in an IMF report on network risk published in January 2026, which treats the concentration of a small number of cloud service providers as a systemic issue. The effectiveness of payment requests depends on the infrastructure underneath them, and that infrastructure also has owners, jurisdictions, and bottlenecks. All of them are influenced by political factors.

The same competition is playing out at a larger scale in computing.

Over the next decade, the most important strategic assets will not be financial instruments alone, but computing capacity—including chips, data centers, energy, water resources, interconnected grids, software stacks, and licenses used to train and run artificial intelligence. Today, every major country treats it as critical infrastructure; the objective of control is security, not commerce.

In May 2026, S&P Global defined computing sovereignty as a structural risk encompassing hardware, software, jurisdiction, and operational control, and noted that suppliers or governments can pause or revoke licenses for advanced chips. This is not theoretical. In late 2025, the United States authorized exports of up to about 35,000 Nvidia advanced chips to Saudi Arabia and the United Arab Arab Emirates for state-owned entities, but with strict security and reporting conditions.

Chips come with security boundaries: licensing, reporting, end-use restrictions, and the exporting sovereign country’s continued discretion.

Computing is the purest example of control over asset availability. The availability of such assets depends on the continued licensing by states, suppliers, energy systems, and regulators. The same attributes that reserve managers discover in frozen accounts are now etched into silicon.

Computing is being financialized.

In March 2026, AI cloud service provider CoreWeave obtained an $8.5 billion delayed draw term loan to expand its platform. The initial borrowing amount was about $7.5 billion, with subsequent increases as its data center assets operate stably. The loan matures in March 2032. It was designed and underwritten jointly by Morgan Stanley and MUFG, with Goldman Sachs and J.P. Morgan as other joint lead arrangers, and Blackstone Credit and insurance companies providing guarantees. The loan received a Moody’s A3 investment-grade rating. It is the first investment-grade financing project backed by high-performance computing infrastructure and related customer contracts.

Capital is shifting toward the entity layer of digital sovereignty.

But this is not without risk. This is not a new US Treasury.

Credit ratings depend on client contracts, utilization, electricity, depreciation, export licenses, and counterparty quality. The strength of core clients may mask the fragility of peripheral ones. Investment-grade ratings borrow the strength of a few dominant platform clients and lend it to hardware that devalues each quarter.

The rebuttal I present here is, in my view, the strongest against lazy reasoning.

Countries are not passively handing over infrastructure and computing power to corporations; they are trying to regain control of these resources. In February 2026, the Brookings Institution noted that almost no country can structurally achieve full AI sovereignty. A more realistic model is controlled interdependence: governments rely on supply chains they cannot replicate and selectively develop domestic capabilities. Canada, the European Union, and Gulf states have all launched autonomous computing strategies.

This story is not about any single institution.

It is a contest. Countries use export controls, industrial policy, licensing regimes, public funding, and security regulations to reassert control over infrastructure that once seemed about to be lost. The most radical version—corporate balance sheets directly inheriting sovereignty—ignores the crucial links above the corporate layer and beyond the chip level: energy, water, land, licenses, interconnected grids, and legal frameworks.

Resistance, not conquest.

Advanced version of usable control

Stack the layers together, and the pattern emerges.

Be sure to state its limitations at the same time, because most writers are lazy about doing so.

Across each layer, the key factors affecting the value of safe assets under stress are not only credit, but also operational control—that is, the holder’s ability to actually exercise their claims when the environment deteriorates.

Controllability depends on where assets are held in custody, which jurisdictions govern the issuers and registrants, who can freeze or seize assets, whether assets can be transferred and settled when a clearing system is weaponized, whether sanctions make them inadmissible, and whether they can be liquidated at the required scale and speed.

Treasuries held long-term in fragile custody chains, digitally frozen dollars, contracts constrained by export licenses, and gold bars that are difficult to seize but hard to move—all sit at different points on this spectrum.

Two assets with the same rating may have different degrees of available control.

That gap is the frontier.

Now, the boundary is clearly drawn.

The pricing of de-risking risk has been confirmed. The IMF has modeled geopolitical risk and sanctions. The ECB and the ESRB closely track market fragmentation. The literature has built news-based geopolitical risk factors and demonstrated their ability to earn premia. If this article merely conveys that political factors have already been priced in, then it is just borrowing someone else’s results and should be discarded.

Open questions—those truly new issues—are narrower and more demanding.

Most existing metrics are passive. They are built from news, realized linkages, and attention indices. The question I cannot answer—and do not intend to answer—is: can a structural score of ex-ante usable control—built from factors such as asset fixed characteristics, custody, jurisdictions, issuer freeze rights, dependency on suppliers, settlement finality, and admissibility under sanctions—predict an asset’s performance in actual de-risking events more accurately than existing geopolitical risk beta measures?

If the structural score provides no additional information beyond known factors, then usable control variables are only a reference, not a source of returns. They may still be useful, but they are not a source of excess returns.

If it can add genuine incremental power across a series of real events (such as the 2022 freezes, continuous sanctions designations, chip controls), and appropriately remove reactive factors, then there is a safety dimension that allocators have not yet measured.

I do not know which one.

And this sentence is the most important sentence here.

Answering this requires cross-sectional event studies of asset-level returns, changes in yield spreads, flows, discounts, and settlement outcomes, controlling for known factors. The research data must come from institutional terminals, and the scoring framework must be built and validated asset by asset. This cannot be summarized in a piece of prose; it is a research plan. It cannot be answered by reasoning or reading alone—both are methods we have already tried.

What can be stated with certainty is that the published work visible from the outside does not seem to have built that kind of precise ex-ante structured scoring.

The ground seems open.

Whether it is open because it has value and remains undiscovered, or because it collapses into existing factors once it touches data—that is the key.

Rather than treat a guess as a discovery, it is better to hand over a clean sample.

What can prove this is wrong

Unfalsifiable arguments are not analysis; they are theology.

So the following will kill it, or sustain it.

If dollar reserve shares remain stable, if Treasury convenience remains stable or recovers, if official demand proves resilient after properly calculating offshore custody channels, if stablecoin demand remains too small or too unstable to matter even at the paper-currency level, and if AI capabilities expand without severe bottlenecks in power, licensing, connectivity, or sovereignty—then this framework is wrong, or at least exaggerated.

If these assumptions hold, then the so-called “de-coupling” is actually more like the ordinary combination of fiscal supply, regulation, and geopolitics. This conclusion deserves serious attention because the most powerful counterargument is based on relevant research by the IMF and the ECB. Geopolitical risk may already be fully priced through known channels, so perhaps there is no need to introduce new factors.

If official sectors continue retreating from the long end while private and stablecoin-linked demand supports the front end; if gold accumulation continues to exceed what valuation can explain; if long-end convenience remains negative; if payment and computing sovereignty are strengthened; and if the behavior of Treasury bills and long-term Treasuries becomes increasingly unlike one asset and more like two different buyers.

These signals are specific, public, and dated—that is what matters, because a framework you cannot monitor is a framework you cannot trade.

Watch the Treasury’s monthly international capital reports for the allocation ratios between official and private funds.

Watch the Fed’s weekly custody data for the official Treasury balance.

Watch the IMF and the ECB’s quarterly data on reserve composition by shares for USD, EUR, US Treasuries, and gold (adjusted for valuation).

Watch stablecoin assets and their impact on Treasury yields.

Closely track power, licenses, interoperability, and export policies—these policies will determine whether computing build-outs can proceed.

Most importantly, pay attention to one thing.

Under the term of the new Fed chair appointed under Powell, this is no longer just theory. Powell was sworn in this May. He has shown both policy continuity and policy change, and has opened discussions on trimming the balance sheet—even as debates continue about how far the Fed can shrink it. The ultimate safeguard for keeping Treasuries available is not the bonds themselves, but the central bank’s willingness to provide dollar loans against Treasuries when the system most needs funding—through swap lines, the FIMA repo mechanism, and broader liquidity architecture. The Fed is re-examining its own size, and its crisis-liquidity posture will be closely watched worldwide.

If the Fed sends signals that, in times of crisis, it is not willing to lend dollars easily, then the availability of the entire dollar reserve layer will change.

That is the catalyst calendar.

The key line

Any sensational claim that a company is safer than the US should be thoroughly refuted, because rejecting that claim is the core of the entire argument.

Such claims usually rest on comparisons of credit default swap spreads, and these comparisons fail once microstructure is involved.

The US single-issuer guarantee market is small. Fed research shows that recently published spreads may have almost no actual trading, making them unreliable as expectation indicators. The deeper mechanism issue is the “cheapest-to-deliver” option. In the 2023 debt ceiling event, assuming the deliverable cheapest bond was a heavily discounted 30-year US Treasury trading around the mid-$50s, the guarantee payout (and the spread) reflected that discounted bond and the settlement mechanism—not any assessment of the US’s ability to pay. The net unliquidated guarantee on deliverable Treasuries is negligible.

Comparing this to corporate financial statements is a category error.

Build a grand argument on this, and a fixed income specialist can be made to dismiss everything else.

Sovereigns have not been replaced by corporations.

What is happening is even quieter, and on a larger scale.

The market is not replacing sovereigns with corporations; it is re-pricing which tools, which institutions, and which channels control the safety measures that were once concentrated in the same safe layer.

The dollar still dominates the financing field.

Treasuries still dominate the front-end market, and they are now further reinforced by a digitally designed dollar designed within a regulatory framework.

Due to increased supply and declining convenience yields, long-term Treasuries are gradually stepping away from their former safe-haven role.

Gold is the insurance layer. It can be used as a hedge against confiscation of assets, but it is difficult to mobilize at scale.

Payments and computing are the battlegrounds for control.

Risk-free assets have not disappeared.

This is de-coupling.

And the job of asset allocators this century is to purposefully seek safety in each layer, rather than hope that a single tool can provide all kinds of safety.

The premium the world used to pay for the old system has vanished because the system itself is disintegrating.

The current task is to price these items.

The question I will explore next is: is available control merely the right way to view this task, or is it also a way to profit from it?

I intend to answer this with data, not convictions.