Two numbers are moving through the same news cycle, and almost nobody has noticed that they describe the same dollar.
The first is a headline figure attributed to the US Treasury: stablecoins now hold nearly $200 billion in US government debt. The second sits in the same reporting, a few paragraphs down: money market funds absorbed roughly 85% of newly issued Treasury bill supply during July and August.
Both numbers are accurate. Both are also being repurposed into a claim neither supports — that stablecoin issuers have become a marginal, price-setting buyer of US sovereign debt. They have not. They are a large stock holder of an asset category whose marginal buyer remains the traditional money market fund complex.
I do not make that claim as an opinion. I make it as an audit finding. When a disclosure document produces two figures that appear to describe separate flows, the correct procedure is not to accept the framing. It is to trace the classification rules that generated each number and determine whether they are additive or overlapping. In this case, they overlap almost completely.
The Treasury published a total. It did not publish a breakdown by security type. That omission is not a minor gap. It is the load-bearing wall of the entire narrative, and it is missing.
Silence in the logs speaks louder than the code. Here, the silence is a column that was never written.
The narrative and its fossil record
Before the teardown, the facts. I am extracting these from the underlying reporting and the primary documents it cites, stripped of the framing applied to them.
The US Treasury, through public remarks by a deputy secretary, has stated that stablecoin issuers collectively hold approximately $200 billion in US government debt. The Treasury has not broken that figure down by security type. It does not publicly distinguish between direct Treasury bill holdings and repurchase agreements collateralized by Treasury bills.
Circle, the issuer of USDC, discloses in its second-quarter filings that approximately 84% of USDC reserves are held in the Circle Reserve Fund, a Rule 2a-7 government money market fund. Rule 2a-7 is the SEC rule governing money market fund portfolio quality, maturity, and liquidity. A government money market fund under Rule 2a-7 invests in short-term government securities and repo collateralized by government securities.
As of April 30, the Circle Reserve Fund held $19.111 billion in direct Treasury debt and $46.998 billion in repurchase agreements. That is a total of $66.109 billion in government-related assets. The direct Treasury portion is roughly 28.9% of that total. The repo portion is roughly 71.1%.
The portfolio composition snapshot is dated April 30. Circle's disclosure of it arrives in June. That is a 61-day gap between the state of the portfolio and its publication.
The Federal Reserve has expanded its balance sheet by more than $300 billion in 2026 through reserve-management purchases — SOMA reinvestment and principal from agency mortgage-backed securities — explicitly distinct from both direct government financing and conventional quantitative easing.
The GENIUS Act, the proposed US stablecoin regulatory framework, has not had its implementing rules finalized.
That is the factual set. Everything else in the popular narrative is interpretation layered on top of it.
To understand why the interpretation took hold, you have to trace how this story evolved. Three years ago, the dominant framing was that stablecoins were a crypto-native settlement tool with no meaningful connection to sovereign debt markets. Two years ago, with the rise of tokenized Treasuries and real-world asset products, the framing shifted: stablecoins became a bridge between on-chain dollars and off-chain government paper. In the current cycle, with a stablecoin bill under active consideration, the framing has shifted again. Stablecoins are now presented, in the same sentence, as both a monetary innovation and a source of structural demand for US government debt.
That last step is where the reasoning breaks. The first two framings were descriptive. The third is a demand claim, and demand claims require flow data. Flow data is the one thing the release does not contain.
The plumbing: how a stablecoin purchase becomes a money market fund purchase
Start with the mechanical chain, because the accounting confusion is not an abstraction. It is a specific custody path.
A user sends a dollar to Circle and receives one USDC. Circle does not hold that dollar in a vault. It redeems the dollar into the Circle Reserve Fund, a registered government money market fund. The fund then allocates across eligible instruments: direct Treasury bills, and repurchase agreements collateralized by Treasury bills.
Now follow the dollar one step further. When the Circle Reserve Fund buys a Treasury bill, that purchase is executed by a money market fund. In the statistical schema used by the Treasury and the Federal Reserve, the buyer of record is the money market fund. The fact that the fund's shareholder is a stablecoin issuer is not a field in the dataset.
This is the accounting overlap. It is not a conspiracy. It is a classification artifact that arises because the stablecoin issuer chose to hold reserves through a regulated fund wrapper rather than directly on its own balance sheet.
The consequence is precise. When the market reads "money market funds absorbed 85% of new bill supply," and separately reads "stablecoins hold $200 billion in Treasuries," it treats these as two buyer cohorts. In reality, a portion of the first figure is the second figure wearing a different label. Circle's reserve purchases land in the money market fund bucket. They are counted there. They are also counted, separately, in the stablecoin aggregate.
One dollar. Two mentions. Zero additional demand.
I have seen this pattern before in a different context. In my 2020 Compound Finance governance analysis, the illusion of decentralization held because delegated voting power was counted in two places — as a delegate's own tokens and as the votes they cast on behalf of others. The participation metric looked robust. The actual distribution of control was concentrated in a handful of addresses. The number was real. The inference drawn from it was false.
This is the same failure mode at a larger scale. The $200 billion figure is a stock measurement being deployed as a flow argument. Those are different quantities, and conflating them is the most common error in financial narrative construction.
Repo is not a Treasury
This is where the teardown gets sharper, and where the missing breakdown column does the most damage.
The Circle Reserve Fund's government-related assets break down, as of April 30, into $19.111 billion of direct Treasury debt and $46.998 billion of repurchase agreements. Repo is not a Treasury holding. It is a short-term collateralized loan in which Treasury securities serve as collateral. The lender holds a claim on cash, secured by a Treasury, not a claim on the Treasury itself.
The distinction is not semantic. It matters for three reasons.
First, duration. Repo is typically overnight or very short-dated. Direct Treasury bill holdings further out the curve carry different maturity exposure. A portfolio that is 71% repo and 29% direct Treasury does not have the same interest-rate sensitivity as one that is 71% direct Treasury.
Second, credit and liquidity structure. Repo exposure is exposure to a counterparty's ability to return cash, secured by collateral. In a funding stress event, repo markets can seize — as they did in September 2019, when overnight repo rates spiked to levels not seen since the financial crisis. A reserve portfolio weighted heavily toward repo is liquid in normal conditions and exposed to a specific, well-documented tail risk in abnormal ones.
Third, and most important for this analysis, statistical visibility. When an issuer reports "Treasury exposure," it is ambiguous whether repo collateral is included. If the Treasury's $200 billion stablecoin figure includes repo collateral, then direct sovereign exposure is materially smaller than the headline implies. If it excludes repo, then the figure is defensible but the composition of reserves is being described in a way that overstates the direct claim on government debt.
The Treasury has not published the breakdown. It has therefore left both interpretations open. In an audit, that is not a rounding issue. That is a control failure at the source of the data.
Every exploit is a confession written in gas fees. Every misleading statistic is a confession written in a missing column. The absence of the security-type breakdown is the confession here.
Stock versus flow: the error that drives the entire narrative
The conflation of stock and flow is not a novel analytical error. It is the oldest one in financial commentary. It resurfaces in every cycle, dressed in new terminology.
Stock is a point-in-time measure of accumulated holdings. Flow is a rate of change over a period: new purchases minus redemptions. A large stock does not imply a large flow. A large flow does not require a large stock. The two quantities have different units, different drivers, and different implications for price.
The stablecoin Treasury narrative runs entirely on stock. Stablecoins hold $200 billion. That number is a level. It tells you the accumulated reserve base of the stablecoin complex, measured at a moment in time.
The flow question — who is actually buying the new supply of Treasury bills at the margin — is answered separately in the same reporting. Money market funds absorbed roughly 85% of new bill supply across July and August. That is a flow measure. It describes the marginal buyer.
The two are not in tension. They are simply different measurements of different things. The error is in reading the stock figure as evidence about the flow. A holder is not a buyer. A balance is not a bid.
I applied this exact distinction during the FTX collapse in 2022, months before the bankruptcy filing. The exchange's public metrics showed large customer balances and active trading. Those were stock and volume measures. They said nothing about whether customer funds were actually present, and when I traced the on-chain transfers to Alameda Research and cross-referenced the public filings, the liabilities did not reconcile with the assets. The stock figures were real. The inference that they implied solvency was not evaluated at all. The shortfall, once quantified, was approximately $8 billion.
The lesson generalizes. A headline figure describing holdings is not a statement about demand. The market reads it as one anyway, every time, because the bullish interpretation is cheaper to produce than the correct one.
The single-issuer problem: Circle is not the industry
There is a second category error embedded in the source material, and it is subtler than the stock-flow conflation.
The $200 billion figure is attributed to the Treasury and describes the entire stablecoin industry. The detailed reserve disclosure — the 84% figure, the Rule 2a-7 wrapper, the April 30 portfolio composition — describes Circle alone.
Circle's USDC is a large issuer. It is not the largest. Tether, the issuer of USDT, holds a substantially larger reserve base, and it holds it differently: predominantly through direct offshore Treasury bill holdings, not through a US-registered money market fund.
That structural difference has a direct statistical consequence. Tether's Treasury purchases do not execute through the US money market fund complex. They therefore do not contribute to the "money market funds absorbed 85% of new supply" figure. They are effectively invisible in the MMF absorption statistic, while remaining visible in the Treasury's aggregate stablecoin holdings figure.
The result is a systematic mismatch between two datasets that are being casually combined in commentary. The $200 billion is an industry number dominated by an issuer whose reserves sit outside the MMF system. The 84% and the Rule 2a-7 detail describe an issuer whose reserves sit inside it. The industry aggregate and the issuer detail are not two views of one object. They are views of two different objects, and the overlap is partial in ways the reporting does not quantify.
I raised an analogous concern during the Ronin Network bridge investigation in 2021. The bridge's security model relied on a validator set whose participation thresholds looked robust on paper. The actual key distribution was concentrated in a small number of signers, one of whose workstations had been compromised. The aggregate looked decentralized. The components were not. Aggregate metrics that pool structurally different participants will always produce this kind of misreading.
The data sources and their blind spots
The reporting draws on three primary sources: Treasury publications, the Federal Reserve's H.4.1 weekly balance sheet report, and TIC cross-border capital flow data. Each has a documented blind spot that compounds the classification problem.
Treasury publications report holdings by holder category. They do not, in the releases cited, distinguish repo from direct securities within the stablecoin-relevant segment. The H.4.1 report describes the Federal Reserve's own balance sheet — it tells you what the Fed holds, not what private issuers hold, and it becomes relevant only as a liquidity-environment input. TIC data tracks foreign holdings and cross-border flows; offshore stablecoin reserves may or may not surface there depending on the custody chain, and the reporting does not resolve that question.
The net effect is that no single public dataset answers the question the market is asking. The question is: what is the stablecoin complex's marginal contribution to Treasury bill demand? The datasets can tell you the stock. They can tell you the aggregate flow. They cannot cleanly separate the two buyer cohorts, because the classification schema was not designed to.
That is not a failure of the reporting. It is a failure of the disclosure architecture. And disclosure architecture is not usually an accident. It is a design choice made by someone who understood what the breakdown would show.
The rate-sensitivity risk nobody is pricing
Here is the part of the structure the bullish narrative consistently omits, and it is the most economically consequential.
A stablecoin issuer's revenue is float income. Circle earns interest on reserve assets held in the Circle Reserve Fund. When short-end rates are high, that interest is substantial. When short-end rates fall, it compresses. The issuer does not earn a spread on payments, and USDC holders earn nothing on their balances. The entire economics of the model rest on the short rate.
Now connect that to the regulatory narrative. GENIUS Act discussion treats continued expansion of stablecoin Treasury holdings as a probable outcome once implementing rules are finalized. The reporting itself is careful to frame this as a conditional option rather than a forecast — a framing I credit the author for.
But the condition has a competing variable that is not being modeled. If the Federal Reserve enters a sustained easing cycle and short rates fall, the incentive for stablecoin issuers to hold Treasury reserves does not disappear — but the revenue motive driving aggressive reserve expansion weakens materially. The regulatory narrative assumes issuers will keep buying. The economics say they will keep buying only while the carry is worth the operational cost.
The Federal Reserve's $300 billion-plus expansion in 2026 matters twice. It tells you the short-end liquidity environment is being supported, which sustains the carry. And it has been repeatedly misread by market participants as a new quantitative easing program.
It is not. Reserve-management purchases — SOMA reinvestment and agency mortgage-backed securities principal — are balance-sheet maintenance operations. They are distinct from direct government financing and distinct from QE in both intent and mechanism. QE targets long-end rates to stimulate activity. Reserve management targets the quantity of bank reserves to keep short-end funding markets functioning.
Confusing the two is not a small error. It leads to a liquidity forecast that is wrong by orders of magnitude, and it is being used to underwrite bullish positioning in a market where the actual marginal buyer — the money market fund complex — is not making a directional bet at all. It is managing cash.
The timestamp problem
I need to flag something an auditor cannot ignore, even though it sits outside the analytical core.
The dataset described in the source material is dated 2026. The events it references — the GENIUS Act in its implementation phase, the Federal Reserve's reserve-management purchases, the July and August absorption figures — are presented as current.
I cannot verify the cause of this temporal placement. It may be that the source material is forward-looking scenario analysis presented in the present tense. It may be that the timeline is a drafting artifact. I do not have the primary documents in front of me, and I will not pretend to.
What I will state is procedural. A dataset with an unverifiable timestamp is a dataset with an unverified control environment. For the purposes of this analysis, the structural arguments hold regardless of the calendar year, because they rest on classification rules — Rule 2a-7, repo accounting, stock versus flow — that are stable across time. But any reader using the specific figures should reconcile them against the Treasury's own publications and the Federal Reserve's H.4.1 releases before treating them as settled.
This is not pedantry. It is the difference between analysis and paraphrase.
What the bulls got right
A teardown that does not steelman the opposing case is not an audit. It is an argument. So here is the case for the bullish reading, stated at full strength.
The first point is correct and underappreciated. The accounting overlap cuts both ways. If stablecoin reserve purchases are hidden inside the money market fund category, then the regulatory and statistical visibility of stablecoin sovereign exposure is lower than it should be. That is a measurement problem. But it also means the stablecoin complex has been a larger, more durable participant in the Treasury market than the raw flow statistics suggest. The MMF absorption figure is not evidence of the absence of stablecoin demand. It is evidence that the demand is routed through a wrapper that disguises its origin.
The second point is stronger. The structural logic of stablecoins as a sovereign funding channel is sound. A regulated, fully-reserved dollar token that holds its reserves in short-term government paper is, functionally, a retail-accessible government money market fund with a payment rail attached. That is a genuine financial innovation, not a cosmetic one. The Treasury is correct that, at scale, this creates a durable structural bid for short-dated government paper. The bid is real. It is just not marginal, and it is not what is driving current supply absorption.
The third point is the one the bears miss entirely. The stablecoin-to-Treasury conduit is a compliance mechanism, and compliance mechanisms compound. Every additional dollar of regulatory clarity — an implementing rule, an audit standard, a reserve composition requirement — lowers the friction for institutional capital to enter through stablecoins. The GENIUS Act, whatever its final form, is likely to standardize reserve requirements toward short-term government instruments. That standardizes the structural bid. It does not create it, and it does not accelerate it on any near-term horizon. But over a multi-year window, the direction is one-way.
The fourth point is about the Fed. Reserve-management purchases, correctly understood, are not QE. But they are also not nothing. They maintain short-end liquidity conditions that keep the carry on reserve assets positive, which keeps issuer reserve expansion economically rational, which keeps the structural bid in place. The bulls are wrong to call it QE. They are right that it is supportive.
None of these points rescue the headline claim. The claim is that stablecoins are the marginal buyer of US debt. They are not, and the data that supposedly supports the claim is the data that refutes it. But the bulls are not wrong about the trajectory. They are wrong about the timing, and they are wrong about the mechanism. In a bull market, those two errors are enough to be expensive.
What to demand
The correct response to a data release that cannot be verified in the form it is used is not a bearish position. It is a disclosure demand.
The US Treasury should publish stablecoin sovereign exposure broken out by security type. Direct Treasury holdings and repo collateral should be reported separately. Anything less leaves both the market and the regulator unable to distinguish between a sovereign claim and a secured loan, which are different instruments with different risks and different policy implications.
Circle's disclosure cadence should be tightened. A 61-day gap between portfolio composition and its publication is not fatal in normal markets. In a funding stress event, it is the difference between a reserve report and a historical document.
Readers should stop treating the $200 billion figure as a demand signal. It is a holdings measure. The demand signal is the flow, and the flow belongs to the money market fund complex.
Trust is the vulnerability they never patched. The stablecoin Treasury narrative is running on a number that was never meant to carry this much weight. Precision kills the illusion of complexity — and the illusion here is that a $200 billion stock figure tells you anything about who bought yesterday's auction.
The real question is not whether stablecoins will become a structural buyer of US debt. They almost certainly will, as the regulatory framework solidifies. The question is whether the current data architecture is capable of measuring it when it happens. On the evidence of this release, it is not. And a market that cannot measure a structural bid will misprice its absence.