The $40 trillion supply shock
Gross US federal debt crossed $40tn in the week of 18 August 2026, having grown $3tn in a year, the fastest pace outside the pandemic, on the Financial Times's calculation. Debt held by the public is above $32tn, roughly the size of the economy.
The number that actually bites is the interest bill. Annualised US net interest payments reached about $1.21tn as of August 2026, against defence spending of about $1.17tn. This is not a first: on a fiscal-year basis the crossover happened in FY2024, when net interest of $879.9bn overtook defence outlays of $850.7bn. It is simply that the gap now gets wider every quarter without anyone voting for it.
The mechanism is dull and therefore easy to underrate. Somebody has to hold every bond that gets issued, and the only lever that clears the market is price. JPMorgan's Jay Barry framed the fiscal side of it plainly: the US runs a 6% deficit in an economy near full employment. We wrote about the arithmetic on 20 August. A government borrowing at that rate at that point in the cycle is not buying insurance against a downturn; it is spending the insurance.
The returns
The term premium is the extra yield an investor demands for owning a thirty-year bond rather than rolling short paper for thirty years. It cannot be observed directly, because nobody publishes the market's expected path of short rates, so it is estimated with term-structure models — the New York Fed's ACM and the Board's Kim-Wright are the two you will see cited. For most of the 2010s the estimates were negative, which is a technical way of saying that buyers were so keen on , and central banks were buying so much of it, that people paid for the privilege of lending long.
That has reversed. The New York Fed's estimate sits around 80 basis points, close to its highest in twelve years, as we reported on 20 August. The 30-year auction on 11 August 2026 cleared at 5.216%, the highest for that maturity since 2001, and the 10-year the following day cleared at the highest financing cost since 2007. This is not a domestic story. The 10-year JGB reached 2.93%, its highest since 1996, and the French 10-year its highest since 2009.
Treasury sells at a single clearing yield. The security trades in a forward "when-issued" market beforehand, and the difference between the auction's and the when-issued yield at the close is . A tail means the auction cleared cheaper than the market had it. Bid-to-cover is the cruder measure: total bids divided by the amount sold.
We should be honest about a discrepancy in our own files. Our dispatches have repeatedly described tailing auctions, and a separate search of the auction record for this entry could not verify any tailed 30-year auction in 2025 or 2026, while turning up four 10-year results: a 1.8bp stop-through on 6 May 2025, a 3.5bp tail on 6 August 2025, a 1.7bp tail on 11 February 2026 and a 0.1bp tail on 12 August 2026, against a recent average of roughly 0.3bp. Two of those are meaningfully worse than average; none is a market break. The clearing yields are not in dispute, and they are the thing that matters.
Meanwhile the , which prices rate volatility, was at 9.83 on 20 August 2026, down 26.9% in a month and 37% below its recent high. A high term premium with falling volatility is a specific diagnosis: nobody is scared of the next Fed meeting, they have simply stopped lending for thirty years at the old price. That is structural, and it is far harder for policy to fix than a scare.
Who is actually holding it
Foreign investors held about 32.2% of marketable Treasury debt in the most recent TIC data available for this entry, $9.371tn as of May 2026. The popular story is that foreigners are walking away. The data are less obliging: Japan's holdings were $1.143tn in May 2026 against roughly $852bn in 2020, and China's $659.3bn against roughly $592bn. Both are up.
The channel everyone worries about — Japan selling Treasuries to raise dollars to defend the yen — has also been quietly plumbed around. Since 3 August 2026 the Bank of Japan can borrow dollars against its holdings through a repurposed Fed , which we covered on 18 August. Japan can now raise dollars against its Treasuries instead of liquidating them. That lowers the risk of one particular fire and does nothing about the supply.
The buyback band-aid
On 19 August 2026 the Treasury said it would increase buybacks of 10-to-30-year bonds "by at least double", from $2bn to at least $4bn per operation, running from 9 September to 4 November — roughly $32bn a quarter by analysts' reckoning (Bloomberg). The 30-year fell almost 10 basis points to 5.187%. By the next day it had climbed as much as seven basis points to 5.27%, briefly reversing the whole move. One day.
A buyback sounds like quantitative easing. It is not, and the difference is the whole point. It creates no money. Treasury buys an old, illiquid long bond and funds the purchase by issuing something else, almost certainly bills. Total debt is unchanged, the average maturity shortens, and the market holds less duration today and more refinancing risk tomorrow. It is a trade, not a solution.
Two objections landed immediately. The first is procedural: the announcement arrived two weeks after the quarterly refunding schedule that was supposed to contain it, breaking a "regular and predictable" issuance convention the Treasury has run since the 1970s on the theory that investors pay more for debt when they know what is coming. "We do not think it is hyperbole to say that this break in communication strategy reduces the overall credibility of their guidance," said Thomas Simons of Jefferies. The second is fiscal. JPMorgan's team wrote that absent real consolidation, "we fear the markets will view this action as lacking credibility". Bloomberg's own opinion desk called the move desperate, noting that the programme was created in 2024 as market-functioning plumbing and was explicitly described by the previous administration as "not intended to ameliorate periods of acute market stress".
There is a second hand doing the opposite thing. On 13 August 2026 the New York Fed said it would halt reserve management purchases entirely from 14 August to 14 September, having already tapered them to about $10bn a month. So the government is taking duration out of private hands and funding it with bills, at the same moment stops supplying the reserves that would normally help absorb those bills. We laid this out on 19 August. The number that tells you whether it is working is , which was at 3.65% and had not moved.
The basis trade overhang
The is the mechanism by which a market with no apparent leverage turns out to have a great deal of it. A hedge fund buys the cash Treasury, sells the corresponding futures contract, and pockets the small spread between them. The spread is tiny, so the position is financed in repo at very high leverage. The Federal Reserve put the size of the trade at about $830bn as of September 2025, roughly twice the previous peak reached in early 2020.
March 2020 is the demonstration of what happens next. In the dash for cash, a broad set of holders sold Treasuries at once, volatility spiked, repo funding tightened and margin requirements rose. Dealers could not absorb the flow, the basis blew out, and the funds had to liquidate cash bonds into the very market that was breaking, which broke it further. The Fed announced on 15 March 2020 that it would buy at least $500bn of Treasuries and $200bn of agency MBS, then went further, ultimately purchasing roughly $1tn of Treasuries in about three weeks to restore function. The Congressional Research Service has a readable account.
The position is now about twice the size it was then. Nobody outside the funds knows exactly how it is distributed, or how much of the repo financing behind it rolls overnight.
Why it matters to this crash
The long end of the Treasury curve is the price everything else is quoted off, so a fiscal argument in Washington now shows up in places that have nothing to do with Washington.
It sets the mortgage rate. Thirty-year Treasuries hit 5.33% in the week of 18 August 2026, the highest since 2007, and pending home sales fell 2.3% in July 2026 to an index of 71.2, the joint second-worst reading in data back to 2001, as we covered on 19 August. Anyone holding a 3% mortgage will not sell into a 7% market, so the housing market clears through volume rather than price and simply stops.
It competes directly with the AI buildout for duration. Goldman analysts put lease commitments at about $1.5tn, up from roughly $200bn five years ago, and the paper is pricing accordingly: the Blackstone-backed QTS "Project Odyssey" bond was marketed at about 7.63% for a five-year, Baa3/BBB− rated, Microsoft-linked data centre (Bloomberg). That is a junk yield on an investment-grade rating, and it is what happens when a borrower has to outbid the US government for the same pool of long money. The same mathematics runs through , where the median share of loans at the twenty largest listed rose from 2.0% to 2.8% between the first and second quarters of 2026.
And it constrains the Fed. The July 2026 minutes showed a 9-3 hold at 3.5–3.75%, the first time since 2016 that three FOMC members dissented in the same direction, and they dissented towards higher rates (New York Times), with inflation above target for five consecutive years. Payrolls fell 23,000 in July 2026 against expectations of about 80,000 added. One arm of the state is buying duration to push the long end down while the other drifts towards pushing the short end up, and the labour market is weakening underneath both. The was at 52bp.
What would make this dangerous
The distinction that matters is between repricing and a liquidity spiral. Right now it is the first. Four things would make it the second.
The debt ceiling binds. Congress set the limit at $41.1tn last year; budget analysts now think borrowing reaches it by early 2027 rather than late, partly because of revenue lost when tariffs were invalidated, as the Washington Post reported. A standoff forces Treasury into extraordinary measures: run down the cash balance, skew issuance to bills, then refill violently once a deal passes. That drains and floods the front end at exactly the moment the buyback programme, which ends on 4 November 2026, has expired.
Repo tightens. SOFR was at 3.65% and unmoved as of 19 August 2026. If it starts printing persistently above where it should sit while the Fed is not adding reserves and bill supply is rising, the financing cost of the $830bn basis position rises with it, and the position was sized on the assumption that it would not.
Volatility joins the term premium on the way up. A high term premium with a falling MOVE is investors demanding a better price. A high term premium with a rising MOVE is investors demanding an exit. The 26.9% monthly fall in the MOVE recorded on 20 August 2026 is currently the single most reassuring number on this beat, and it is the first one we would expect to break.
The collateral loop starts. The template is not 2008, it is September 2022 in London, when a rise in gilt yields triggered margin calls on leveraged liability-driven investment strategies at pension funds, which sold gilts to meet them, which raised yields, which triggered more calls. The Bank of England had to announce temporary purchases of long-dated gilts of up to £65bn on 28 September 2022 to stop it. The spark was a rate move; the accelerant was leverage and collateral mechanics. The US version has both, at twice the March 2020 scale, and its lender of last resort is currently trying to make its balance sheet smaller.