Interest Now Takes a Fifth of US Revenue. The Long End Cannot Be Cut Out of It.

Debt service reached 20.9% of federal receipts in the second quarter, by our own calculation from the BEA accounts. What follows from that ratio is why cuts at the front end will not rescue the back, and why the next recession may lift long yields rather than lower them.

Start with one ratio. Take everything the US government collects in a year, then divide it by what it pays in interest on its debt. On the most recent quarterly accounts the answer is 20.9% — roughly a fifth of all federal revenue goes to debt service before a single pension is paid or a single road is paved.

The rest of this piece is about what that number does to the bond market. Two terms carry the argument. Fiscal dominance means the central bank's decisions are constrained by the government's borrowing needs, rather than the other way round. The term premium is the extra yield an investor demands for lending long instead of short. Put the two together and you get a market where the front end can be cut while the back end still rises.

That is the claim this piece tests — and it comes with a named source, a falsifiable set of rules, and one figure we could not reconcile.

A fifth of revenue, on our own arithmetic

Interest expense ran at a $1.25 trillion annual rate in the second quarter, against federal receipts of $5.98 trillion, both seasonally adjusted. That is 20.9%. Total outlays were 129.8% of receipts, an implied deficit of $1.78 trillion.

The path matters as much as the level. For most of the past 2 decades the ratio sat in single digits. It crossed 10% after the pandemic borrowing and has been climbing since the Fed began raising rates into a stock of debt that reprices far faster than it used to, because so much of it was issued short.

That is the whole mechanism in one line: the same stock of debt, repriced at a higher rate. No new spending programme, no new crisis — merely the passage of time through the maturity wall.

Why a rate hike does not tighten

Luke Gromen, of Forest for the Trees, puts the claim bluntly on a podcast published this month: both cuts and hikes are inflationary at this point, because the Fed is more cornered than it looks. A cut with inflation above target weakens the dollar. A hike raises the government's own funding cost.

He cites an estimate from the investor Mike Green that a rate rise adds roughly $50 billion to the deficit the following year. That is not a rounding error in a debate about whether policy is restrictive. Interest is the second-largest line in the federal budget after entitlements, which means every policy decision the Fed makes moves a fiscal line item.

There is no version of this in which handing more interest income to bondholders restricts the private sector in the way a rate rise is supposed to. It transfers spending power from taxpayers to holders of the debt, and it raises the deficit doing it.

The second-order effect is the one that matters for markets. A stronger dollar forces foreign borrowers who are short dollars to sell Treasuries to defend their own currencies. That adds to the supply the market must absorb, which pushes the long end higher. A hike that raises the long end is not a hike that tightens financial conditions in the direction intended.

The figure we could not reconcile

Gromen's sharpest claim is that interest and interest-like obligations — entitlements and veterans' affairs — already exceed 105% of receipts, putting the US at the choice between printing the interest and defaulting on it.

We could not reproduce it. Interest alone is 20.9% of receipts on the BEA accounts. Non-interest mandatory spending adds roughly 53% on the same basis, for a total near 74%. The gap looks like a window mismatch: he is describing 9 months of outlays in the fiscal year to date, which cannot be compared against an annualised receipts figure.

The direction of his claim survives the check; the number does not. We are publishing ours and stating the window, which is the rule here. Anyone repeating 105% should ask what period each side of the ratio covers.

Why the long end cannot be rescued

This is the strongest part of the argument, because it does not depend on a forecast — only on who already owns the bonds. The US net international investment position sits at a record negative 80% of GDP. Foreigners hold about $22 trillion of US assets net, including $9.4 trillion of Treasuries, and against that they are short an estimated $13 trillion to $14 trillion of offshore dollar-denominated debt.

Raise US rates and the dollar rises. Foreign borrowers who are short dollars then watch their own currencies fall, and they have to buy dollars to defend them. They sell what they can, not what they want, and what they can sell is Treasuries. Japan did exactly this in August, which is why the Treasury Secretary became involved in buybacks.

Domestically the same reflex runs the other way. Since foreign central banks stopped adding Treasuries around 2014, US regulators moved banks, money market funds and pensions into government paper — banks now hold roughly $4 trillion of Treasuries and mortgage-backed securities as high-quality liquid assets, explicitly so they can be sold in a downturn. Hedge funds hold about 7% of the Treasury market, double their share 5 years ago, much of it in basis trades levered between 20 and 100 times.

So the published deficit is not the number that has to be financed. Start at $2.1 trillion. Add a trillion a year of foreign selling if the dollar stays strong. Add degrossing by hedge funds and cash-raising by pensions. Reach roughly $4 trillion. Then note that the last three recessions widened the deficit by 600, 800 and 1,000 basis points of GDP — $1.8 trillion to $3 trillion on a $30 trillion economy. That is how you arrive at $5.8 trillion to $7 trillion in a downturn, and why the long end is more likely to rise than fall when the economy breaks.

The same dollar, two claims

Layered on top is the AI buildout. Gromen cites SemiAnalysis: roughly $11 trillion of AI capital spending between 2024 and 2029, about $6 trillion funded from cash flow and $5 trillion borrowed. Spread the borrowing over the 5 years and it is a trillion a year competing with the Treasury for the same pool of capital.

That is the same flow Bloomberg's credit desk describes from the other end. AI accounts for nearly half of all investment-grade bond issuance this year, according to Apollo Global Management data reported by Bloomberg — which means an investor buying investment-grade credit is buying an AI capital cycle, whether or not that was the intention.

Cleveland Fed President Beth Hammack said the quiet part plainly on 25 September: long-term yields are being driven higher by growth, by debt concerns, and by a government that has to compete for funding with artificial intelligence projects. Gromen's framing and hers are the same fact, one measured at the Treasury's funding cost and the other at a credit spread.

What would prove this wrong

Gromen sets his own conditions, and they are worth stating as he states them: the thesis breaks if the Fed and Treasury stand aside and let a Treasury auction fail — which would be deflationary, and which would also mean equities falling hard; if China returns to the kind of buying it did in 2004; or if the entitlement arithmetic changes, for example through a demographic shift. He notes that all three require the physical world to change rather than a policy announcement.

Two corrections belong in the same piece, because a mechanism can be right while its label is wrong.

The first is the word. What he describes — negative real rates, a state inflating a debt burden down, holders of government paper losing in real terms — is financial repression, and it is the historical norm rather than an aberration. Hyperinflation requires a state losing the capacity to tax and transact. His own cited history makes the point: of 53 countries reaching 130% debt-to-GDP over 220 years, 52 defaulted, mostly through periods of high inflation. High inflation and negative real returns are not hyperinflation, and the distinction is the difference between a thesis and a title.

The second is Bitcoin. Gromen calls it the hardest asset, with gold second. But he also concedes that it trades like a risk asset when inflation and yields rise — which is what happened this month, with Bitcoin down 1.6% on rising yields and a firmer dollar while gold rose and gold options carried more than five calls for every put. Central banks buy gold. They do not buy Bitcoin. The asset ranking inverts on its own evidence.