The US Debt Trap

Finance
Economics
Author

Josh Pearlson

Published

Aug 21, 2026

In May 2025, Moody’s stripped the US of its last triple-A credit rating, cutting it to Aa1 and joining S&P and Fitch. In fiscal 2025, the federal government spent more servicing its debt than it spent on national defense. Gross federal debt is now a bit over $39 trillion.

Each of those gets recycled into a headline roughly once a quarter, always with the same unstated implication: that the debt is compounding out of control and the interest rate is the thing doing it.

Half of that is true. The half about the interest rate is not, at least not yet, and getting it wrong leads people to watch the wrong number. My claim here is narrow: write down the actual equation that governs debt-to-GDP and the popular version of this debate turns out to be arguing about the term that is currently helping.

All market and fiscal figures are as of August 21, 2026, using the most recent complete fiscal year (FY2025, which ended September 30, 2025) for budget data.

The Interest Bill

Gross federal debt sits at 122.6% of GDP as of Q1 2026. That is above the World War II peak of 119.1% in fiscal 1946, though slightly below the all-time high of 125.9% we hit in 2020.

Gross federal debt includes money the government owes itself, mostly the Social Security and Medicare trust funds. Debt held by the public is what is actually out in the market being financed at market rates. Gross debt is 122.6% of GDP; debt held by the public is 98.7%. The second number is the one that drives interest costs, and it’s the one I’ll use for the math below. The 1946 comparison on that basis is 106%, and CBO projects we blow through it and reach 120% of GDP by 2036.

The raw stock, on either measure, undersells the problem. What matters is the flow: the interest bill.

In fiscal 2025, net interest hit $970.4 billion against $916.6 billion of national defense outlays, on $5.24 trillion of total federal revenue. That works out to 18.5 cents of every dollar the federal government collects going straight back out the door as interest.

Here is the first thing that surprised me when I actually pulled the series instead of trusting the headline: we have been here before. Interest ate 18.4% of federal revenue in fiscal 1991. It’s 18.5% now. On this metric alone, 2025 is a tie with the early Bush Sr. years, not a record by some terrifying margin.

What worries me is that a similar interest burden now comes from much more debt at a lower average rate. In 1991 that 18% burden came from a debt pile of 58% of GDP paying an average 7.5%. Today’s comes from a debt pile twice that size paying 3.3%. Same bill, half the rate, double the principal.

The 1991 burden was a hangover. Rates had already peaked in 1981 and spent the next thirty years falling, so every refinancing made the problem smaller. Today’s burden is being generated at rates that are, by the standards of the last fifty years, still cheap, and every refinancing makes the problem bigger.

Which raises the obvious question: cheap compared to what, and for how long?

How the Debt Ratio Changes

The debate usually gets compressed into a single condition you’ll see repeated everywhere:

\[r < g\]

Where \(r\) is the interest rate on the debt and \(g\) is the economy’s growth rate. Grow faster than you borrow and the debt shrinks relative to GDP without anyone ever paying it off. That comparison leaves out the primary balance. The debt ratio can still rise if new borrowing exceeds the reduction from growth outpacing interest costs.

The actual law of motion for the debt-to-GDP ratio is:

\[\Delta d = \underbrace{\frac{r - g}{1 + g} \cdot d}_{\text{snowball}} \; - \; \underbrace{pb}_{\text{primary balance}}\]

where \(d\) is debt as a share of GDP, and \(pb\) is the primary balance, meaning the budget balance excluding interest, as a share of GDP. A primary deficit makes \(pb\) negative, which makes \(-pb\) positive, which pushes \(d\) up.

The first term is the snowball: what compounding does to the existing stock. When \(r < g\) it’s negative and it eats the debt ratio for you. The second term is the new borrowing you do this year to cover the gap between what you spend and what you collect, before any interest at all.

\(r < g\) is necessary. It is not sufficient. You can have a perfectly favorable snowball and still watch the ratio climb, if the primary deficit is bigger than the snowball is helping. That is exactly what is happening in the United States right now.

Let’s plug in real numbers rather than argue about it. I’ll use nominal terms throughout, because the identity works the same way and it avoids having to pick an inflation deflator:

  • \(d = 0.998\), debt held by the public as a share of GDP at the close of FY2025 (the 98.7% quoted earlier is the more recent Q1 2026 reading; the math needs the fiscal-year figure to match the flows)
  • \(r = 3.31\%\), the effective interest rate the Treasury actually paid, which is $970B of net interest over $29.3T of average debt outstanding
  • \(g = 4.89\%\), nominal GDP growth over fiscal 2025
  • \(pb = -2.65\%\), a primary deficit of $805B, which is the $1,775B total deficit minus the $970B of interest

Snowball term:

\[\frac{0.0331 - 0.0489}{1.0489} \times 0.998 = -0.0150 = -1.50\text{ pp}\]

Primary term:

\[-pb = +2.65\text{ pp}\]

Net:

\[\Delta d \approx -1.50 + 2.65 = +1.15\text{ percentage points per year}\]

So: the snowball is currently working in our favor to the tune of 1.5 points a year, and the debt ratio is still climbing, because the primary deficit is nearly twice as large. Everyone yelling about \(r\) versus \(g\) on financial television is arguing about the term that is currently helping.

(The identity is an approximation; actual debt-held-by-public rose about 1.9 points in FY2025 rather than 1.15. The residual is stock-flow adjustment: cash balances, student loan revaluations, and other financing that doesn’t show up in the deficit. The decomposition is for intuition, not accounting.)

Why Refinancing Is Raising Interest Costs

Notice which \(r\) I used. Not the 10-year Treasury yield, which is around 4.6%. The effective rate, which is what the Treasury pays on average across a stock of debt issued over the last thirty years, much of it during the zero-rate decade.

As older debt matures, refinancing at higher rates raises the Treasury’s effective rate.

The effective rate bottomed at 1.63% in fiscal 2021 and has doubled to 3.31% in four years. It’s still more than a point below today’s 10-year yield, and the only reason for that is the queue: old, cheap paper that hasn’t matured yet.

Watch how the dotted line leads the solid one in both directions. In 1981 the market yield hit 13.9% while the Treasury was still only paying 9.2%, and the effective rate kept climbing for another year after yields had already turned. That lag is why the interest bill in the first chart peaked in 1991, a full decade after rates did. The pipeline runs in both directions, and it is currently pointed the wrong way.

The queue moves fast. The weighted average maturity of Treasury debt is about 71 months, roughly six years, and 33% of it comes due within twelve months. Treasury has to refinance $9.7 trillion of maturing securities in fiscal 2026 alone, on top of raising about $1.9 trillion in new money.

Every one of those dollars gets repriced from a 2019-vintage coupon to a 2026 one. There is no policy decision involved and no way to opt out. Absent a large decline in market rates, the effective rate keeps grinding toward the market rate, and the snowball term keeps shrinking toward zero and then flipping sign.

Here’s what that does to the arithmetic. Holding the primary deficit exactly where it is:

Scenario Snowball Primary \(\Delta d\) per year
Today (\(r\) = 3.3%, \(g\) = 4.9%) −1.50 pp +2.65 pp +1.15 pp
Growth cools to 4.0%, \(r\) holds −0.66 pp +2.65 pp +1.99 pp
\(r\) rolls up to 4.0%, \(g\) = 4.0% 0.00 pp +2.65 pp +2.65 pp
\(r\) rolls up to 4.5%, \(g\) = 4.0% +0.48 pp +2.65 pp +3.13 pp

Nothing in that table requires a crisis, a recession, or a policy mistake. It requires only that existing debt mature and be refinanced at prevailing rates, which is the single most predictable event in all of finance.

This reframes the standard “we need a primary surplus” talking point into something more precise and, I think, more useful. At today’s rates you do not need a surplus. The primary balance that holds debt/GDP flat right now is a deficit of 1.5% of GDP, so the required tightening is only about 1.1% of GDP. But once the effective rate reaches 4.5% against 4% nominal growth, the stabilizing primary balance becomes a surplus of 0.5%, and the required tightening is 3.1% of GDP. The bill for waiting is roughly two points of GDP, call it $650 billion a year at today’s output.

Four Ways Out

The equation gives us four ways to reduce the debt ratio: faster growth, a higher primary balance, lower real rates, or default.

1. Raise \(g\). Grow out of it. This is the only painless option and it’s why everyone reaches for it first. The problem is that the snowball term is already favorable and it isn’t enough. To close a 2.65% primary gap through growth alone, you’d need \(g\) to exceed \(r\) by about 2.7 points, sustained. With \(r\) heading toward 4.5%, that means nominal growth above 7%, or real growth around 5% at target inflation. The US has not sustained that outside a postwar reconstruction. An aging workforce and flat labor-force growth make it harder, not easier. AI might genuinely change the productivity math, and if it does I’ll happily revise this, but the repricing schedule is running on a six-year clock and productivity revolutions do not keep to schedules.

2. Raise \(pb\). Run a primary surplus. Several European countries did versions of this in the 1990s to hit the Maastricht convergence criteria of 3% deficits and 60% debt.

The US precedent is the late 1990s, and it’s worth being honest about what actually happened there, because it gets cited as proof of political will. Brookings’ accounting of the 1992-to-2000 swing from a 4.5% deficit to a 2.3% surplus attributes roughly 60% of the improvement to the tech-boom revenue surge, mostly capital gains, and about a quarter to the post-Cold-War defense drawdown, which took defense from 6.0% of GDP in 1986 to 2.9% by 2000. Clinton’s 1993 tax increase contributed about 0.7% of GDP, roughly a tenth of the total.

So the one modern example of a US fiscal turnaround was mostly a stock bubble and a peace dividend wearing a trenchcoat. There is no defense budget left to halve, and I would not build a fiscal plan around a second dot-com bubble. Meanwhile Social Security, Medicare, and Medicaid are roughly 46% of federal outlays and are the third rail of American politics. A 3% of GDP tightening has to come from somewhere, and every candidate is either politically radioactive or too small to matter.

3. Lower \(r\) below inflation. Inflation and financial repression. Hold nominal rates below the inflation rate so the real value of the outstanding stock erodes and bondholders are repaid in cheaper money. This is a tax on savers that never appears in any tax bill and never gets voted on.

4. Default. Argentina did it in 2001 on roughly $80 billion, and Greece restructured in 2012. Neither issued the world’s reserve currency, and neither could print the currency its debt was denominated in. The US can always make a nominal payment. Default is a choice, not a constraint, which is why it isn’t really on the menu, though a technical default via debt-ceiling brinkmanship remains an entirely separate and entirely self-inflicted risk.

Three of the four doors involve doing something hard. One of them happens on its own if the inflation gods cooperate, and it’s the only one with a track record at this debt level.

How Financial Repression Reduces Debt

The US has run the repression playbook before, and it worked. Gross debt went from 119% of GDP in 1946 to 31% in 1981. That is an 88-point reduction with no default, no austerity program, and no year in which anyone announced a policy of debt liquidation.

The popular telling is that inflation ate it. That’s roughly right but the mechanism is more specific and more interesting. Real rates in that era weren’t dramatically negative most of the time. They were pinned near zero for three decades, which, compounded against nominal GDP growth that averaged 7.9% a year from 1946 to 1981, does the same work more slowly and much less visibly.

The contrast between the two shaded eras and the gap between them is the entire argument. From 1948 to 1981 the real 10-year yield averaged +1.0% and the real 3-month bill averaged +0.04%, which is to say savers earned nothing for three decades. From 1982 to 2000, the real 10-year averaged +4.6%. Same country, same instruments, wildly different deal for the bondholder.

The tools that made the first era possible were explicit. Regulation Q capped what banks could pay on deposits starting in 1933, so savers had nowhere to run for a better rate. Bretton Woods capital controls meant they couldn’t leave the country either.

Reinhart and Sbrancia put numbers on the whole thing and their result is worth stating carefully, because it’s stronger than the “inflation ate it” version and it’s measured differently from my chart above. They don’t use a market benchmark at all. Their objection is worth quoting: there is no single government interest rate, a bill or a bond, that is appropriate to apply to a hybrid debt stock. So they rebuild a synthetic portfolio from actual coupon rates, maturities, and the marketable-versus-nonmarketable split, which captures the below-market paper that never traded and that my 10-year series can’t see.

On that measure, the real return on US government debt never once reached 3% in any year between 1945 and 1980, and was below 1% in roughly two-thirds of them. In the quarter of years it went outright negative, the transfer from bondholders to the Treasury ran about 3.2% of GDP, or 18.9% of federal tax revenue. For scale, that is larger than today’s entire primary deficit. Repression, run properly, doesn’t just slow the debt. It pays it down.

And if you think this is all ancient history, look at the amber band. In fiscal 2021 and 2022, nominal GDP grew 8.0% and 10.9% while the Treasury’s effective rate sat at 1.63% and 2.04%. The snowball term contributed −5.8 and −7.8 percentage points to debt/GDP in those two years. Debt held by the public fell from 98.6% of GDP to 95.1% despite the government running primary deficits the entire time.

That was a real, live, two-year repression episode, and most people experienced it as “groceries got expensive” rather than “the Treasury quietly retired three and a half points of debt/GDP at my expense.” Which is precisely the point. It’s the only debt reduction mechanism that requires no legislation, no vote, and no one’s name on it.

The formal term for the endgame is fiscal dominance: when the debt is large enough that the central bank can no longer set rates for price stability, because doing so would push the interest burden to crisis levels. Rate policy stops serving the inflation mandate and starts serving debt management. Nobody announces it. You infer it from the fact that rates keep being lower than the inflation data would justify.

Why the 1945 Playbook Is Harder to Run Now

The strongest objection to my argument is that today’s bondholders have more ways to avoid accepting negative real returns.

Financial repression worked in 1945 because the bondholder was trapped. Today they aren’t, in four specific ways.

The duration is short. Post-war Treasury debt was long. An inflation surprise hit a stock of bonds that had decades left to run and destroyed their real value in one shot, with no chance to reprice. Today’s weighted average maturity is about six years and a third of the stock matures inside twelve months. An inflation surprise now gets you maybe one good year of erosion before the market demands compensation and the whole thing reprices into higher coupons. You saw this happen in real time: the 2021-22 episode was highly effective and it lasted about eight quarters before the effective rate started climbing.

Some of the debt is inflation-indexed. TIPS didn’t exist in 1945. As of June 30, 2026 they stood at $2.16 trillion out of $31.1 trillion of marketable debt, just under 7%. That slice is contractually immune to being inflated away, by design.

The investor base can leave. Reg Q is gone, phased out between 1980 and 2011. Capital controls are gone. Foreign investors hold roughly 31% of federal debt held by the public, about $9.5 trillion, and they have no patriotic obligation to accept a negative real return. Domestic savers have money market funds, foreign bonds, equities, and crypto. There is no captive audience left to tax.

Everyone can see it coming. Reinhart and Sbrancia’s era ran under a monetary regime nobody had a name for yet. We now have a well-developed literature on fiscal dominance, a liquid inflation-swap market, and a bond market that reads Fed minutes for a living. Repression works best when it’s invisible, and it is much less invisible in 2026.

None of this makes repression impossible. It makes it slower, more expensive, and more likely to be resisted by the price of the very thing it operates on.

Low Inflation and Rising Interest Costs

To engineer repression you need sustained negative real rates. That requires inflation to run hot while the Fed declines to fully respond.

The economy is not currently cooperating with that plan. Real GDP is growing about 2% year over year, the labor market is holding up, and the Fed’s target range is 3.50-3.75% against inflation near target. That combination produces mildly positive real rates, which is the opposite of what the debt needs. The 2023-2026 average real 10-year yield is +1.1%.

And the bond market is pricing exactly that continuing. Ten-year TIPS breakevens, which are the yield gap between a nominal Treasury and an inflation-protected one and therefore the market’s live inflation forecast, sit at 2.34%. The market is saying: we believe the Fed, inflation returns to target, and repression never arrives.

The growth door needs productivity we don’t have. The surplus door needs a 3% of GDP tightening from a Congress that has not managed one since a stock bubble did it by accident. The default door doesn’t exist for a reserve currency issuer. And the repression door, the only one with a track record at this debt level, requires an inflation regime the Fed is explicitly mandated to prevent, on a stock of debt that is now structured specifically to resist it.

The uncomfortable synthesis is that a good economy makes the debt math worse. Strong labor markets keep the Fed from cutting, keep real rates positive, and keep the snowball term marching toward the point where it stops helping and starts compounding. The resilience everyone is celebrating is what’s closing the last door.

For now, one number is worth holding onto: 1.5 percentage points. That’s the size of the snowball cushion we’re currently running on, and it’s shrinking every time a 2019 note matures.