The Interest Bill

The federal government paid $970B of net interest in fiscal 2025 — 18.5 cents of every dollar it collected in taxes, a hair above the 1991 peak and the highest since the series begins in 1940, and, by the Congressional Budget Office’s reckoning, on its way to 25 cents by 2035 and 37 cents by 2056.

Interest is debt times rate, so the levers are the debt (which Congress sets through the primary balance), the rate (which markets set, with the Federal Reserve and the Treasury at the margin) and growth (which shrinks the burden of both). The consequences are taken from the record — Canada in 1995, Britain in 2022, the United States in May 2025 — and from CBO’s own estimates of what borrowing does to investment and rates.

Net interest was $970B in FY2025: 3.2% of GDP and 13.8% of all spending ($970.1B ÷ $7,011.1B), more than national defense; $2.7B a day ($970B ÷ 365); $2,830 per resident ($970B ÷ 342.7 million, January–August 2026 average, BEA via FRED POPTHM). Since 2021, about 60% of the rise was higher rates; the rest the larger debt. A third of the debt refinances within a year, at 4–5.6% against a 3.49% average. A tenth of a point on interest rates is $379B on the ten-year deficit — the biggest of CBO’s rules of thumb, and the one Congress controls least — and one point adds 47 points of GDP to the debt by 2056. Of each borrowed dollar, 33¢ comes out of private investment.

Where the bill stands

Two ways of measuring the bill, and both are at highs, if only by a hair. Against receipts, interest took 18.5% in FY2025 — a whisker above the 18.4% of 1991, the previous high since 1940, which was set with a debt half today’s size relative to the economy but at rates around 8% ($194.4B of net interest ÷ $2.41T of debt held by the public at the end of FY1990). Against GDP it is 3.2%, a record by the same hair. The dashed line is CBO’s extended baseline from February 2026: it assumes rates settle at about 4% and the primary deficit — the deficit before interest — stays around 2% of GDP. On those assumptions interest alone reaches 6.9% of GDP by 2056, more than the government spends today on Social Security.

Why it grows: debt times rate

Interest is the debt multiplied by the rate on it, and the last five years show which factor has been doing the work. Gross interest went from $562B in FY2021 to $1.4T in FY2026. The debt grew 41%; the effective rate — interest paid over debt owed — rose from 2.0% to 3.4%, as the average rate on interest-bearing debt climbed from about 1.6% to 3.49%. Split symmetrically, each factor weighted by the midpoint of the other, more debt accounts for $315B of the increase and the higher rate for $483B: about 60% of the rise was higher rates (50–71% depending on how the interaction is assigned); the rest the larger debt. (The split depends on method: crediting the whole interaction to the rate gives 71%, crediting it to the debt 50%.) And the price is still catching up. The average rate is a blend of old cheap debt and new dear debt, and the old debt keeps maturing.

A third of the marketable debt comes due within twelve months and is refinanced at whatever the market charges that day: 4.17% on bills, 5.24% on ten-year notes, 5.61% on thirty-year bonds on October 1, 2026, against the 3.49% average. Bills are 22% of marketable debt, up from 13% in 2014, because bills are the flexible way to raise cash quickly; they are also the thing that reprices first when rates rise. The weighted average maturity is 71 months. GAO counts $9.7 trillion maturing in FY2026 alone.

When the average rate on the debt (r) is below the growth rate of the economy (g), a country can run a modest primary deficit and still see its debt ratio fall; the economy outgrows the interest. When r rises above g the debt grows on its own, and the primary balance has to be in surplus just to hold the ratio still. For most of 2002–2025 the United States had r below g, which is why 2010s deficits did not feel like 1990s deficits. The gap has narrowed from nine points in 2022, when inflation flattered growth, to about a point and a half in 2025 — and at the margin it has closed: new ten-year money costs 5.24% against nominal growth near 4.7%. CBO’s baseline assumes r and g both near 4% for the next thirty years, and with r equal to g every dollar of primary deficit is a dollar of permanent debt.

The levers, and how big each one is

CBO publishes rules of thumb for how the ten-year deficit moves when the economy moves. A tenth of a percentage point on interest rates is worth $379 billion over 2027–2036, in either direction; a tenth of a point of productivity growth, $317 billion; a tenth of a point of labor-force growth, $166 billion. Inflation is on the list too, at $311 billion per tenth of a point — but as a cost, not a cure: higher inflation raises nominal rates, and with a third of the debt repricing each year the Treasury pays the new rate long before inflation has eroded the old principal. The old idea that a country can inflate its debt away assumes long, fixed-rate debt and lenders who are slow to notice. Neither holds well here: a third of the debt reprices within a year, and lenders now price inflation in quickly.

The lever Congress actually holds is the primary balance. CBO’s September 2026 scenarios make the point from both ends: rates one point above the baseline take debt to 222% of GDP by 2056 instead of 175%, with a total deficit of 14% of GDP by then; cutting the primary deficit to about 0.2% of GDP holds the debt flat at 101% and, on CBO’s estimate of how debt moves rates, brings rates down on its own by about 0.3 points on average over 2026–2056. The Committee for a Responsible Federal Budget prices the goals: $9.5 trillion of ten-year savings to hold debt at 100% of GDP, $10 trillion to bring the deficit to 3% — of which $8.5 trillion is policy and $1.5 trillion is interest the Treasury then does not pay. Interest savings are the one part of deficit reduction that compounds.

Lever, Congress: the primary balance. Stop adding to the principal. Every dollar of primary deficit avoided saves the dollar and the interest on it forever; CRFB counts $1.5T of a $10T package as interest saved. The only lever fully in Congress’s hands, and the only one whose savings compound.

Lever, growth: grow the denominator. A tenth of a point of productivity is $317B over ten years and widens the margin of growth over the rate. Immigration, investment and regulatory policy act here; the effect is slow and indirect.

Lever, Treasury: maturity and credibility. Terming out the debt locks today’s 5% for decades; staying in bills keeps the option to refinance cheaper later but reprices a third of the debt each year. A credible path lowers the rate itself: CBO’s stable-debt scenario shaves 0.3 points on average over 2026–2056.

Not a lever: inflation. Raises the bill rather than cutting it: $311B per tenth of a point over ten years in CBO’s arithmetic, because nominal rates and the repricing debt move first. It also falls on savers and on wages that lag prices.

Not a lever: leaning on the Federal Reserve. The Fed holds 14% of the debt (GAO, Federal Debt Management: Treasury Is Meeting Borrowing Needs but the Deteriorating Fiscal Outlook Poses Risks, GAO-26-107529, March 2026). Asking it to hold rates down to ease the Treasury’s bill — “fiscal dominance” — buys lower interest with higher inflation, which raises the bill back through the previous lever. The 1970s are the usual cautionary tale.

Not a lever: default or restructuring. Even a delayed payment in a debt-limit standoff would raise what lenders charge; GAO lists debt-limit impasses among the standing risks to the Treasury market. The world’s cheapest borrower is cheap partly because it has never chosen not to pay.

What happens when it runs too high for too long

The first consequence is the quiet one. Every dollar the Treasury borrows is a dollar someone did not lend to a business or a homebuyer: CBO’s estimate is that 33 cents of each borrowed dollar comes out of private investment, with the rest covered by extra saving and foreign lenders. Each percentage point of debt-to-GDP raises long-term rates by about two basis points (CBO Working Paper 2024-05, Revisiting the Relationship Between Debt and Long-Term Interest Rates), so the twenty points CBO projects by 2036 (100.6% of GDP in 2026 to 120.2% in 2036) is roughly 0.4 points on the ten-year Treasury rate.

The second is the squeeze. Interest is the one line in the budget that cannot be cut, so as it grows it takes the room from everything else. It passed national defense and Medicare (net of premiums) in FY2024. On CBO’s path it passes Medicare again in 2028, stays above it through 2056, and passes Social Security in 2047. By 2056 it is 25% of all federal spending (CBO’s long-term outlook, February 2026: interest 6.9% of GDP of total spending of 27.9%). Canada reached that point first: by the mid-1990s nearly a third of every dollar Ottawa spent went to interest, and federal debt was 66.6% of GDP (Canada’s Fiscal Reference Tables, 2025 edition). The 1995 budget that followed cut program spending hard, and the books did not balance until 1997–98.

The third is the loss of the benefit of the doubt. Moody’s took away the last AAA rating on May 16, 2025, and its stated reason was the rise in debt and in this number: federal interest at 9% of revenue in 2021, 18% in 2024, about 30% by 2035 — and, counting all levels of government, 12% of revenue against 1.6% for other Aaa countries (Moody’s rating action of May 16, 2025, as quoted by Expansión the same day). Britain in September 2022 showed what the loss looks like when it comes fast: a jump of more than a point in long gilt yields within days, pension funds’ LDI funds forced to sell gilts into a falling market, and the Bank of England buying the long bonds it had been preparing to sell. None of that was a default. It was the market repricing the risk, and the repricing was the punishment. GAO’s own list of standing risks to the Treasury market now includes the fiscal path, debt-limit impasses and the dollar’s falling reserve share (72% in 2001, 58% in 2024); GAO also finds investors less willing to accept lower yields to hold Treasuries.

The other reading. Not every economist reads the bill this way. Olivier Blanchard argued in 2019 that while the rate on the debt stays below the growth rate, debt carries little fiscal cost; Jason Furman and Lawrence Summers (2020) proposed measuring the burden as real interest — net of the inflation that erodes the debt — and keeping it under about 2% of GDP. Using one year’s inflation, FY2025’s bill was about 0.7% of GDP (our calculation: $970B less 2.7% inflation on the $28.2 trillion of debt held by the public at the start of the year, against GDP of $30.3 trillion), well inside their line; Furman and Summers average inflation over five years (4.4% for FY2021–25), which puts FY2025 below zero, at about −0.9%. The disagreement is about timing more than direction: the r-against-g chart shows the cushion between growth and the rate thinning from nine points to one and a half, and on CBO’s path, with CBO’s own projection of inflation, real interest crosses 2% of GDP in 2036 (net interest of 4.59% of GDP less 2.26% inflation on $53.1 trillion of debt, against GDP of $46.7 trillion) and keeps rising.

The exhibits support one conclusion. The interest bill is not a forecast that can be argued with; it is last year’s debt times this year’s rate, and both are set. Of the three levers, growth is slow, the rate belongs to the market, and only the primary balance is in Congress’s hands — and it is the one lever that compounds in the right direction. A country that waits pays twice: once in interest, and once in the cuts the interest eventually forces.