The Fed has cut since 2024. Interest costs kept rising

A policy-rate change need not immediately change the coupons on existing fixed-rate debt.

The natural inference is that companies are paying less to borrow. Across a consistent panel of 191 listed non-financial companies, they are paying 40.3% more than in 2021, and 5.4% more than in 2024.

The relationship between the policy rate and what a company actually pays is much looser than it looks, and the gap is measurable in filings.

What the panel is

Filers reporting interest expense in all five years, 2021–2025    885
less: companies under $100M of interest expense in 2025        −601
less: unlisted subsidiary filers                                −25
less: banks and other financials (SIC 6000–6799)                −68
──────
Panel                                                             191

Two exclusions need stating.

Financials are removed. For a bank, interest expense is the cost of funding the lending business and rises alongside interest income. Left in, the aggregate measures financial-sector balance sheet growth rather than corporate borrowing costs — and it dominates: Freddie Mac ($108.4B) and Goldman Sachs ($66.8B) alone reported $175.2 billion of interest expense in 2025, nearly twice the entire non-financial panel.

The panel is fixed. Only companies that reported in all five years are included, so the series measures change at the same companies rather than change in the population. That also means it excludes anything that listed, delisted or was acquired during the period.

The panel by year

Interest expense rose 0.7% in 2022, 19.4% in 2023, 10.6% in 2024 and 5.4% in 2025. These observations do not establish which part of the change came from monetary policy.

Why the lag runs in both directions

Most corporate debt is fixed-rate and issued with a maturity of five to ten years. The coupon is set on the day it is sold and does not move afterwards. What changes a company’s average interest cost is not the policy rate but which bonds mature and what the replacement costs.

Refinancing and new borrowing are possible contributors to the pattern. Establishing their individual contributions would require a maturity schedule and debt-balance analysis for each company; this panel does not provide that evidence.

Put differently: the Fed sets the rate on new money. A company’s income statement reflects the weighted average of every dollar it has already borrowed.

The turn has started, at the company level

The aggregate hides something. Look at how many individual companies are past their peak:

Nearly half the panel is now paying less than it did a year earlier. The aggregate keeps rising because the largest borrowers are not among them.

Deere’s interest expense is 3.19 times its 2021 level. Nothing about the Fed’s last two years is visible in that number yet.

We revisited this in August 2026, once the first half of that year was filed. The aggregate did not turn — it accelerated to 10.1% — but the median company was flat and 47.7% were paying less. The largest borrowers did not join the companies paying less. They went the other way, because several of them became new borrowers.

What would make this wrong

• The frames API assigns fiscal years to the nearest calendar frame, and not every company closes in December. Deere’s “CY2025” covers October 2024 to November 2025; Oracle’s covers June 2025 to May 2026. Off-cycle filers were not excluded, so this is not a clean calendar-year series.
• A fixed panel has survivorship in it. Companies that failed, delisted or were acquired are not here, and highly levered companies are more likely to be among them. If anything this understates the increase.
• Interest expense is a gross figure. It says nothing about interest income, cash balances, or whether the borrowing funded something productive. A company can have rising interest expense and improving net interest position at the same time.
• The $100M floor is arbitrary and removes 601 smaller filers. It keeps the aggregate from being driven by rounding at very small companies, at the cost of excluding real ones.
• Two companies were spot-checked, not 191. Deere and Oracle were verified against their own filings; the rest come from the frames data as filed.
• This is a description of what happened, not a forecast. Nothing here says when the aggregate turns down, and this site does not make that kind of claim.

Check it yourself

The frames endpoints are linked below and return raw JSON. Take InterestExpense and InterestExpenseNonoperating for CY2021 through CY2025, keep the companies present in all five years, drop SIC codes 6000–6799, and sum. The live endpoints do not preserve the original historical population, so these steps are not a complete reproduction package.

If a figure here does not match, tell us and it will be corrected on the article and on the corrections log, with the date. The tag-migration problem that this panel had to work around is the same class of issue described in our note on how a number gets verified.