How Companies Actually Get Financed

Key takeaways

• There are two unrelated credit markets: U.S. nonfinancial corporations owed $14.2 trillion at the end of 2025, 56% of it in tradeable bonds, while 59% of small firms carrying debt had to sign a personal guarantee to get it.
• Equity never sends an invoice and is still the expensive money: Damodaran's January 2026 U.S. dataset puts the market-wide cost of equity at 8.02% against an after-tax cost of debt of 3.97%.
• The interest tax shield has a legal ceiling. Section 163(j) caps the deduction at 30% of adjusted taxable income, and Congress changed the definition of that income twice in four years, so the same debt got cheaper, then dearer, then cheaper again.
• Covenants are a price before they are a tripwire: one filed revolver charges 1.25% over SOFR below 2.5x leverage and 2.25% above 4.0x, and jumps straight to the top rate if the quarterly compliance certificate arrives late.

There is no such thing as the credit market

"How companies get financed" is a single phrase covering two markets that share almost nothing but a verb. At one end, a company issues a security that strangers buy, trade and rate. At the other, an owner signs a personal guarantee at a branch. Both are called borrowing. Almost no other feature is the same.

The large end is a securities market. The Federal Reserve's Financial Accounts of the United States put nonfinancial corporate business debt at $14.2 trillion in the fourth quarter of 2025, of which corporate bonds were $7.9 trillion, or 56% of the total, with loans, mortgage and non-mortgage, making up 38%. That debt grew at just a 1.0% annual rate in the quarter, well below its recent average. SIFMA, tracking the whole U.S. corporate bond market including financial issuers, reports $11.7 trillion outstanding as of the first quarter of 2026 and $1,681.0 billion of new issuance in the first seven months of 2026, up 26.9% year over year. In this world the lender is a market. Price is set by auction, the instrument is transferable, the borrower's name is a rating, and nobody has signed away their house.

The small end is an underwriting market, and its numbers describe a different animal. The Federal Reserve Banks' 2026 Report on Employer Firms, drawn from 6,525 responses to the 2025 Small Business Credit Survey, found that 60% of firms applied for financing in the prior twelve months. Of those applicants, 42% received the full amount they sought, 36% received some or most, and 22% received nothing. Of the firms that carry debt, 59% secured it with a personal guarantee and 51% pledged business assets. Thirty-one percent of employer firms carry no debt at all, up from 21% in the 2020 survey. Underneath that, the Small Business Administration reported guaranteeing 77,600 7(a) loans worth $37 billion in fiscal year 2025, plus 6,750 504 loans worth $7.8 billion.

The difference that matters is not size, it is who bears the search cost. In the bond market, price discovery is done for you: a thousand buyers argue about what your risk is worth, continuously, in public. In the small-business market there is no price discovery. There is one lender's opinion at a time, and you find out what the money costs by living with it. The same survey has the sharpest evidence of that asymmetry anywhere in public data: 60% of firms that borrowed from an online lender said their actual borrowing costs came in higher than expected, against 37% at small banks and 32% at large banks. Meanwhile the share of applicants approaching online fintech lenders climbed from 17% in the 2020 survey to 29% in the 2025 one. More borrowers are walking into the corner of the market with the worst price information, and they are systematically surprised in one direction.

Which is why advice imported from one world is malpractice in the other. "Optimize your capital structure" is a sensible instruction to a treasurer with a rating and a revolver. Said to an owner whose only lever is whether to sign the guarantee, it is noise.

What each kind of money costs, in numbers

Debt has a price printed on a page. Equity has a price nobody ever bills you for. Both are real, and the invisible one is usually larger.

Start with the visible one. In the Federal Reserve's H.15 release for 17 August 2026, the effective federal funds rate was 3.63%, the three-month Treasury bill 3.72%, the ten-year Treasury constant maturity 4.72%, and the bank prime loan rate 6.75%. Almost no business borrows at any of those. What a business signs is a benchmark plus a margin: prime plus a spread for small commercial credit, Term SOFR plus a spread for corporate facilities. That distinction is the whole game. You did not agree to a rate; you agreed to a spread over something that moves. A borrower at prime plus two is paying 8.75% today and will pay whatever prime does next year, plus two, forever, regardless of how the business is doing.

Now the invisible one. Aswath Damodaran's January 2026 cost-of-capital dataset, covering 5,994 U.S. firms, puts the market-wide cost of equity at 8.02%, the pre-tax cost of debt at 5.29% and the after-tax cost of debt at 3.97%, on a capital structure that is 73.98% equity and 26.02% debt, for a blended cost of capital of 6.96%. The industry spread inside that average is the interesting part: general utilities carry a cost of capital of 4.36%, systems and application software 9.34%. That gap is not a preference. It is the market saying a dollar of software cash flow five years out is worth materially less today than a dollar of regulated utility cash flow, because one is more certain and can safely carry more debt.

Illustrative only, with round numbers: a company funded by $60 million of equity and $40 million of debt, with a cost of equity of 10%, a pre-tax borrowing rate of 7% and a 21% tax rate. The after-tax cost of debt is 7% x (1 - 0.21) = 5.53%. The weighted average is 0.60 x 10% + 0.40 x 5.53% = 6.00% + 2.21% = 8.21%. Now the trap. Push the mix to 50/50 and the arithmetic appears to lower the blended cost, so the model says value went up, so you do it again. It does not work, because the cost of equity is not a constant: levered equity is riskier equity, and the 10% rises as debt grows. Every extra turn of leverage buys a cheaper slice and makes the remaining slice dearer. The naive spreadsheet that leaves the cost of equity fixed will recommend borrowing until the company is 100% debt, which is also a description of bankruptcy.

The second thing textbooks bury: a weighted average cost of capital is the discount rate for the average project of the average firm. Apply the parent company's 7% to a venture whose risk looks like the 9.34% column and the project clears a hurdle it should not have cleared. That is how a stable business talks itself into a speculative one: not through recklessness, but through using one number twice.

The tax shield and its ceiling

Interest is deductible. Dividends are not. That single asymmetry is the entire tax case for debt, and it is where most leverage models stop. It should not, because the deduction has a statutory ceiling and the ceiling has been moving.

The rule is section 163(j). Per the IRS, deductible business interest expense in a taxable year cannot exceed the sum of the taxpayer's business interest income, 30% of adjusted taxable income (ATI), and floor plan financing interest expense. Interest disallowed in the current year is carried forward to the next taxable year, still subject to the limit there.

What makes this live rather than academic is that ATI has been redefined twice inside four years. For tax years beginning after 31 December 2021 and before 1 January 2025, depreciation, amortization and depletion were not added back when computing ATI, giving an earnings-before-interest-and-tax-shaped base. For tax years beginning after 31 December 2024, the One, Big, Beautiful Bill restored the add-back, making ATI an EBITDA-shaped base again and raising the cap. Identical company, identical debt, identical interest bill; the deductible amount changed because a definition changed.

Illustrative only, round numbers, and note that ATI is a defined tax term computed from taxable income rather than literally EBITDA or EBIT, so this example only approximates the shapes. Take a capital-heavy company with $50 million of EBITDA, $20 million of depreciation and amortization, so $30 million of EBIT, paying $12 million of interest. On the EBITDA-shaped base, the cap is 30% x $50 million = $15 million, and all $12 million of interest is deductible. On the EBIT-shaped base, the cap is 30% x $30 million = $9 million, and $3 million is disallowed and carried forward. At a 21% federal rate that is roughly $630,000 of cash tax a year the model said would not exist: recurring, not one-off, and largest in exactly the capital-heavy businesses that leverage up in the first place. The asymmetry is cruel: the companies whose ATI base swings most with the definition are the same companies that borrowed against the shield.

Most owner-operated businesses never meet this rule. The limitation does not apply to an exempt small business meeting the section 448(c) gross receipts test, meaning average annual gross receipts of $25 million or less over the previous three years, inflation-adjusted to $30 million for 2024 and $31 million for 2025. Certain regulated utility trades or businesses are excepted outright, and eligible real property and farming businesses may elect out, though electing real property businesses must then depreciate the relevant assets under the alternative depreciation system and lose bonus depreciation on them, which is the elegant part of the design: you may keep the interest deduction or the fast depreciation, not both.

The practical takeaway for a growing company is unglamorous. The gross receipts test looks at a three-year average, so the year you cross it is decided by the three years behind you. A company that models its interest cost as fully deductible in perpetuity is building a shield into its cash forecast that a revenue milestone can switch off.

Covenants are a price before they are a tripwire

Most people picture a covenant as a rule you either break or do not. In a real credit agreement it is mostly a dial that resets your interest rate every quarter, whether or not you are anywhere near trouble.

Here is a filed one. Centuri Holdings (NYSE: CTRI), a utility infrastructure services company, attached its credit agreement as an exhibit to the Form 10-K it filed on 26 February 2026. Section 9.13 sets two financial covenants, tested on the last day of each fiscal quarter. The consolidated interest coverage ratio (four consecutive quarters of consolidated EBITDA divided by consolidated interest expense) may not fall below 2.50 to 1.00. The consolidated total net leverage ratio may not exceed 4.50 to 1.00 for quarters ending before 30 September 2026, and 4.00 to 1.00 for quarters ending on or after that date, with an option to step back up to 4.50 to 1.00 in connection with a permitted acquisition carrying more than $150 million of cash consideration.

The part that costs money every single quarter is the pricing grid. On the revolving facility, the margin over Term SOFR is set by the same leverage ratio: 1.25% at or below 2.50x, 1.50% between 2.50x and 3.00x, 1.75% between 3.00x and 3.50x, 2.00% between 3.50x and 4.00x, and 2.25% above 4.00x. The commitment fee charged on the undrawn portion moves alongside it, from 0.150% to 0.350%. Read that as an operator and the message is blunt: roughly 25 basis points per turn of leverage, applied to everything drawn, with no breach involved at all. Illustrative only: on $300 million drawn, sitting at the top pricing level rather than the bottom costs about $3 million a year in extra interest, entirely inside compliance.

The detail nobody mentions is in the same clause. If the officer's compliance certificate is not delivered when it is due, pricing snaps to the most expensive level until the certificate arrives. Late accounting is a priced event. A finance team that treats the quarterly close as an internal housekeeping chore is, in a facility structured this way, converting administrative slippage directly into interest expense.

And the structure is reflexive in a way that punishes exactly the wrong moment. Both ratios are computed on trailing four-quarter EBITDA. One bad quarter simultaneously raises leverage and lowers coverage; the grid then raises the margin; the higher margin raises interest expense; higher interest expense lowers coverage again next quarter. Nothing operational has to get worse for the numbers to keep getting worse. The Federal Reserve's May 2026 Financial Stability Report describes the population where this bites: investment-grade borrowers, holding nearly 70% of outstanding debt, kept robust interest coverage, but the median coverage ratio for non-investment-grade firms stayed low, in the bottom quartile of its historical distribution, and for leveraged loan borrowers it remained near its historical low. The same report notes elevated use of payment-in-kind provisions in private credit, arrangements where interest is added to principal rather than paid in cash. PIK is not a feature. It is a symptom that has been given a product name.

What actually breaks

Companies rarely fail because their discount rate was fifty basis points wrong. They fail on timing, and on the fact that the rate was never fixed.

Refinancing is the first failure mode, and it is not the one people model. A term loan does not blow up because you cannot pay the interest; it blows up because the principal comes due and the market will not renew it. That is why maturity dates cluster into what the market calls walls: a cohort of borrowers who all financed in the same window arrive at the same window to be judged again, and the judging is done under whatever conditions happen to exist on that date rather than the conditions that justified the loan. A business whose plan requires a refinancing it has not yet secured has an unstated assumption about someone else's future appetite sitting in its base case.

Floating rates are the second. Illustrative only: a company with $200 million of floating-rate debt, $45 million of EBITDA and $15 million of interest expense is running at a 3.0x coverage ratio. A 200 basis point move in the underlying benchmark adds $4 million of annual interest, taking coverage to about 2.4x. Nothing happened to the business. No customer left, no margin compressed, no plan failed. The Financial Stability Report is direct about which borrowers this reaches: debt-servicing capacity was lower among riskier firms relying on floating-rate debt such as leveraged loans and private credit.

The third is the credit window itself, which is a market condition and not a decision. The Federal Reserve's July 2026 Senior Loan Officer Opinion Survey reported that banks left standards basically unchanged for commercial and industrial loans to firms of all sizes, with significant and moderate net shares of banks reporting narrower spreads to large and middle-market firms and to small firms respectively, and moderately stronger demand from large and middle-market borrowers. Earlier in the same year, the January and April surveys had reported modest net tightening. Standards move on a cycle nobody at your company controls, which is the entire argument for arranging credit in a good quarter you do not need it. A revolver signed while the numbers are strong is a call option on your own future distress.

The equity side breaks differently and more quietly. The price of equity is not the percentage sold; it is the percentage sold multiplied by the eventual outcome, plus the terms governing who gets paid first. Two rounds sold at "20% each" do not add to 40%: the founder retains 0.80 x 0.80 = 64%, so 36% is gone. Then liquidation preferences reorder the payout, and in every outcome except a very good one the founder's economic share is smaller than the cap table's percentage implies. Nobody breaches anything. The arithmetic simply resolves against you in the states of the world that were always most likely.

The order companies actually follow

Firms fund from internal cash first, then borrow, and issue equity last. That ordering looks like conservatism. It is closer to logic.

Stewart Myers and Nicholas Majluf set out why in 1984, in NBER working paper 1396, "Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have." Their model assumes only that management knows more about the firm's value than potential investors do, and that investors respond rationally to that fact. The consequence is that issuing stock carries a signal: management is most willing to sell shares when it believes those shares are generously priced, so investors discount any issuance, so issuing is most expensive precisely when management believes it is least deserved. Their paper concludes that firms may therefore refuse to issue stock and pass up valuable investment opportunities entirely, and that the model explains "the tendency to rely on internal sources of funds, and to prefer debt to equity if external financing is required."

The behaviour shows up at the small end of the market too, where nobody has read the paper. In the 2025 Small Business Credit Survey, 31% of employer firms carried no outstanding debt at all, up from 21% in the 2020 survey, and among firms that did not seek financing, most said they simply had sufficient funding already. Retained earnings are not merely the cheapest capital. They are the only capital that costs nothing to explain.

What breaks the ordering, correctly, is the shape of the cash flow being financed. Debt is a promise: it must be paid on a schedule that does not care what quarter you are having, and its cost is capped and knowable. Equity is a partnership: it demands nothing in a bad month and takes a permanent share of every good one. The right question is therefore never "which is cheaper" but "is the cash flow that repays this a forecast or a fact." A profitable service business funding a predictable expansion with equity is buying insurance it does not need at venture prices. A speculative product bet funded with a term loan whose first payment lands next month has borrowed a certainty it does not have.

One honest caveat, because this material invites overconfidence: everything above is general education, not financial or tax advice. Section 163(j), covenant definitions and pricing grids are all documents with defined terms, and defined terms are where the money actually is. Read the agreement, not the summary of the agreement.

Put it to work

Pull your loan documents and find two things: the benchmark your rate floats over, and the grid or covenant test that resets it, then work out how far one bad quarter moves you along it. Model your interest deduction against the section 163(j) cap, not as unlimited. Before selling equity, price it at the exit: percentage sold times a plausible exit value, against the interest on the debt you refused.

Sources & references

Linked entries open the named source directly. Entries without a link say exactly what kind of reference they are — and how to check them yourself.

• Federal Reserve — Financial Accounts of the United States (Z.1), recent developments, Q4 2025
• SIFMA — U.S. Corporate Bonds Statistics (issuance, trading, outstanding)
• Federal Reserve Banks — 2026 Report on Employer Firms (2025 Small Business Credit Survey)
• Federal Reserve — H.15 Selected Interest Rates
• Aswath Damodaran, NYU Stern — Cost of Equity and Capital (US), January 2026
• IRS — Basic questions and answers about the limitation on the deduction for business interest expense (section 163(j))
• SEC EDGAR — Centuri Holdings, Inc. credit agreement (financial covenants and pricing grid), exhibit to Form 10-K filed 26 Feb 2026
• Federal Reserve — Financial Stability Report, May 2026: Borrowing by Businesses and Households
• Federal Reserve — July 2026 Senior Loan Officer Opinion Survey on Bank Lending Practices
• Stewart C. Myers & Nicholas S. Majluf (1984), NBER Working Paper 1396 — the pecking order model
• U.S. Small Business Administration — FY2025 7(a) and 504 loan volumes