Venture Debt: Costs, Warrants and Loan Terms for Startups

In November 2025 the tutoring company Nerdy signed a $50.0M venture debt facility with Hercules Capital (Nerdy, 2025). Interest is the greater of prime plus 3.50% or 10.75%, and prime stood at 6.75% at the end of 2025, so the floor applies. The loan runs 48 months, the first 36 interest-only.

On top come a 7.50% end-of-term charge, a $0.3M facility charge and a prepayment charge that falls over time. Two clauses decide how much of the $50.0M is really there. The second tranche, $20.0M or 40% of the headline facility, “may be made available subject to the approval of the lenders”, and Nerdy must keep the greater of $15.0M or six months of liquidity in cash. A material adverse effect is an event of default. The filing holds the whole argument in miniature: venture debt is priced at a coupon well above its headline once the fees are counted, and the money beyond the first draw arrives at the lender’s discretion.

The pattern runs through the lenders’ own books. Two listed venture lenders reported effective yields of 13.3% and 15.3% for 2025, warrants included, and the loan agreements other companies have filed carry the same conditions as Nerdy’s. Venture debt beats equity as a bridge to a round that is already likely; as insurance against a miss it is weaker, because the documents let the lender step back when the plan slips.

Venture debt is underwritten on the next round, and the lenders say so

Silicon Valley Bank states the rule in its founder guide. Venture debt “follows equity” and does not replace it, and it “relies on a company’s access to venture capital as the primary repayment source for the loan” (SVB). The bank sizes loans at 25% to 35% of the most recent equity round, repayable over three to four years after an interest-only period of 6 to 12 months.

Darian Ibrahim, a law professor who interviewed seven of the 13 major venture lenders, found the same arrangement from the lenders’ side. Venture capitalists make an implicit promise to “repay venture loans out of their present and future equity investments” (Ibrahim, 2010). No contract binds the venture firm; lenders trust the promise because a firm that lets one loan default can be shut out of venture debt across its portfolio.

SVB Financial Group’s last annual report, signed on 24 February 2023, said repayment of its “investor dependent” loans “is often dependent upon receipt by our borrowers of additional financing from venture capitalists or others”. An early-stage client’s cash commonly dipped below the bank’s threshold for a pass-rated credit while it raised a round. The bank added: “we expect that each of our early-stage clients will reside in our criticized portfolio during a portion of their life cycle” (SVB Financial Group, 2023). Hercules Capital may downgrade a borrower “as it approaches a point in time when it will require additional equity capital to continue operations” (Hercules Capital, 2026).

The dependence has been measured in a crash. Yael Hochberg, Carlos Serrano and Rosemarie Ziedonis followed 1,519 venture-backed software, semiconductor and device startups founded from 1987 to 1999 (Hochberg, Serrano & Ziedonis, 2018). After the Nasdaq collapse of March 2000, they compared 119 startups by the age of their investors’ funds. Where the investors had recently closed a fund, the annual rate of new loans rose from 10% to 13%. Where they had not, lending fell from 17% a year in 1997 to 1999 to 1.5% in the three years after the crash. On that evidence, a startup whose lead investor has little left to deploy finds a loan much harder to get.

Lenders charge 13% to 15% because the next round repays them before startups fail

Ibrahim put the puzzle in his interviewees’ terms: “most start-ups fail, yet lenders can afford few defaults”. One lender gave him the answer: “Whether a company fails is not as important as when it fails.” Loans go in early, follow-on rounds repay them, and most failures come later, after the lender has left. His interviewees put the industry’s default rate below 5%, and one put the rate on loans that “bore the brunt of the dot.com bust” at 12%.

Survival statistics has a name for this structure. Peter Austin, Douglas Lee and Jason Fine define it: “A competing risk is an event whose occurrence precludes the occurrence of the primary event of interest” (Austin, Lee & Fine, 2016). For a venture loan, repayment from the next round is the competing event. Once a round refinances the loan, the company’s later failure no longer counts as the lender’s default, so the lender’s default rate can sit far below the startups’ failure rate. One difference matters: in a clinical study nobody chooses the competing event, while here a venture firm decides whether it happens.

Exhibit 1. What venture lenders earn and lose

Lender and loan book	Yield earned	Losses	Non-accrual at year end
Hercules Capital, 2025	13.3% effective; 12.5% core	Gross realized losses 1.73% of average cost; net 1.00%	0.2% of cost
Trinity Capital, 2025	15.3% effective	Gross realized losses 3.46% of average cost; net 3.10%	0.7% of debt at fair value
SVB, early-stage investor-dependent loans, 2022	Not disclosed	Net charge-offs 1.80% of average loans	Not disclosed by class
SVB, growth-stage investor-dependent loans, 2022	Not disclosed	Net charge-offs 0.57% of average loans	Not disclosed by class

Source: Hercules and Trinity 10-Ks for 2025; SVB Financial Group 10-K for 2022. Loss rates on average cost are our arithmetic. Effective yields include fees, and both lenders accrete the warrants’ fair value at issuance into interest income.

At Hercules, a year of gross realized losses came to about one-eighth of its effective yield (our arithmetic). Trinity, which also finances equipment, lost twice the share of its book. Warrants count for less than either side claims: Hercules held them in 108 companies, with a fair value of $41.1M against a $4.28B debt book, or about 1%. Jesse Davis, Adair Morse and Xinxin Wang reach the same verdict in their study of the market: “The warrants generate a small fraction of expected payoffs” (Davis, Morse & Wang, 2020). A lender’s low loss rate probably says more about its exit timing than about its borrowers’ chances.

Fees and warrants add 2.5 to 7 points to a venture loan’s coupon

Nerdy’s terms make a usable price list, with one caution: Nerdy is listed and larger than a Series A startup, so its terms are a reference point, short of a quote. Suppose a startup, hypothetical, has just raised a $20M Series A at $2.00 a share, an $80M post-money valuation. It borrows $6M, 30% of the round, on Nerdy’s pricing but with a younger company’s structure. That means 12 months of interest only, then 36 equal monthly payments, a 1% facility charge, the 7.50% end-of-term charge, and warrants for 5% of the loan at the Series A price, or 150,000 shares.

Exhibit 2. The all-in cost of a hypothetical $6M venture loan

Scenario	Cash paid to the lender	Annual cost (APR)	Points above the 10.75% coupon
Held 48 months; warrants expire worthless	$8.20M	13.25%	2.50
Repaid at month 24 from the Series B, with a 2% prepayment charge	$7.80M	15.44%	4.69
Held 48 months; company sold at $6.00 a share in year 5	$8.20M plus $0.60M of warrant gain	15.37%	4.62
Repaid at month 24; company sold at $6.00 a share in year 5	$7.80M plus $0.60M of warrant gain	17.96%	7.21

Hypothetical company and loan. Pricing from Nerdy’s 2025 agreement with Hercules; interest-only period from SVB’s guide; warrant coverage inside Hercules’s 3% to 20% range; the prepayment charge is assumed. Cash paid includes the facility charge. APR is 12 times the monthly internal rate of return. Every figure is computed from these terms.

Held to term, the loan costs 13.25%, within a tenth of a point of the 13.3% effective yield Hercules reported for 2025, which suggests the lenders’ reported yields are a reasonable guide to price. Repaying early raises the annual cost to 15.44%, because the fixed charges are spread over fewer months, and repayment from the next round is the plan the loan is sold on. The warrant costs nothing if the company fails and adds 2.1 to 2.5 points a year at a $6.00 exit.

Rates, though, are the wrong unit: the choice between debt and equity is paid in shares. Selling $6M of equity at $2.00 issues 3.0M shares, 6.98% of the enlarged company. Borrowing issues 150,000 warrant shares now, but every dollar paid to the lender has to be replaced, sooner or later, by equity sold at the next round’s price. The question is how much higher that price must be, as a multiple of today’s, for the loan to cost the same number of shares as selling equity now.

Breakeven next-round price ÷ today’s price = (total cash paid to the lender ÷ principal) ÷ (1 − warrant coverage)

For the loan held to term, that is 1.367 divided by 0.95, or 1.44: the Series B must price above $2.88 a share before the loan beats the equity. Repaid at month 24, the breakeven falls to 1.37, or $2.74.

Exhibit 3. Shares given up for the $6M loan, against 3.0M shares of equity sold at $2.00

Series B price	Step-up	Shares for the loan, held 48 months	Against the equity route
$1.50	0.75x	5.62M	1.87x as many
$2.00	1.00x	4.25M	1.42x as many
$2.88	1.44x	3.00M	Breakeven
$4.00	2.00x	2.20M	0.73x as many

Hypothetical company from Exhibit 2. Shares for the loan = cash paid ÷ Series B price + 150,000 warrant shares. Ignores the time value of money.

In a flat round, the loan costs 42% more shares than the equity it replaced; in a down round at $1.50, 87% more. On these terms, debt is the cheaper capital only for a company about to become more valuable.

But the cover lapses when the plan misses

Lenders also sell venture debt as a hedge. Trinity Capital lists “As an Insurance Policy” among its uses, to protect a company “from potential mishaps or delays” and avoid an emergency bridge or a down round (Trinity Capital). The agreements filed with the SEC describe a narrower policy, of which Nerdy’s lender-approved tranche is one clause.

• A lender’s judgment on every draw. Couchbase’s 2021 agreement with SVB makes each advance under its $40M line conditional on the bank determining “to its reasonable satisfaction that there has not been a Material Adverse Change”. The definition includes “a material impairment of the prospect of repayment of any portion of the Obligations”, and the same event is a default (Couchbase, 2021). The policy pays out at the insurer’s reasonable satisfaction.
• The plan as a condition. HawkEye 360’s mezzanine loan from First-Citizens, Hercules and others was subordinated under an agreement of 18 December 2025. It conditions each advance on no “material adverse deviation by Borrower from the business plan”, and adds 4.0 points to the rate after a default (HawkEye 360, 2026).
• The investors’ intentions. Pandora’s 2009 credit line from Bridge Bank, at half a point over prime, made it a default if the bank “determines that it is the intention of Borrower’s main investors to not continue to fund Borrower” (Pandora Media, 2011). Ooma’s 2015 amendment let SVB declare a default when the bank determines “in its good faith business judgment that there is a lack of Investor Support” (Ooma, 2015).

SVB’s 2022 report describes the general form: “milestone tranches of Investor Dependent loans, which are tied to company performance or additional funding rounds”. An undrawn tranche is therefore the lender’s option, and a plan that counts it as runway may find the gap in its worst month.

How often do venture lenders actually pull the cover?

The best evidence for venture debt comes from Davis, Morse and Wang. Early-stage firms, they find, use venture debt in one-third of financing rounds, and “treatment with venture debt lowers closure hazard by 1.6-4.4%” while raising successful exits by 4.3% to 5.3%. Ibrahim’s interviewees describe contracts enforced with restraint. Even when a MAC clause is tripped, lenders treat it as a “chance to start a conversation”, because “lenders will sacrifice a particular start-up to preserve their broader relationships with the VCs that send them future business.”

That evidence deserves weight: lenders rarely call loans on a MAC, and on average borrowers fare better. It has three limits. Adair Morse’s 2024 review notes evidence of “a VC selection of startups (those with lower likelihood of failure) into venture debt”, so the averages flatter the instrument (Morse, 2024). Restraint lasts only while the venture firm is worth more to the lender than the loan, and the 2000 data show what happens when it is not. And the lender can vanish. Morse quotes Andrew Metrick’s verdict that “SVB never faced serious concerns about the quality of its loans”, yet California regulators closed the bank on 10 March 2023, 14 days after signing the annual report quoted above (FDIC, 2023).

Breakeven, tranches and covenants to settle before signing

• Timing. Borrow when the next round is already likely, usually just after a raise.
• Price. Price the loan at the repayment date you expect, with fees and a warrant value at a realistic exit. On Exhibit 2’s terms, repayment from the Series B costs 15.44% on a 10.75% coupon.
• Breakeven. Work out the next-round price at which the loan costs the same shares as equity. If the plan’s next round prices below it, equity is the cheaper capital. The free dilution calculator below shows the shares behind each route.
• Runway. Count only drawn cash as runway, and treat later tranches as the lender’s option.
• Clauses. Negotiate investor-support and plan-deviation conditions out, or replace them with objective triggers such as a cash floor.
• Investors and boards. Treat a portfolio company’s loan as partly underwritten on the fund’s reserves, put the breakeven price and covenant headroom in the board pack, and ask what happens if the lender fails or is acquired.

A bridge to a likely round, with cover that can lapse

Venture debt is cheap for a reason the lenders’ books make plain: the next round repays them before startups fail, and 13% to 15% a year prices that timing. As a bridge to a round the company would raise anyway, it costs fewer shares than equity whenever the next round prices at about 1.4 times the last or more. As insurance against a miss it is weakest where insurance must work, because tranches, MAC clauses and investor-support conditions let the lender step back when the plan slips. Whether a given lender would step back in a given company’s bad month turns on how much that company’s investors are worth to it, and no filing discloses that.

Notes and sources

• Austin, P. C., Lee, D. S. and Fine, J. P. (2016). Introduction to the analysis of survival data in the presence of competing risks. Circulation 133(6), 601–609.
• Couchbase, Inc. (2021). Amended and Restated Loan and Security Agreement with Silicon Valley Bank, 29 January 2021. Form S-1, Exhibit 10.13.
• Davis, J., Morse, A. and Wang, X. (2020). The leveraging of Silicon Valley. NBER Working Paper 27591.
• FDIC (2023). Press release PR-16-2023 on the closing of Silicon Valley Bank, 10 March 2023.
• HawkEye 360, Inc. (2026). Mezzanine Loan and Security Agreement with First-Citizens Bank & Trust Company as agent, Hercules Capital and others. Form S-1, Exhibit 10.12.
• Hercules Capital, Inc. (2026). Form 10-K for the year ended 31 December 2025.
• Hochberg, Y. V., Serrano, C. J. and Ziedonis, R. H. (2018). Patent collateral, investor commitment, and the market for venture lending. Journal of Financial Economics 130(1), 74–94; NBER Working Paper 20587 (2014).
• Ibrahim, D. M. (2010). Debt as venture capital. University of Illinois Law Review 2010(4), 1169–1210.
• Morse, A. (2024). Venture debt. NBER Working Paper 32183.
• Nerdy Inc. (2025). Form 8-K dated 3 November 2025, Loan and Security Agreement with Hercules Capital.
• Ooma, Inc. (2015). Second Amendment to Amended and Restated Loan and Security Agreement with Silicon Valley Bank, 5 January 2015. Form S-1, Exhibit 10.10.2.
• Pandora Media, Inc. (2011). Amended and Restated Loan and Security Agreement with Bridge Bank, N.A., 10 September 2009. Form S-1/A, Exhibit 10.16.
• SVB. Understanding venture debt financing (founder guide).
• SVB Financial Group (2023). Form 10-K for the year ended 31 December 2022.
• Trinity Capital Inc. (2026). Form 10-K for the year ended 31 December 2025; and Venture debt (web page).

Yields, losses and loan terms come from the lenders’ 10-Ks and the loan agreements companies filed with the SEC, and lender web pages are vendor material. Rates, shares and ratios printed without a citation are worked out from the filed figures. The company and loan in Exhibits 2 and 3 are hypothetical, priced from Nerdy’s filed terms.