AI and tech lending

Why tech lending is different

Many software and AI companies borrow before they're profitable. EBITDA may be negative, so leverage multiples on EBITDA don't work. Lenders instead underwrite recurring revenue, the quality of that revenue, and how long the company can run before it needs more money.

These loans are often called recurring-revenue loans, sized as a multiple of annual recurring revenue (ARR) and protected by liquidity and revenue covenants rather than EBITDA tests.

Documents in the packet

• An ARR bridge: starting ARR, new customers, expansion, contraction and churn, ending ARR.
• Cohort data: how revenue from each year's customers grows or shrinks over time.
• Gross margin, including compute and hosting costs.
• Compute or cloud contracts: committed spend, pricing and term.
• Monthly burn and cash runway.

Key metrics

Typical risks

• Customer concentration: a few large customers are a large share of ARR.
• Compute cost and supply: model and hosting costs that rise with usage, or capacity that can't be secured.
• Obsolescence: a better model or a platform change makes the product less valuable quickly.
• Burn without a path to cash flow: the loan only works if the next equity round happens.
• Weak covenants: without minimum liquidity or ARR tests, the lender finds out too late.

A worked sizing example

Illustrative numbers only, not from any case. Suppose a software company has $50M of ARR growing 30% a year, net revenue retention of 110%, gross margin of 70% after compute, and burns $2M a month. It has $30M of cash and asks for a $25M term loan.

1. Leverage on ARR: $25M ÷ $50M = 0.5x. Many lenders cap recurring-revenue loans somewhere around 1x to 2x ARR, depending on retention and margin.
2. Runway: ($30M + $25M) ÷ $2M a month ≈ 27 months, before any growth in burn.
3. The question that matters: does the company reach break-even, or raise equity, well before the loan matures? If runway ends before maturity, the loan depends on the next round.

Covenants that matter

• Minimum liquidity: a floor on cash, so the lender acts before money runs out.
• Minimum ARR or revenue: tested quarterly, to catch a shrinking base early.
• Conversion to EBITDA tests: many recurring-revenue loans switch from ARR-based to EBITDA-based covenants once the company is profitable.
• Negative covenants: limits on further debt and on moving intellectual property out of the borrower group.

Questions an IC will ask

• What happens to runway if growth halves and burn doesn't?
• How much of ARR comes from the top five customers, and when do their contracts renew?
• If compute prices rise, how much of the increase can the company pass on?
• What does the lender own if the company fails: code, customer contracts, or just a brand?

Common memo mistakes

• Treating gross bookings or one-off revenue as recurring.
• Quoting software-style gross margins that leave out compute costs.
• Looking at growth without retention. Fast growth with leaky retention is fragile.
• No view on what happens if the next funding round is late or smaller.
• Missing the link between runway, the loan's maturity and the covenant package.

Practice on the AI and tech case in the library. The model template for that case rolls ARR forward year by year, builds gross margin from usage and compute cost per unit, tests committed cloud spend for shortfalls, and runs cash after interest against the liquidity and debt-to-ARR covenants.