Burn Multiple for AI Startups: Calculate and Improve It

Growth alone rarely closes a round anymore. Investors want to know what that growth cost. The burn multiple is the simplest way to answer that question, and it is one of the first numbers a careful investor will compute from your data room, whether or not you put it on a slide.

I'm Sergei Mochtchenkov, a CFA Charterholder and fractional CFO for AI-native startups. This guide explains how to calculate burn multiple correctly, what investors read into it, where AI companies get it wrong, and the levers that actually improve it.

What burn multiple measures

David Sacks of Craft Ventures introduced the metric in a 2020 post, The Burn Multiple. His definition:

Burn Multiple = Net Burn / Net New ARR

In his words, it answers: "how much is the startup burning in order to generate each incremental dollar of ARR?" A lower number means more efficient growth. Sacks built it by flipping Bessemer's Efficiency Score (net new ARR divided by net burn) so the focus lands on burn.

Both inputs are measured over the same period, usually a quarter. A quarter is long enough to smooth out timing noise from a single large contract and short enough to show change between board meetings.

How to calculate it

The formula is simple. The definitions are where companies drift.

Net burn is the cash the business consumed in the period from operations: cash out minus cash in from customers. It excludes equity raised and, in most presentations, debt drawn. If you use venture debt, show net burn before financing, because that is what investors will reconstruct.

Net new ARR is the change in annual recurring revenue over the period:

• plus new ARR from new customers
• plus expansion ARR from existing customers
• minus contraction ARR
• minus churned ARR

Here is an illustrative quarter:

Item	Amount
Cash paid for payroll, compute, tools and rent	2.4M
Cash collected from customers	0.9M
Net burn	1.5M
New ARR	0.8M
Expansion ARR	0.3M
Churned and contracted ARR	(0.2M)
Net new ARR	0.9M
Burn multiple	1.7x

Two quick checks keep the number honest. First, the ARR bridge should tie: opening ARR plus net new ARR equals closing ARR. Second, net burn should tie to the change in cash on your balance sheet after removing financing flows. If either check fails, fix the inputs before you discuss the ratio.

What investors read into it

Sacks offered guideposts in plain language. Burning 2M to add 1M of ARR in a quarter is "a 2x Burn Multiple — reasonable for an early-stage startup," while burning 5M to add the same 1M is "terrible." He also warned that if "extraordinary investment (3x burn or more) is required to deliver that growth, it's an indicator that product-market fit isn't quite what it appears to be."

He also describes the metric improving by stage: a seed company "might have a Burn Multiple of 3 because it just started selling," dropping to around 2 after a Series A, with expectations rising after a Series B. My AI startup benchmarks page tracks where current investor expectations sit for AI-native companies.

The reason investors like the ratio is that, as Sacks puts it, it is "a catch-all metric." Weak gross margin, poor sales efficiency, churn and stalling growth all show up as a worse burn multiple. A single number will not tell an investor which of those is the cause, but it tells them where to look.

Why AI startups need extra care

AI companies face three distortions that traditional SaaS companies see less often.

Compute sits inside burn. Inference and training costs are real cash. When an AI product scales, compute can grow with every new customer. If gross margin after inference is thin, burn rises nearly as fast as ARR and the multiple does not improve with scale. That is exactly the "gross margin problem" Sacks describes. Track compute separately so you can show whether the multiple is driven by headcount investment or by unit economics. My guide on AI startup gross margin covers how to measure margin after inference.

Usage revenue is not automatically ARR. Many AI products charge by tokens, seats plus usage, or outcomes. Annualizing one strong month of usage overstates ARR and flatters the burn multiple. Be explicit: count committed contract value and a conservative run-rate for usage, and disclose your method. Investors will discount a burn multiple built on an aggressive ARR definition, and a definition that changes between quarters is worse.

Prepaid compute and annual billing shift cash timing. Buying reserved capacity can spike burn in one quarter. Collecting annual contracts upfront can make burn look low. Neither changes the underlying economics. Show a trailing four-quarter burn multiple next to the quarterly figure so one-off timing does not drive the story.

When a high burn multiple is acceptable

Sacks notes that before revenue, "net new ARR will be zero so the ratio will not even compute," and that a weak multiple is mostly acceptable "in the earliest years, when sales are incipient and the product is still being built."

For AI startups, heavy early research or model work can justify a high multiple for a while. The test is whether the trend is improving. A seed-stage company with a high multiple that falls each quarter tells a better story than a company with a moderate multiple that is drifting up. If it is moving the wrong way, explain the reason and the plan before an investor finds it.

How to improve your burn multiple

Every improvement either lowers burn or raises net new ARR without raising burn as much. In practice:

• Fix gross margin before you scale volume. Routing simpler requests to smaller models, caching, batching and renegotiating compute commitments reduce the cost of each new dollar of ARR. My AI compute budgeting guide walks through the planning.
• Reduce churn and contraction. Churn subtracts directly from net new ARR. Onboarding, usage alerts and expansion paths for existing accounts often move the multiple more than new hiring.
• Price for margin, not just adoption. Free tiers and uncapped usage plans can add users while adding burn. Minimum commitments and usage tiers help ARR quality.
• Sequence hiring against milestones. Hiring ahead of pipeline inflates burn before revenue arrives. Tie each hiring wave to a measurable trigger.
• Watch cash timing. Annual billing improves cash, but report it transparently so the multiple is not overstated.

Cutting costs shows up quickly. As Sacks notes, the burn multiple "instantly adjusts to the most recent period" and "doesn't care about sunk costs." That makes it a useful board metric for checking whether changes are working.

How to present it to investors

Put the burn multiple next to the inputs that explain it: net burn, net new ARR broken into new, expansion and churn, gross margin after compute, and runway. Show quarterly and trailing four-quarter views, and state your ARR definition in a footnote. If your forecast shows the multiple improving, make sure the drivers in your startup financial model actually produce that result rather than assuming it.

Burn multiple and runway are two sides of the same plan. Runway tells you how long you have; burn multiple tells you whether the spend is buying growth. My runway and burn rate guide covers the cash side.

If you are preparing for a raise and want a second set of eyes on your efficiency metrics, you can book a call.

FAQ

What is a good burn multiple?

There is no single cutoff. Sacks called 2x "reasonable for an early-stage startup" and treated 3x or more as a warning sign. Investors expect the multiple to fall as the company matures.

Should I use gross burn or net burn?

Net burn. The metric compares cash consumed after customer collections with the recurring revenue added in the same period.

Can I calculate burn multiple before I have ARR?

Not meaningfully. With no net new ARR the ratio doesn't compute. Before revenue, focus on runway and milestone progress instead.

Does compute spend count in burn?

Yes. Cash paid for inference, training and reserved capacity is part of net burn. Showing it as a separate line helps investors see whether unit economics are improving.

How often should I report it?

Quarterly is the common board cadence, with a trailing four-quarter figure to smooth out timing from annual billing or prepaid compute.