A Good Burn Multiple for an AI Startup Is Under 1.5x

A good burn multiple for an AI startup is under 1.5x, which means the company burns less than $1.50 of net cash for every $1.00 of net new annual recurring revenue (ARR) it adds. On the scale popularized by investor David Sacks, under 1x is amazing, 1x to 1.5x is great, and 1.5x to 2x is still good. AI startups face a stiffer version of this test than classic software companies, because inference and GPU compute live inside cost of goods sold, so each dollar of growth consumes more cash than it would in a pure SaaS model.

What is a good burn multiple for an AI startup?

A good burn multiple for an AI startup is under 1.5x. The burn multiple measures how much net cash a company burns to add each new dollar of annual recurring revenue, calculated as net burn divided by net new ARR over the same period. A reading under 1x means the business is generating more than a dollar of new ARR for every dollar it burns, which is exceptional at any stage. Between 1x and 1.5x is great, and between 1.5x and 2x is still acceptable for a company that is deliberately spending to grow. For AI-native startups, where inference and GPU compute sit inside cost of goods sold and pull gross margin well below the 70 to 80 percent that classic software typically enjoys, clearing 1.5x is a stronger signal of discipline than the same figure would be for a traditional SaaS company. Anything above 2x deserves a hard look.

How to calculate the burn multiple

The formula is deliberately simple. Burn multiple equals net burn divided by net new ARR. Net burn is the cash a company spent minus the cash it brought in over a period, usually a quarter or a full year. Net new ARR is the increase in annual recurring revenue over that same window. If a startup burned $2 million in a quarter and grew ARR by $1.6 million, its burn multiple is 1.25x, which lands in the great band. If it burned the same $2 million but only added $700,000 of ARR, the burn multiple jumps to roughly 2.9x, which is a warning sign.

The metric is powerful because it collapses growth and efficiency into one number. A company can post fast growth and still look reckless if it is buying that growth at three or four dollars of burn per dollar of revenue. It can also grow slowly yet efficiently and score well. That is why the burn multiple has become a favorite board-level shorthand. It sits alongside the rest of the unit economics every founder should model from day one, and it is one of the cleanest early reads on whether a business compounds or leaks.

The David Sacks burn multiple bands

Investor David Sacks popularized the burn multiple in a 2020 essay and paired it with a scale that founders and investors still use as shorthand. The bands turn a raw ratio into a verdict, which is exactly why they travel so well between a founder and a board. What Sacks is on record for is the formula and those five efficiency bands, not a fixed table of per-stage targets.

Read the number against stage, though, because the same ratio can carry a different verdict at different points. As a rough venture convention, an earlier-stage company still searching for product-market fit is given more tolerance, and the expectation tightens with every round, trending toward zero as the business approaches profitability. So a 1.8x burn multiple is not just a number. It is a company that is officially "good" but still has room to tighten before the next round.

Why AI startups have a harder time hitting a low burn multiple

The burn multiple was defined in a SaaS world where the marginal cost of serving one more customer was close to zero. AI changes that math. Every prompt, every agent run, and every generated token carries a real compute cost, and that cost belongs in cost of goods sold, not in a line item you can wave away. Agentic products make it sharper still, because a single task can trigger many model calls and burn far more tokens than a simple chat response.

The result is that an AI company adding a dollar of ARR often spends materially more to deliver it than a classic software company did, which pushes the burn multiple up before sales and marketing even enter the picture. A rising burn multiple also shortens runway, so the compute bill is not an accounting curiosity, it is a survival variable. Founders who treat inference as a fixed background cost rather than a per-transaction COGS line tend to discover the problem only after the burn multiple has already drifted past 2x, when the fix requires renegotiating pricing or re-architecting the product under pressure.

How a venture studio bends the burn multiple down

This is where the build model matters. At Avante, we co-found AI-native companies for Brazil and Latin America, and we underwrite capital efficiency at the moment of building rather than after the fact. Two levers move the burn multiple in a studio's favor.

The first is the shared cost base. When engineering, design, data, and go-to-market infrastructure are spread across a portfolio instead of rebuilt inside every new company, the net burn side of the ratio starts lower, so the same revenue growth produces a better multiple. The second lever is model routing and inference discipline. Because the studio underwrites cost of goods sold before a product ships, portfolio companies route each workload to the cheapest model that clears the quality bar and measure themselves on cost per successful task rather than raw token spend. That keeps gross margin defensible and the burn multiple honest as the company scales.

A burn multiple under 1x is also what makes it realistic to build an AI-native company without rushing into a Series A, because a business that adds more ARR than it burns controls its own timeline. Capital efficiency stops being a fundraising talking point and becomes the thing that buys optionality.

What a good burn multiple looks like in 2026

Benchmarks have tightened as capital has gotten more selective, and the direction of travel is clear even where precise 2026 medians are still settling. Efficient AI companies at Series A are expected to operate close to 1x, strong ones sit in the 1x to 1.5x band, and the median company is under more pressure than it was in the cheap-money era, because heavier inference COGS eats into the growth-per-dollar it can show. Treat the specific number for your stage as a moving target and the band as the anchor.

The practical test is direction plus level. If your burn multiple is under 1.5x and trending down, you are compounding capital and you can raise on your own schedule. If it is drifting above 2x, the honest question is whether that spend is buying durable growth or quietly subsidizing revenue that does not carry its own weight.

The bottom line

A good burn multiple for an AI startup is under 1.5x, and the best companies push under 1x. The scale is simple, but AI economics make it harder to earn, because inference sits in COGS and every unit of growth costs real compute. That is why the metric rewards discipline built in from the first architecture decision rather than bolted on during a later cost-cutting scramble. Measure net burn against net new ARR every quarter, watch the trend more than any single reading, and treat any drift past 2x as a prompt to interrogate both your pricing and your model spend.

Preguntas frecuentes

What is a good burn multiple for an AI startup?

A good burn multiple for an AI startup is under 1.5x, meaning it burns less than $1.50 of net cash for every $1.00 of net new ARR it adds. On David Sacks' scale, under 1x is amazing, 1x to 1.5x is great, and 1.5x to 2x is good. AI companies have to work harder to reach it because inference and GPU costs sit inside cost of goods sold and depress gross margin, so anything above 2x deserves scrutiny.

How do you calculate a burn multiple?

Divide net burn by net new ARR over the same period. Net burn is cash out minus cash in, and net new ARR is the increase in annual recurring revenue over that window. For example, burning $2 million while adding $1.6 million of ARR gives a burn multiple of 1.25x.

Why is the burn multiple higher for AI startups than for SaaS?

Because inference is a real per-transaction cost that belongs in cost of goods sold. Every prompt, token, and agent run consumes compute, so each dollar of new revenue costs more to deliver than it does in classic SaaS. That pushes the ratio up before sales and marketing spend is even counted.

What burn multiple do investors want to see at Series A?

Investors generally want Series A companies operating near or below 1.5x, with the most capital-efficient closer to 1x. A reading drifting above 2x invites questions about whether the spend is buying durable growth or subsidizing revenue that does not carry its own weight.