Forecasting & Risk · 2 min read
Runway isn't a number — it's a distribution
Your runway is usually reported as a single figure. For an AI-native company, that number is a lie of precision. Here's how to model it honestly.
By Akhil Anand · September 9, 2026
Every founder can recite their runway to the month. "We have fourteen months." It's on the board deck, it's in the investor update, it's the number that helps you sleep. And for a traditional software company, a single number is roughly fine — your costs are mostly fixed, so dividing cash by burn gives you something close to the truth.
For an AI-native company, that single number is a lie of precision.
Why AI burn breaks the simple math
Traditional runway assumes your monthly burn is stable. But AI inference cost is a variable, usage-driven line item. It moves with customer behavior, model mix, context growth, retries, and the raw unpredictability of how people use a product that talks back.
So when you divide cash / burn, you're dividing by a number that isn't stable. You're treating a noisy, fat-tailed process as if it were a constant. The result feels precise — "fourteen months" — but it quietly hides the thing you actually care about: the chance you run out sooner than you think.
The question isn't "what is my runway?" It's "what is the probability my runway is shorter than X?"
Model the distribution, not the point
The honest version of runway is a distribution of outcomes, not a point estimate. You get there by simulating many possible futures instead of assuming one.
At AtlasBurn we run a Monte Carlo over a 365-day horizon — thousands of paths, each one a plausible trajectory of daily burn given your observed volatility, growth, and shock structure (outages, retry cascades, traffic surges). The output isn't a number; it's a shape.
// A single simulated path: daily cost is stochastic, not fixed.
for (let day = 0; day < 365; day++) {
const dailyCost = sampleDailyCost(mu, sigma); // log-normal draw
capital -= dailyCost;
capital += dailyRevenue(day); // MRR grows + churns
if (capital <= 0) { insolventAt = day; break; }
}
Run that ten thousand times and you can read the truths a single number can't show you:
- P50 runway — the median outcome. Your "fourteen months."
- P5 / stress runway — the unlucky-but-plausible tail. Often dramatically shorter.
- Survival probability — the share of futures where you make it to your next milestone.
What this changes
Once runway is a distribution, the decisions change. "Fourteen months" might really be "a median of fourteen, but a one-in-twenty chance of nine." That gap is the difference between raising on your terms and raising in a panic.
Reporting a point estimate isn't wrong because the math is hard. It's wrong because it throws away the one piece of information — the spread — that tells you how much to worry.
Runway is a distribution. Model it like one.