A headline says an AI company expects to reach a very large annual revenue within a few years. The line on the chart climbs like a ramp. It is easy to nod along, yet many AI revenue projections look exactly like that.
Pause there. A projection is a prediction, built from assumptions, by people who often have a reason to want the number to look big. That does not make them dishonest. It does mean AI revenue projections should be read as arguments, not facts.
What AI revenue projections really are
A revenue projection starts with a few inputs: how many customers the company has now, how fast that number grows, how much each customer pays and how long they stay. Multiply these across future months and you get a line.
Change one input slightly and the result moves a lot. If growth is assumed to be eight percent a month instead of six, the end number can change dramatically, because small differences compound. This is why two analysts looking at the same company can produce very different AI revenue projections.
The important point is that the arithmetic is rarely the problem. The assumptions are. When you read a projection, look for the assumptions behind it and ask whether each one is believable. A general picture of how this works is in the article on forecasting, which also explains why long-range predictions are harder than short ones.
Annualized numbers and what they hide
Many AI revenue projections lean on an annualized figure. A company earns a certain amount in a recent month or week, multiplies it to cover a full year, and calls it a run rate.
This can be fair for a stable subscription business. For a fast-changing AI product, it can mislead. A single strong month might include one-time deals, a launch spike, or heavy promotions. Multiplying it by twelve assumes the strength continues, which may not be true.
Usage-based pricing adds another wrinkle. Many AI products charge for what customers use, not for a fixed plan. Usage can jump when a customer runs a big experiment and fall when it ends. A run rate built on a peak month is a measure of a good month, not of a year.
We explain this in more detail in our piece on how to read an annualized revenue number, and the article on what $100M ARR really means shows what such a milestone does and does not tell you.
Questions to ask about AI revenue projections
When you see AI revenue projections, run through a short list.
What counts as revenue? Some companies include signed contracts that have not started, pilots that may not convert, or free credits given to customers. Cash received and revenue recognized are different things.
How concentrated is it? If a large share comes from one or two customers, losing one wipes out the forecast.
How sticky are customers? Retention matters more than sign-ups. If customers try the product and leave, growth is a leaky bucket.
What are the costs? Revenue is not profit. Running large models can be expensive, and an AI company may spend much of every dollar on computing. Strong revenue with thin margins is a different business from strong revenue with healthy ones.
What happens if prices fall? AI services have often become cheaper over time, and competition can push prices down. A forecast that assumes stable prices and rising volume deserves a second look.
Who made the forecast and why? A projection in a fundraising deck serves a different purpose from one in an audited report.
Red flags in AI revenue projections
Some patterns recur across weak forecasts.
A hockey-stick curve with no explanation for the sudden bend. If growth speeds up sharply in year three, ask what changes.
Vague terms. “Expected revenue”, “contracted pipeline” and “potential” can describe very different things. Precise terms, like billed revenue, are more reliable.
Comparisons that pick only the best case. Saying a company is growing faster than a famous giant did at a similar stage ignores the many that did not.
A single big number without a range. Honest forecasts usually offer a base case, a low case and a high case.
Heavy reliance on the market being huge. “If we capture just one percent of a trillion-dollar market” is a story, not a plan.
Reading it like a pro
Treat the number as the end of a chain and inspect each link. Ask for the monthly figures, not just the yearly. Look at the trend in customer counts, churn and average spend. Compare the claim against what the company reports in regulated filings, if it files them.
Also, keep the time frame in mind. A one-year forecast based on signed contracts deserves more trust than a five-year forecast built on market share assumptions.
You do not need to distrust every number. You need to ask what would have to be true for it to come out right. If the answer is a long list of things going well at once, discount the projection accordingly.
Good AI revenue projections are modest about uncertainty, specific about assumptions and open about costs. When a forecast has all three, you can take it more seriously than any headline number, however large.
