ARR vs Annualized Run Rate: What Each Means and Why It Matters
ARR (annual recurring revenue) counts only revenue that repeats under contract, such as subscriptions, annualized. An annualized run rate takes recent revenue of any kind, usually the latest month or quarter, and multiplies it up to a full year. ARR is a statement about contracts; run rate is arithmetic on a recent period. The two numbers can match, but they often do not.
The gap matters most for AI companies, where revenue frequently depends on how much customers use a product rather than on a fixed seat fee. That is why headlines about AI revenue so often quote a run rate and not ARR.
What ARR means
ARR is the yearly value of recurring revenue a company expects to keep receiving: subscriptions, service contracts and multi-year commercial agreements. By the usual definition it excludes one-time and non-recurring payments such as implementation fees, professional services and hardware.
The simplest version is monthly recurring revenue (MRR) times 12. If 400 customers each pay a $250 monthly subscription, MRR is $100,000 and ARR is $1.2 million. Annual contracts are counted at their yearly value, and multi-year deals are normally counted per year, not as a lump sum.
ARR is only as solid as the contracts behind it. A customer can cancel, downgrade or stop paying, and ARR does not capture that risk by itself.
What an annualized run rate means
A run rate takes a short, recent slice of revenue and projects it forward as if nothing changes. The common forms are:
- Monthly: last month’s revenue × 12
- Weekly: last week’s revenue × 52
- Quarterly: last quarter’s revenue × 4
Nothing in the formula asks whether the revenue is recurring, contracted or repeatable. A run rate built on one month of revenue includes everything booked in that month: subscriptions, overage charges, one-off projects and metered usage. It is a snapshot of pace, not a promise.
ARR vs annualized run rate: the differences
- Basis. ARR is built from contracts and subscriptions. Run rate is built from recent revenue actually recognized or billed.
- Treatment of one-offs. ARR strips them out. Run rate keeps them unless the company adjusts.
- Usage-based revenue. ARR has no clean place for it, since the amount changes every month. Run rate captures it directly.
- Forward-looking strength. ARR implies commitment. Run rate implies only that the recent month could repeat.
- Standardization. Neither is a measure under US GAAP or IFRS, so each company chooses its own definition. Two firms quoting “$100 million” may mean different things.
Why AI companies quote run rate
Many AI products are priced by the token, by the API call, by the task or by the hour of compute used. A customer who spends $5,000 one month may spend $50,000 the next. That revenue is real but not contractually recurring, so a strict ARR figure would either leave it out or understate the business.
Run rate fits that pricing better, and it also reflects rapid growth. When revenue doubles every few months, a figure based on the latest month looks much larger than one based on the trailing twelve. Companies in that position have a clear incentive to lead with the more recent number. That does not make it dishonest, but it does mean the reader should check what period it was drawn from.
The same question applies to products built on other firms’ models. If you are comparing revenue claims from application-layer startups, our explainer on what an AI wrapper is covers why margins and defensibility vary, and why headline revenue can say little about either.
A worked example
The numbers below are invented to show the arithmetic.
A startup sells an AI assistant with a $100 monthly seat price and also bills usage on top.
- It has 500 paying seats: 500 × $100 = $50,000 of subscription revenue per month.
- Usage charges in the latest month came to $30,000, including $12,000 from a one-time spike during a customer pilot.
- It also earned $8,000 from a one-off integration project.
The figures then work out as follows:
- ARR from subscriptions only: $50,000 × 12 = $600,000.
- Run rate, all revenue in the month: ($50,000 + $30,000 + $8,000) × 12 = $1,056,000.
- Run rate, excluding the one-off project and pilot spike: ($50,000 + $18,000) × 12 = $816,000.
Same company, same month, and the figure ranges from $600,000 to just over $1 million depending on definition. This is why a bare “$1 million run rate” is much less informative than the sentence that explains what is inside it.
Common pitfalls when reading a run-rate claim
- Pilots and trials. Paid pilots can inflate a month and then not convert to contracts.
- Seasonality. A retailer’s December multiplied by 12 says very little. Education, tax and travel products have the same problem.
- Gross vs net revenue. If a company resells a service and books the full amount instead of only its margin, run rate can overstate the business. For AI products, this can happen when a startup passes through model provider costs.
- Reseller and partner revenue. Revenue routed through a cloud marketplace or channel partner may be booked differently by each side.
- Churn. A run rate does not subtract customers who have given notice but have not yet left.
- Peaks. Using the best week or best month of the period makes the number look better than the typical one.
How investors check the number
Investors rarely accept a single headline figure. Typical follow-up checks include:
- Net revenue retention (NRR): how much of last year’s revenue from a group of customers remains this year after churn, downgrades and expansion. Above 100% means existing customers are spending more in total.
- Gross margin: especially for AI products with large inference bills.
- Burn multiple: net cash burned divided by net new ARR in the same period, a rough read on how efficiently growth is being bought.
- Revenue concentration: how much comes from the top few customers.
- Billing data and contracts: in diligence, the number is tied back to invoices, not slides.
Revenue claims also travel with valuations. Our guide to OpenAI’s valuation history shows how revenue claims are used alongside valuations, and the AI Business section tracks how AI companies describe their revenue as they raise money.
How to describe revenue honestly
If you are a founder preparing an investor update or a press statement, state the period and the formula. For example, “annualized run rate of $X based on revenue in September 2026, including usage fees and excluding one-time services.” Keep ARR for contracted recurring revenue, and report run rate separately. Readers on the other side can apply the same test: ask which month, which revenue and what was left out. More startup finance coverage sits in the Funding category.
Frequently asked questions
Is annualized run rate the same as ARR?
No. ARR is based on recurring contract revenue, while annualized run rate multiplies recent revenue of any type by the number of periods in a year. They can coincide for a pure subscription business with no one-off income.
How do you calculate annualized run rate?
Take revenue for a recent period and scale it to a year: monthly revenue × 12, weekly × 52 or quarterly × 4. The result assumes the period is typical, which is often not true.
Is run rate revenue reliable?
It is reliable only as a measure of current pace. It ignores seasonality, churn and one-time income unless the company adjusts for them, so it should be read together with retention and the stated period.
Why do AI startups report run rate instead of ARR?
Much of their revenue comes from usage-based pricing that varies month to month, which does not fit a strict ARR definition. Run rate also reflects fast growth sooner than a trailing figure would.
Can a company have ARR higher than its run rate?
Yes, depending on the company’s definition. If it counts signed annual contracts that have not started billing yet, or if revenue in the latest month was unusually low, ARR can exceed a run rate built on that month.