
Why ARR No Longer Tells the Whole Story for AI Startups
October's OpenAI revenue shock revived doubts about ARR. RTP Global's Kenza Zayani sets out what investors will ask instead: committed revenue, churn and cohort retention.
- By
- Tom Whitfield
- Filed
- Channel
- Startups & Funding
- Read
- 3 min read
Reports in October that OpenAI's annualized revenue was running well below earlier estimates sent AI-linked stocks sharply lower and reopened a question investors have circled for months: is ARR still the right way to describe an AI business? Kenza Zayani, an early-stage investor at RTP Global in London, argues in a new analysis that the metric is wearing thin, because most AI companies no longer earn money the way ARR assumes.
The episode mattered beyond one company. It showed, Zayani writes, how much weight public markets now place on a single private company's numbers. A cohort of AI startups is also reaching the growth-capital stage, and the questions they face in diligence are changing. Founders who understand this now hold an advantage over those who wait.
What was ARR built for?
ARR became standard in the mid- to late-2000s, when software shifted from licence fees to recurring subscriptions. A customer signed for 12 months, revenue became predictable, and annualizing it was simple arithmetic rather than forecasting.
The first generation of post-ChatGPT companies inherited the metric without much examination, despite earning money differently. Most now run hybrid models: a subscription base plus consumption charges billed against API calls, tokens processed or compute used. A growing number charge per outcome. An annualized figure therefore blends revenue a customer has committed to with revenue that depends on what they use next month.
Zayani's core point is mechanical. What gets presented as ARR is usually a revenue run rate: recent revenue multiplied out to an annual figure. That describes momentum reasonably well, but it is a less reliable guide to the year ahead. The first question an investor tends to ask is what the figure actually counts.
Partner sales add another wrinkle. Companies selling partly through partners can report either the full value of those sales or only the share they keep. Both conventions are legitimate, but the two are not comparable, and the same business can produce very different headline numbers.
Why two identical numbers can hide different businesses
Zayani offers a thought experiment. Take two companies each reporting a $10 million annualized run rate. In the first, most of that revenue is contracted, and comes from customers whose usage has expanded steadily over 12 months. In the second, the number rests on a handful of customers whose consumption spiked in the last quarter.
These are not similar businesses. The first has demonstrated that customers keep coming back and spend more over time. Nothing in the headline figure reveals that difference. The takeaway: a single annualized number cannot separate a durable business from a lucky quarter.
What should investors measure instead?
The answer depends on stage. At pre-seed and seed, the number still matters. A steep trajectory can signal market demand, customers willing to pay, and a team beginning to crack go-to-market. Nobody expects a run rate at that size to predict the year.
Later, investors need more. What matters is the revenue behind the number:
- customer concentration;
- how much revenue is contracted rather than consumed;
- how predictable that revenue is;
- whether gross margins hold as volume rises.
Retention and expansion are the centerpiece, Zayani writes. They show how essential a product has become and whether it will still be in use in a few years. Churn is diagnostic in a way a growth figure never is: it can point to demand the product fails to answer, pricing that does not match the value, a need that was never truly recurring, or a market shallower than it first appeared.
What changes next?
ARR has anchored how high-growth tech companies demonstrate momentum for so long that it is unlikely to disappear soon. What is changing is how much weight the single number can carry on its own.
Zayani expects more investors to ask what a figure actually counts, how much of it is committed revenue, and what retention looks like by cohort. Founders who arrive with those answers ready, she concludes, will tell a far more convincing story than a single number ever could.
Source: news.crunchbase.com
More from Tom Whitfield
Show full bio
Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,063 articles
Related articles
openai-s-reported-q3-revenue-rose-50-to-about-10-billion
OpenAI's reported Q3 revenue rose 50% to about $10 billion
gartner-puts-genai-model-spend-at-28-3-billion-c7818787
Gartner Puts GenAI Model Spend at $28.3 Billion for 2026
us-business-ai-spending-hit-record-up-just-1-65
US Business AI Spending Hit Record in October, Up Just 1.65%


