AI startups are racing to show extraordinary revenue growth, but the number being used to crown winners is becoming less reliable just as valuations depend on it most.
The AI funding market has found its favorite scoreboard, and it is Annual Recurring Revenue. A startup that can say it has reached $20 million, $50 million, or $100 million in ARR is no longer just reporting traction. It is making a claim about durability, category leadership, and the valuation it deserves.
The problem is that many of those claims are now being stretched. According to TechCrunch, founders, investors, and startup finance professionals say some AI companies are presenting committed contracts, unpaid pilots, usage spikes, and annualized run rates as if they were traditional recurring software revenue. That may help a company look larger in a headline, but it also makes the business harder to judge.
For founders raising capital, this is not just a debate over accounting language. It changes the market around them. If one company in a category calls contracted but not yet live revenue ARR, competitors feel pressure to do the same. If investors reward the biggest number, the cleanest number starts to look weak. That is how a metric built to create clarity becomes a tool for narrative building.
Traditional SaaS ARR had a simple purpose. It gave investors a way to understand the annualized value of recurring subscription revenue from customers who had already committed to paying. It was never perfect, but it was useful because it pointed to predictable revenue.
AI has made that harder. Many AI products are not sold only as seat-based subscriptions. They use consumption pricing, outcome pricing, API usage, enterprise pilots, implementation-heavy contracts, or a mixture of all of them. Those models may be perfectly legitimate. The issue is whether the resulting revenue behaves like recurring revenue.
Carta's startup guidance says ARR should focus on subscription-based revenue and should leave out one-time fees, short-term revenue, refunds, taxes, and other items that do not recur. It also warns founders not to confuse bookings with revenue, because a customer's promise to pay is not the same as cash that has actually come in.
That distinction matters more in AI because many deployments take time. A large enterprise may sign a contract but not fully roll out the product for months. A customer may test a tool in a free or discounted pilot and decide not to continue. A usage-based customer may spend heavily during one period, then reduce activity the next. Annualizing those moments can produce an impressive number that does not survive contact with renewal behavior.
TechCrunch reported that one investor had seen companies where committed ARR was 70 percent higher than actual ARR. It also cited an example where marketing materials showed $50 million in ARR while the real figure available to investors was $42 million. Those gaps are not always fraud. Sometimes investors have the accurate books. But employees, customers, and later investors may only see the cleaner story.
Valuations Depend On The Definition
The incentive is obvious. If a startup raises at a $1 billion valuation on $20 million of ARR, the deal looks like 50 times ARR. If the real durable revenue is closer to $10 million, the same valuation is 100 times ARR. One number can make a round look aggressive. The other can make it look reckless.
This is why the ARR debate is now tied directly to venture returns. AI companies are being expected to grow faster than earlier software businesses. Jennifer Li of Andreessen Horowitz told TechCrunch earlier this year that not all ARR is equal and warned founders to be skeptical of spectacular revenue claims posted online. Her point was not that fast growth is fake. It was that retention, business quality, and durability still matter.
That is the part the market keeps trying to compress. A founder can raise money on a sharp revenue curve before there is enough evidence that customers will renew, expand, and absorb higher prices. A fund can mark up a portfolio company when the outside world believes the company is running away with the category. Talent follows the apparent winner. Customers follow it too. In venture, perception can become a temporary advantage.
But temporary advantage is not the same as company value. If usage slows, pilots fail to convert, or customers churn after a discounted contract period, the next round has to face the real revenue base. That is when inflated ARR becomes a valuation problem, not a communications problem.
The Market Needs Cleaner Revenue Language
There is no single accounting body solving this for private AI startups right now. ARR is not a GAAP metric, and public-company disclosure rules do not apply cleanly to private fundraising decks, founder posts, or press announcements. That leaves the burden on boards, investors, CFOs, and founders to define the number clearly.
A practical standard is already visible. Companies should separate live ARR from committed ARR, usage-based run rate, bookings, pilots, and remaining contract value. They should disclose whether usage revenue is based on the last month, last quarter, or a longer period. They should show churn, downsell, renewal rate, and gross margin beside the topline number.
Limited partners will care because they ultimately fund the funds that are marking these companies. If portfolio value rests on ARR, LPs need to know what kind of ARR is being used. A fund reporting exposure to fast-growing AI software looks very different if much of that growth is contracted but not live, or based on customer usage that can fall quickly.
None of this means AI revenue is weak. Some AI companies are growing at speeds that older software markets rarely saw. But the stronger the business, the less it should need a blurry metric to explain itself. The next phase of AI investing will not be about who can announce the biggest ARR number. It will be about who can prove the revenue renews, expands, and turns into cash.
Also read: Tencent's L2P makes pixel-space image generation practical again • Apple's $4.5 trillion valuation raises the bar for AI founders • Micron brings advanced U.S. memory production into the AI supply race
