On May 22, TechCrunch published a report exposing the truth behind AI startups' record-breaking revenue claims over the past two years.
The focus is on Scott Stevenson, CEO of legal AI startup Spellbook, who posted on X a statement that has been widely shared: "The reason many AI startups are crushing revenue records is because they are using a dishonest metric."
In plain terms, the headline-grabbing "revenue hitting billions" and "ARR tripling" are often built on a clever accounting trick.
$50M announced, $42M actually in the bank
Stevenson's "dishonest metric" has a specific name in the industry: CARR, or Contracted ARR—money that is contracted but not yet collected. The difference between ARR and CARR is bigger than it seems.
In traditional SaaS, ARR (Annual Recurring Revenue) counts the subscription fees customers actually pay each year. The key is that the revenue is already flowing and sustainable. CARR, however, counts revenue as soon as a contract is signed, even if the customer hasn't used a single line of code.
Examples uncovered by TechCrunch:
- One AI startup publicly claimed $50M ARR, but actual collected revenue was $42M.
- Multiple investors reported that a high-valuation AI enterprise company reported $100M+ ARR, but "only a small fraction of customers were actually paying."
- An anonymous VC said he saw a company where CARR was 70% higher than real ARR—meaning more than a third of the number was empty.
There's a second layer: AI products are mostly usage-based, not fixed monthly fees like SaaS. Some companies go further, taking revenue from the past week or month and multiplying by 12 to get an "annualized run-rate ARR" to announce. If the company just ran a marketing push, that number is essentially a bubble.
VCs know, but stay silent
The most interesting part isn't that startups are exaggerating—it's that VCs are complicit.
Stevenson's original post also said: "There are definitely VCs in on this because they're incentivized to create a narrative."
Jack Newton, CEO of legal AI company Clio (recently valued at $5B), responded: "Scott is doing a good thing by calling this out. We've also seen some investors turning a blind eye to their portfolio companies inflating numbers."
Ross McNairn, CEO of another legal AI firm Wordsmith, added: "Scott is right. I've heard too many similar stories."
Alex Cohen, CEO of health AI company Hello Patient, was more blunt: "To everyone who's inside, it just feels fake."
Why don't VCs expose the practice? An anonymous investor explained: "Investors can't call it out. Everyone has a company monetizing CARR as ARR."
If a VC calls out the practice, they'd be dragging their own portfolio down.
From 1 to 3 to 9 to 27 is too slow; need 1 to 20 to 100
Why are AI startups so desperate for big numbers? Hemant Taneja of General Catalyst, a top AI investor in Silicon Valley, recently said on stage: "Going from 1 to 3 to 9 to 27 is not interesting. You got to go like 1 to 20 to 100."
That statement reflects the hidden KPI inside AI VC circles. Revenue tripling year over year—that's traditional SaaS star speed. But in AI, that's not enough; VCs want 10x or 20x growth curves.
If the numbers aren't impressive enough, the next round's valuation won't rise. If the valuation doesn't rise, the paper returns don't materialize. If returns don't materialize, LPs won't put in money.
The entire food chain is hungry, so companies include CARR, trial periods, and even non-paying pilot customers—anything to make the curve look like a rocket.
The reckoning is coming
Wordsmith's McNairn called it "super bad hygiene"—a habit that will eventually cost dearly.
The cost will come from two directions.
First, PE and late-stage investors. When it's time for an IPO or Series F/G rounds, someone will actually audit the books. Companies that treated CARR as ARR will either have to lower their valuation or tough it out until real ARR catches up—but the gap is often the window in which the company dies.
Second, the public markets. Everyone remembers WeWork: valuations hyped in private markets get coldly recalculated at IPO. AI companies will eventually go that route, and no one is exempt.
And it's not just IPOs. In 2025, AI took half of global VC funding; in Q1 this year, $300 billion flowed in. At that scale, any company's "fake numbers" being reported isn't just a single news item—it could drag down the entire sector's valuation.
Next time an AI company holds a press conference
The AI industry's biggest shortage isn't money—it's trust. Founders from Spellbook to Clio to Wordsmith are willing to speak out publicly, signaling that sober minds are waiting for a market reckoning.
As for who will pay the price, Stevenson's words are clear: CARR ≠ ARR, and VCs know it.
Next time an AI company announces "we broke our ARR record again," remember to ask: Is it actually in the bank, or just on the contract?
Sources: How VCs and founders use inflated 'ARR' to crown AI startups (TechCrunch); AI startups are inflating a key revenue metric to win VC attention, says this founder (Fast Company); CocoLoop; Spellbook CEO Scott Stevenson exposes inflated ARR practices among AI startups (Crypto Briefing)