The venture capital industry has spent the past three years insisting it learned hard lessons from the 2021 bubble. Valuations would be disciplined. Due diligence would be rigorous. Unit economics would matter. Then along came generative AI, and the industry promptly forgot everything.
The latest symptom: a quiet epidemic of inflated "annual recurring revenue" figures at AI startups, where the definition of "recurring" has become so elastic it now encompasses pilot programs that may never renew, one-time enterprise contracts, and in some cases revenue that hasn't actually been collected yet. Founders and their investors are playing a definitional shell game, and everyone involved seems content to keep the music playing.
The ARR illusion
ARR became the north-star metric of SaaS investing because it promised predictability—a customer paying $10,000 per month represented $120,000 in ARR, and if retention rates held, that revenue would compound. The genius of the metric was its simplicity and its forward-looking nature.
But AI startups operate differently. Many sell large proof-of-concept engagements to enterprises testing whether generative AI can actually deliver on its promises. These deals are often structured as six-month pilots with no contractual renewal obligation. Some startups are counting the annualized value of these pilots as ARR anyway, reasoning that the customer could renew. Others are including professional services revenue—custom model fine-tuning, integration work—that by definition cannot recur in the same form.
The result is a landscape where a startup claiming $20 million in ARR might have $5 million in genuinely recurring contracts and $15 million in creative accounting.
Why investors play along
The cynical explanation is that VCs are complicit because inflated ARR justifies inflated valuations, which justifies larger fund sizes, which justifies larger management fees. There's truth to this, but the fuller picture is more interesting.
Many investors believe they're buying optionality on transformative technology. If an AI startup's product genuinely works, today's pilot customers will become tomorrow's enterprise contracts. The ARR figure isn't a lie so much as a forecast dressed up as a fact. Investors are betting that the gap between reported and real ARR will close—that the fiction will become true before anyone has to mark down the position.
This works until it doesn't. The 2021 cohort of fintech and e-commerce startups that played similar games eventually faced a reckoning when growth slowed and the one-time revenue didn't repeat. The AI sector is now building the same trap, with the added complexity that many customers are themselves uncertain whether AI tools deliver lasting value.
Our take
Metrics exist to illuminate reality, not to obscure it. When founders and investors conspire to redefine ARR until it means whatever is convenient, they're not being aggressive—they're being dishonest, and they're setting up employees, later-stage investors, and eventually public-market shareholders to absorb the consequences. The AI boom is real, but the revenue supporting many of its most celebrated companies is substantially less real than the pitch decks suggest. Someone will eventually have to reconcile the spreadsheet with the bank account.
