The most valuable metric in Silicon Valley has always been growth, but in the AI gold rush of 2026, even growth itself has become negotiable. A growing number of AI startups are presenting investors with revenue figures that bear only a passing resemblance to the Generally Accepted Accounting Principles that govern public markets—and venture capitalists, hungry for the next foundational model company, are choosing not to ask too many questions.
The practice centers on Annual Recurring Revenue, or ARR, the north star metric for software businesses. In its traditional form, ARR represents contracted, repeatable revenue from customers who have committed to ongoing subscriptions. It is predictable, defensible, and the basis upon which enterprise software valuations have been built for two decades. But in the current AI landscape, founders are stretching the definition to include one-time compute credits, pilot programs with no renewal commitment, and even projected revenue from customers still in negotiation.
The mechanics of metric inflation
The sleight of hand works because AI companies have genuinely novel business models that do not map cleanly onto traditional SaaS economics. A customer might pay a large upfront sum for model fine-tuning, then consume inference on a variable basis. Is that recurring? It depends on who is asking. When the audience is a Series B investor competing with three other term sheets, the answer tends to be yes.
Some firms have begun distinguishing between "ARR" and "contracted ARR," but the distinction often gets lost in pitch decks and press releases. The result is a landscape where companies claiming comparable revenue figures may have wildly different underlying economics. One firm's ARR might represent enterprise contracts with Fortune 500 companies; another's might be largely composed of free credits that converted to paid plans for a single month.
Why investors tolerate the ambiguity
The venture capital industry is not naive about these practices. Partners at top-tier firms privately acknowledge that they apply significant haircuts to reported ARR figures when conducting due diligence. But the competitive dynamics of AI investing create powerful incentives to play along publicly. A fund that refuses to validate inflated metrics risks losing deals to competitors who will. And in a market where the winners may capture extraordinary value, the cost of missing a generational company outweighs the cost of overpaying for a few that do not pan out.
This tolerance has historical precedent. The late 2010s saw similar flexibility around metrics like Monthly Active Users and Gross Merchandise Value, categories that allowed WeWork and others to tell growth stories that eventually collapsed under scrutiny. The difference now is that AI companies are approaching public markets faster, and the gap between private valuations and public market discipline is narrowing.
Our take
Metric inflation is not fraud—it is marketing. But the distinction matters less than the outcome. When AI startups eventually face the cold scrutiny of public markets or down-round investors, the companies with real recurring revenue will be separated from those with creative accounting. The founders who resist the temptation to inflate will look conservative today and prescient tomorrow. The venture capitalists enabling the current game are making a calculated bet that they will exit before the music stops. Some of them are right. Most of them are not.
