In the AI startup world, “traction” has always been a moving target. But in the last couple of years, as generative AI products have shifted from demos to deployments, a particular metric has become both more important and more slippery: annual recurring revenue, or ARR.
ARR is supposed to be a clean shorthand for how much revenue a company can expect to generate each year from customers on an ongoing basis. In classic SaaS, it’s often close enough to reality that investors can use it to benchmark growth, forecast runway, and compare companies with some confidence. In AI—where pricing models can be usage-based, contracts can be structured around pilots, and “recurring” can mean anything from monthly seats to variable consumption—ARR can start to drift away from its original intent.
A new TechCrunch report argues that some AI startups and their venture backers have learned to treat ARR less like a strict accounting measure and more like a narrative instrument. The headline framing is provocative, but the underlying story is more nuanced: the report suggests that inflated or re-framed ARR isn’t necessarily a surprise to investors. Instead, it’s part of a shared understanding of how the market works when traditional revenue definitions don’t map neatly onto AI business models.
What makes this story worth paying attention to isn’t just the possibility of overstatement. It’s the way the metric is being used to “kingmake” outcomes—shaping fundraising momentum, influencing hiring and partner decisions, and affecting which companies get treated as category leaders before their financials fully catch up.
To understand what’s happening, it helps to separate three things that often get blended together in public discussions: revenue, recurring revenue, and forward-looking expectations.
Revenue is what you earned. Recurring revenue is what you expect to keep earning because customers are committed to ongoing usage or subscriptions. Forward-looking expectations are what you believe you’ll earn next quarter or next year based on pipeline, pilots, and contract negotiations.
In AI, those lines blur quickly. A company might sign a contract that looks like recurring revenue on paper but is actually contingent on performance, adoption, or continued model usage. Another company might have customers who are “committed” to a pilot for six months, but the pilot is really a procurement process with uncertain conversion. A third might sell enterprise access where the customer pays monthly, but the amount fluctuates dramatically based on usage patterns that are still stabilizing.
None of these situations are inherently deceptive. They’re common in early-stage AI commercialization. The problem arises when the metric used to describe traction becomes detached from the economic reality it’s meant to represent—and when the detachment is large enough to change how outsiders interpret the company’s trajectory.
The report’s core claim is that some founders and VCs have been willing to stretch ARR definitions or present ARR in ways that make the company look healthier than a strict reading would suggest. And crucially, it implies that investors are not operating under naive assumptions. They may know that the ARR being discussed is not perfectly comparable across companies, and they may accept that as long as the story is coherent and the underlying business is progressing.
That’s a key point: the report isn’t simply about “lying.” It’s about incentives and interpretation. When everyone understands that a metric is imperfect, the metric can still function as a coordination tool—helping the market decide who is winning, who deserves capital, and who should be taken seriously.
So how does ARR become a flexible narrative tool?
One mechanism is timing. In many AI deals, revenue recognition and cash collection don’t align neatly with the moment a contract is signed. A company might secure a multi-month agreement that will generate meaningful revenue over time, but the “annualized” figure can be presented immediately as if it were already stable recurring revenue. Annualizing a contract value is not automatically wrong—many businesses do it—but the difference between “annualized contract value” and “run-rate ARR” matters. If the contract is short, conditional, or likely to change, the annualized number can overstate what the company can reliably sustain.
Another mechanism is contract structure. AI companies often sell bundles that include model access, integration work, support, and sometimes professional services. Depending on how the contract is written, parts of the deal may be recognized as revenue differently. If a company emphasizes the portion that behaves like recurring subscription while downplaying components that are one-time or variable, the ARR story can become more flattering than the full economics.
A third mechanism is usage-based pricing. Usage-based contracts can be recurring in the sense that customers keep paying as they use the product. But usage-based revenue is inherently volatile early on. Customers ramp up slowly, internal adoption takes time, and usage can spike or drop depending on how the product is deployed. If a company reports ARR based on current usage without acknowledging that usage is still in a ramp phase, the ARR can look more stable than it truly is.
Then there’s the question of what counts as “recurring” at all. In classic SaaS, recurring usually means renewals and subscriptions that continue unless a customer churns. In AI, “recurrence” can be tied to ongoing experimentation. A customer might keep paying for a pilot because it’s useful, but the pilot could end once procurement cycles complete or once the customer decides whether to scale. If a company treats pilot commitments as recurring ARR, it may be describing a future that is plausible but not guaranteed.
The report’s most interesting angle is that investors reportedly aren’t being blindsided by these differences. That suggests a market where the metric is understood to be approximate, and where the real question is whether the approximation is “good enough” to justify valuation and capital allocation.
This is where the concept of kingmaking comes in. In venture capital, the market doesn’t just fund companies based on fundamentals; it also funds companies based on perceived momentum. Momentum attracts talent, partners, and additional investors. It creates a feedback loop: the more credible the story, the easier it is to raise the next round, which funds the next phase of product and sales, which then makes the story more credible.
ARR is one of the most visible signals of momentum. Even when sophisticated investors know the metric is imperfect, they still use it as a shorthand. If a company’s ARR narrative is strong, it can become a proxy for execution quality. If the narrative is weak, it can become a reason to pass—even if the underlying business is improving.
In other words, ARR can influence outcomes beyond the company’s own financial statements. It can affect who gets hired, which candidates choose to join, which partners decide to integrate, and which customers feel comfortable committing. For AI startups trying to win enterprise deals, credibility matters. A clean ARR story can reduce perceived risk for buyers who are evaluating whether the vendor will survive and scale.
But credibility built on a stretched metric has consequences too.
For one, it can distort benchmarking. If Company A reports ARR that includes annualized pilot commitments and usage-based estimates, while Company B reports ARR based on contracted renewals with stable consumption, comparing them directly can lead to incorrect conclusions. Investors may adjust for this internally, but the market at large often doesn’t. Public messaging, press coverage, and even recruiting materials can propagate the simplified version of the metric.
Second, it can create pressure to maintain the narrative rather than the economics. If ARR is the headline number, teams may optimize for actions that improve the ARR figure in the short term—signing contracts that can be annualized, structuring deals to look recurring, or emphasizing certain revenue components—without ensuring that the revenue is durable. That doesn’t mean every founder is gaming the system. It means the system rewards the appearance of durability.
Third, it can complicate diligence. When ARR is used as a primary indicator, diligence teams must spend more time unpacking definitions, contract terms, and customer behavior. That increases friction and cost. It also increases the chance that different investors apply different levels of skepticism. Two funds might look at the same ARR number and reach different conclusions depending on how they interpret the underlying contracts.
The report’s implication that investors are “fully aware” points to a fourth consequence: the market may be normalizing a level of ambiguity that would be unacceptable in other contexts.
In traditional SaaS, ARR is often treated as a relatively standardized metric. There are still differences in calculation methods, but the industry has developed conventions around what counts as recurring, how to handle expansions, and how to treat one-time fees. AI disrupts those conventions. But if the industry responds by loosening definitions without clear disclosure, the metric becomes less informative.
And when a metric becomes less informative, it becomes easier to use it strategically.
There’s also a deeper issue: AI companies often operate in a world where “product-market fit” can look like “distribution-market fit.” Many AI startups can demonstrate value quickly in a pilot. The hard part is scaling adoption across departments, integrating into workflows, and sustaining usage. That means early revenue can be real but not yet stable. In that environment, ARR can be both a useful signal and a misleading one.
The unique take here is to recognize that the tension isn’t only about ethics. It’s about measurement under uncertainty.
AI startups are selling something that evolves. Model capabilities improve, costs change, and customer expectations shift. A contract signed today might be priced based on assumptions that won’t hold after a model update or after the customer changes its deployment strategy. In such a dynamic environment, any single metric will be imperfect. The question is whether the imperfections are acknowledged transparently—or whether they’re smoothed over to create a cleaner story.
Transparency is the missing ingredient in many of these debates. If a company clearly labels its ARR as run-rate, annualized contract value, or includes specific components like usage-based estimates and pilot commitments, the market can adjust. If it presents ARR as if it were equivalent to classic SaaS recurring revenue without qualification, the metric becomes a lever.
The report suggests that some companies and investors are comfortable with that lever because the market is already trained to interpret ARR flexibly. That comfort may come from