Inside the sketchy world of ARR and inflated AI startup accounting

Beginning in 2024, a stream of ‘holy shit’ growth metrics from VC-backed startups began to pop up on X [formerly Twitter]. In less than three years, Midjourney’s ARR went from zero to $200 million. In 20 months, ElevenLabs, a voice AI startup, saw its ARR soar from zero to near $100 million. In three months, vibe coding darling Lovable went from zero to $17 million in ARR, this summer hitting $100 million in ARR. In its first six months, Decagon hit “seven figures” in ARR, the company reported. The most famous example: AI coding tool Cursor went from nada to $100 million in ARR in a year. But who needs a year, anyway? Two VCs Fortune spoke to highlighted the claim made by Andreessen Horowitz-backed AI “cheat on everything” tool Cluely, which claimed over the summer to have doubled ARR to $7 million over a week. 

“There is all this pressure from companies like Decagon, Cursor, and Cognition that are just crushing it,” said one VC. “There’s so much pressure to be the company that went from zero to 100 million in X days.”

All the examples have one thing in common: ARR, or “annual recurring revenue.” The metric came to be a favorite of VCs and startups through the software-as-a-service (SaaS) wave starting in the 2000s, when it was widely accepted as a trusted proxy for a stable startup, with a reliable source of revenue and a reasonably shored up future. 

But as billions flowed across the venture capital ecosystem into AI startups, some mere months old, the vaunted, trusted ARR metric has morphed into something much harder to recognize. There’s now a massive amount of pressure on AI-focused founders, at earlier stages than ever before: If you’re not generating revenue immediately, what are you even doing? Founders—in an effort to keep up with the Joneses—are counting all sorts of things as “long-term revenue” that are, to be blunt, nothing your Accounting 101 professor would recognize as legitimate.

Exacerbating the pressure is the fact that more VCs than ever are trying to funnel capital into possible winners, at a time where there’s no certainty about what evaluating success or traction even looks like. Throughout the 90s, VC as an industry grew to more than 700 firms managing about $143 billion. Today, there are more than 3,000 VC firms according to the National Venture Capital Association, managing more than $360 billion, with some projections suggesting venture will be a more than $700 billion industry by 2029. 

Creative accounting, of course, has a long history of cropping up during a boom, a tradition dating far back, to the Gilded Age when inflating assets, understating liabilities, and bribery were commonplace. More recently the dotcom boom and the leadup to the Great Recession brought to light such practices as “channel stuffing,” “roundtripping revenue,” and who can forget “special purpose entities.” Now industry watchers are starting to raise red flags about ARR. “The problem is that so much of this is essentially vibe revenue,” one VC said. “It’s not Google signing a data center contract. That’s real shit. Some startup that’s using your product temporarily? That’s really not revenue.” Or rather there is revenue (the first ‘R’) but it’s not recurring revenue (the second ‘R’).

One example of what this looks like: A top-firm VC described an early-stage defense tech startup he was looking at, where the founder claimed $325,000 in ARR. “The first time he said it, he didn’t even make a big deal out of it,” the VC said. “He was like: ‘Oh, by the way, we have a contract with this company and it’s worth this much.’ In the second meeting, I said, ‘Wait, let’s go back to that customer, that big contract. How did that deal happen?’ A very common question. He said: ‘Oh, it was super easy. It was a two-week pilot. And we have it on good authority that they’re going to keep paying us that much.’”

Record scratch: “I was like: What does that mean?” the VC said. “Hold the phone, man. The good authority I subscribe to is a signed piece of paper.” 

How ARR became the favored metric in AI

A number is almost never just a number in tech. And it never has been. 

Behind every revenue figure you’ve ever seen, especially when talking about privately-held tech startups, there is a little bit of science—and a lot of art. These companies aren’t monitored the way that public companies are, reporting to the Securities and Exchange Commission quarterly. Investors also don’t necessarily audit the companies they invest in, either. A financial due diligence process, before a VC invests, may involve an informal audit, but more likely it’s a game played with trust. 

And in the SaaS era, technically starting in the 1990s and gaining steam through the 2000s, trust in ARR came comparatively easy. There was an agreed upon set of conventions. For example, annual per-seat pricing was standard, where one user pays for one year and then accounts expanded by adding multiple users. And there was clear separation between ARR, CAR (signed contract value before activation) and recognized revenue (actual revenue booked). Typically 80 to 90% of CAR would convert to ARR, and you could predictably chart a company’s expansion, relying on low churn rates and steady customers.

There were, in short, standardized methods of calculating ARR. 

“We did settle on these terms that everyone agreed on in the SaaS world,” said Anna Barber, partner at VC firm M13. “It was a lot harder to fudge, because people had a general understanding of what things had to mean. Today, we don’t know what things have to mean in the same way. So, there’s a lot of confusion and, maybe, obfuscation.”

Now, here’s the wrinkle: The SaaS era wasn’t a halcyon time of absolute revenue clarity either. As the cloud wave started to take shape, ARR started to get a little funkier. Especially for consumer-facing companies (like restaurant software company Toast) there were questions about whether subscription revenue was a good proxy for ARR. But it was the emergence of AI that created a whole new layer of uncertainty. 

“Investors wanted to keep evaluating companies as SaaS-predictable, so they tried to shoehorn those elements into ‘recurring’ revenue,” said Nnamdi Okike, managing partner and co-founder at 645 Ventures. “It doesn’t truly work, but it worked well enough for investors to keep doing it. Now AI has shown up with a whole new set of elements, and it would be better for investors to finally create new metrics to represent this new reality.”

ARR is in what could be described as an awkward phase, where there are some AI startups that are trying to use the metric with sincerity, but their business dynamics are just too different from traditional SaaS. Prospective customers are still in an experimentation phase, trying all sorts of products on short-term pilots, creating high churn risk. And AI services have unpredictable token usage, which refers to the amount of text that AI processes to understand language. (More tokens equals more usage, and more complicated queries require more token usage, by extension.) So, a few “inference whales” like OpenAI and Anthropic have massive pricing power and can skew costs, making AI startups’ financial structures fundamentally different from traditional SaaS businesses. 

“The classic SaaS model is dying as we speak,” said Priya Saiprasad, general partner at Touring Capital. “We shouldn’t be using classic SaaS terms to measure these companies, we shouldn’t be using the language of it. So we should all, collectively as an industry, evolve to a new set of metrics we feel comfortable measuring these companies by.” 

The result? Founders are counting pilots, one-time deals, or unactivated contracts as recurring revenue, six VCs told Fortune. And there are lots of hairs to split here. For example, some startups are claiming “booked ARR”—numbers based on what customers might pay in the future rather than what they actually are paying now—even though contracts frequently have provisions that let customers opt out at any time for any reason.

“Companies are signing contracts with kill provisions, so they’re claiming booked ARR, but giving their customers an out,” said one early-stage focused VC. “So it’s like, okay, so I’m claiming that I just booked this million-dollar-a-year contract. But by the way, it says in three months, you can cancel for no reason. Does that count?”

It’s important to say: While there is a massive amount of variation across industries, there are also widely accepted, optimal accounting principles. In general terms, there are normal red flags around revenue that accounting experts watch for. 

“If there’s speculation that revenue’s being inflated, that’s a primary concern among external auditors,” said Jonathan Stanley, director of Auburn’s Harbert College of Business School of Accountancy and KPMG endowed professor. “There are always many things you’re looking for, but a company potentially trying to manipulate the revenue numbers to achieve a goal that really contradicts objective reporting is always a red flag.”

And revenue itself, on a fundamental level, features both core truths—and discretionary realities. 

“You book revenue when the service is provided and/or when the goods are delivered,” said Bradley Bennett, accounting department chair and professor at the University of Massachusetts, Amherst’s Isenberg School. “Depending on how the contracts are written, depending on how clear those stated objectives or benchmarks are noted, and/or just the industry in general, there’s some room for discretion and perhaps misreporting, intentional or not. Also, there are often incentives tied to revenue for management and members of the sales teams.”

Until proven otherwise, there’s nothing illegal about taking the rosiest view of revenue, and many would even say it’s a time-honored tradition. But that doesn’t mean it can’t cause problems (or crises) down the line.

The circular startup ecosystem

There are also broader sociological changes making the ARR shenanigans possible. One VC says part of the fault lies in well-established accelerators which have standardized “what to say” to raise money, encouraging metric manipulation.

Y Combinator, this VC says, “standardized the approach to building companies to such a degree, mostly for the betterment of our industry, for the record,” said one VC. “But they’ve also productized company-building to the extent that these people know exactly what to say. They’ve been in YC for ten weeks, so they think they know and they figure that annualizing whatever they’ve got in week nine feels like a reasonable thing to do.” (YC did not return a request for comment.)

And the fact that lots of these startups ultimately sell to other startups circuitously makes things even more insular. “More than private equity, more than even banking, venture has an ‘in’ crowd,” said NYU Professor Alison Taylor. “A certain sort of person gets funded with a certain sort of business model.” 

The emphasis on ARR, ultimately, is reflective of a wider reckoning in venture overall. Not only are there more VCs (and more capital) than ever, but priorities are in flux. “Generally, historically, there’s been an important tradeoff in the venture capital industry between profitability and growth,” said Dr. Ilya Strebulaev, a professor at the Stanford Graduate School of Business and co-author of The Venture Mindset. But roiled by geopolitical tensions and macroeconomic uncertainty, “that pendulum has been changing over time. I think venture capitalists are now spending more effort on profitability today than in the past, and are spending more effort on revenue. But that doesn’t mean the tradeoff between profitability and growth has evaporated—absolutely not.”

In the end, as University of Virginia economist Dr. Anton Korinek points out, this isn’t about ARR at all, but one step in a much bigger (and even more consequential) design. “The big picture question is, why are valuations so high?” said Korinek. “This is one of the symptoms of that. The bet is AGI or bust…’If I want to give you even more money, because there’s so much liquidity sloshing around and we are really, really eager to invest in this, then please give me more ARR, and I’ll give you a higher valuation.’”

One VC says he feels like he’s going a little mad—but that’s the business. “It’s like going to a carnival and saying ‘wait a second, this game where I’m supposed to throw a ring around the milk bottle—that’s not a real milk bottle, that’s not a real ring.” 

The consensus among VCs seems to be that ARR won’t ultimately be the way forward at all: Smart investors will develop new ways to assess AI businesses, focusing on retention, daily active usage, and unit economics. 

Until then, it may be the VCs and founders who have pumped up ARR that will feel the most pain if a bubble bursts. This is an equity-driven boom, says Korinek, and “the main losers in equity-driven booms like the one right now are the ones who made the bets.”