The $100 Million Question: How Emergent’s ARR Debate is Forcing Re-evaluation of SaaS Metrics

When an AI start-up says it has hit $100 million ARR (annual recurring revenue) in eight months, the number sounds like a classic breakout signal. But the Emergent case shows that in AI, the same acronym can mean very different things, and that difference may decide how the market values the next generation of start-ups.

Emergent’s claim that it had crossed $100 million in ARR within eight months was supposed to read like a milestone. Instead, it became a referendum on one of the most contested questions in AI start-up finance: what, exactly, counts as ARR when the product is sold on usage rather than subscription?

The dispute began when Emergent flashed on its website that it had reached the $100 million mark. After that, several reports started pointing out that the company was not describing classic SaaS ARR, but an estimated annualised revenue run rate based on recent usage.

That subtle distinction is what turned the announcement into an industry debate.

What is the Debate About?

In SaaS, ARR traditionally implies recurring, contractually committed revenue that investors can underwrite with some confidence. However, In Emergent’s case, the number appears to have been derived from realised monthly revenue, annualised forward.

Emergent CEO Mukund Jha’s explanation to Moneycontrol was that the company had roughly $8.3 million-$8.4 million in revenue in a month, which annualises to about $99.6 million. More importantly, he said, this was “money that has hit our bank,” not contracted annual revenue.

That statement matters because it shifts the debate away from whether the revenue is real and toward whether the label attached to it is doing too much work.

That is the heart of the argument. Emergent’s argument says the company is being judged by an old SaaS rulebook that does not fit AI-native businesses. In a usage-based model, customers are not buying a fixed seat count or a multi-year licence; they are consuming tokens, credits or credits tied to model usage.

Revenue can therefore spike quickly when usage spikes, which makes annualised run-rate figures feel economically intuitive, even if they are not contractually locked in.

The counter-argument, however, says the problem is not the math. It is the framing. A number annualised from one strong month can make a company look far more durable than it may actually be, especially if usage was helped by discounts, heavy power users, or an unusually favorable period of demand.

That is why some reports described the claim as “smoke and mirrors” or “vibe accounting.”

Their concern is not that Emergent’s reported figure is fictional. It is that the headline may smuggle in a SaaS-era sense of predictability that usage-based AI revenue does not yet deserve.

The Argument

The loudest pro-Emergent voices on X have argued that the old grammar of software metrics is too rigid for the AI era.

For instance, Hemant Mohapatra, Partner at Lightspeed India has said token consumption is what really matters, framing Emergent’s revenue as real, banked money rather than theoretical future revenue.

Similarly, Vinod Khosla, founder of Khosla Ventures has a more permissive stance on the issue. He stated that there are many ways ARR is measured now, he has argued, but cash collections are indisputable.

Freshworks’s co-founder Girish Mathrubootham too, has pushed back against over-reading private-company numbers, suggesting that outsiders should allow founders to execute rather than litigate every disclosure in public.

The Counter-Argument

Manav Garg, founding partner at Together Fund occupies a more nuanced middle ground. His X posts support the view that Emergent’s growth is impressive, but he also asks what the real number is.

In comments to ET, Garg has said the company mixes subscription pricing with usage-based credits, making the metric better understood as a run rate than classic seat-based ARR. That is an important distinction because it shows the debate is not neatly split between believers and skeptics.

Even some supporters are effectively saying: the business may be real, but the label needs qualification.

That qualification is where the Indian AI start-up ecosystem is running into a broader global problem. Tokens sit at the center of the new software economy. A token is, in simple terms, a unit of text processed by an AI model. In token-based products, a customer’s bill can rise the moment usage rises. Supporters say this is precisely why token consumption is the most honest metric for AI businesses: it reflects actual product engagement and actual cash collected.

The counter-argument in this debate is that token-driven revenue may be real, but it can also be more volatile, more discounted and less predictable than classic ARR.

Need for New Metrics

One month of strong usage does not necessarily translate into a stable annual base. A customer may be enthusiastic today and gone tomorrow. Pricing may normalize. Power-user concentration may fade. And because AI infrastructure costs remain high, gross revenue growth may not translate into software-like margins.

That is why Gopal Jain, managing director and chief executive of Gaja Capital, brings a useful investor lens to the debate. Responding more broadly to the question of how AI start-up revenue should be judged, Jain said the real markers of quality are gross margins, retention, and customer concentration.

“Gross margin reveals whether a company has a product or a service dressed up as one, AI-native companies often carry significant infrastructure costs that compress margins well below traditional SaaS benchmarks, and that compression determines how much of the top line compounds into long-term earnings,” he said.

He added that net revenue retention is “the most honest signal of product-market fit” in a market where model commoditisation is accelerating and switching costs remain low. Customer concentration, he warned, is the hidden risk behind impressive headline numbers, because a handful of large contracts can produce a strong run rate while leaving the underlying business fragile.

Jain also drew a line between different revenue models, saying usage-based pricing is often the right structure for AI-native products. He pointed to outcome-based pricing as the more mature evolution, because it aligns incentives more cleanly and makes revenue easier to underwrite. But his larger warning was about comparability.

When companies define ARR differently, contracted, annualised, or run-rate, investors are not looking at the same thing even when the label is identical. His framing captures the core issue around Emergent: the number itself is not the controversy. The assumptions inside it are.

That is why this debate matters beyond one start-up. Emergent has become a proxy for a bigger transition in software finance. SaaS taught investors to prize predictability, retention, and contracted revenue. AI, especially usage-based AI, changes the cadence. Revenue can arrive quickly, but it may not arrive smoothly. It can be “real” without being “recurring” in the old sense. It can be banked without being durable. It can be impressive without being fully underwritable.