The Traditional SaaS Blueprint
For years, startups aiming for major success followed a predictable path. Companies invested heavily in product development and customer acquisition before targeting revenue milestones like $100 million in Annual Recurring Revenue (ARR). In Silicon Valley terms, this figure signaled not just scale but crucial predictability—a business that could consistently sell, retain, and grow its customer base.
Traditional Software-as-a-Service (SaaS) ARR relies on committed, contracted income. A typical scenario involves subscription fees, where revenue is locked in for set periods, allowing for reliable forecasting. Whether through numerous per-user licenses or a single large enterprise deal, this contracted model provided a stable foundation.
AI's Usage-Driven Disruption
That established rhythm is rapidly changing. A new generation of AI-native companies is compressing timelines dramatically. Emergent, an Indian AI app-building platform, announced reaching $100 million in ARR within eight months of launch. Midjourney, an AI image generator, and ElevenLabs, specializing in AI voice tools, are global examples that scaled to hundreds of millions far faster than their SaaS predecessors.
More recent entrants, such as Lovable, a prompt-based app builder, reportedly hit approximately $400 million in ARR just over a year after launch. Decagon, which develops AI agents for customer support, achieved seven-figure ARR in roughly six months and is now said to be operating at over $30 million. Across these successes, a consistent pattern emerges: rapid early revenue growth driven by usage, not long-term contracts.
The Shifting Definition of ARR
This change in business models fundamentally alters what ARR represents. Instead of primarily charging for access, many AI companies price based on usage—tokens consumed, queries run, or applications built. Revenue is therefore intrinsically linked to how much a product is used, creating a less predictable income stream compared to fixed subscriptions. "What many AI companies are calling ARR is really an annualised run rate combining subscription and usage," notes Girish Mathrubootham, co-founder of Freshworks and an investor.
This has led to ARR often being calculated as an "annualised run rate," where recent monthly performance is multiplied by 12. While this captures momentum, the underlying revenue can be spiky. Spending may surge during intensive build or deployment phases and ease once that work is complete. "If it includes token or compute costs, that's fine. Investors just have to underwrite the margins and volatility accordingly," suggests Aakrit Vaish, founder of AI-focused venture fund Activate.
Challenges in Measurement and Interpretation
Varied reporting methods exacerbate the complexity. Some companies report the full amount billed, while others deduct infrastructure costs, leading to different ARR figures for similar customer activity. This lack of standardization creates a "metric in flux." Rituraj Biswas, founder of AI video platform Hypergro, argues for a new, standardized way of evaluating AI companies, especially consumer AI, stressing clear disclosure. Dhananjay Yadav of Neosapien, however, maintains that core financial definitions like revenue and ARR should not change, even as valuation multiples adapt.
Ultimately, while ARR still signals growth, its interpretation in the AI era requires deeper context. Investors must now scrutinize how revenue is generated, how usage patterns evolve, and the underlying consistency and margins of these usage-driven models to truly understand business health.