The End of the MAU Era, and Now MRR Can't Be Trusted Either?

Analysis is emerging that when measuring the growth of AI companies, ARR (Annual Recurring Revenue), the core metric of traditional SaaS, can distort the actual health of the business. As seen in the cases of Anthropic and OpenAI, a gap is emerging between the announced ARR and the actual cumulative revenue and cash burn rate.

The 3 reasons ARR shows limitations in the AI era are as follows.

• High marginal cost: Unlike SaaS, AI incurs GPU and cloud costs for every inference call, so costs increase along with revenue.
• Cost variance by customer: Even when customers pay the same fee, the infrastructure costs incurred vary greatly depending on their usage patterns, making profitability difficult to predict.
• Instability of recurring revenue: Switching costs between services are low, making revenue continuity (Recurring) weaker than in the past.

Accordingly, next-generation metrics to measure the actual value of AI businesses are drawing attention.

• Productivity per Dollar Spent: Measures actual efficiency by dividing ARR by the sum of labor costs and AI infrastructure costs.
• First Year Value: Checks whether a customer experiences sufficient value within the first 12 months to renew.
• Gross margin per token: A unit-economics-focused metric that concentrates on how much profit is left per token, rather than simply throughput.

Ultimately, in the AI era, reading gross margin structure and per-customer profitability rather than revenue scale will become the key factor determining business success or failure.