Recall.ai resembles video infrastructure
Recall.ai
This cost structure makes Recall.ai look less like classic SaaS and more like payments or video infrastructure, where revenue scales with usage but so does the bill to serve it. Every live meeting bot runs on cloud machines that join the call, capture audio, video, and metadata, and push that data back through Recall.ai’s API. That is why pricing is per minute, gross margins are more moderate, and engineering work on CPU and network efficiency directly changes the business model.
-
A normal SaaS seat can often be added with little extra cost once the software is built. Recall.ai instead has to spin up compute for each concurrent meeting, with early workloads around 4 vCPU per bot. More customers means nearly proportional infrastructure scaling, not just more database rows.
-
This is why optimization matters so much. Recall.ai cut per bot CPU usage in half by changing how internal data moved between services, materially reducing AWS spend. In this model, small systems changes can expand gross margin far more than a typical feature launch.
-
The closest analogs are usage based infrastructure APIs like Mux, not high margin application SaaS. In video infrastructure, expensive workloads in bandwidth, compute, and storage force companies to design around COGS from day one. Recall.ai faces the same discipline because every meeting is a live media workload.
Going forward, the upside comes from shifting more of each meeting from raw capture into higher value software layers. Desktop recording, mobile capture, transcription, and AI agent workflows can raise revenue per conversation without requiring a silent bot for every use case, which is the clearest path to better margins and a broader platform position.