Sesame

Sesame operates a consumption-based infrastructure model: customers pay per unit of voice synthesis — typically per character, per second of audio, or per API call. They sell low-latency, multilingual voice models to voice-agent builders and conversational AI developers who integrate text-to-speech into their own products. Revenue is recognized as customers consume inference compute on the platform, either through prepaid credit packages or postpaid invoicing. The core product is a set of proprietary neural voice models exposed over REST APIs and SDKs, with developers billed on usage volume. In this model, marginal cost is the compute required for each inference — GPU cycles for token generation and waveform decoding. Gross margins improve as inference throughput per GPU increases and as the company optimizes model architecture for lower cost per character. No up-front hardware or per-seat fees; the model scales linearly with each customer's usage. The economic shape is similar to other API-first AI companies: high gross margin on unit economics once inference cost is covered, with the key lever being model efficiency and inference engine optimization rather than user count or seat expansion.