Workload Usage Does Not Equal Revenue

The key point is that Speechmatics is still a mixed business, not a pure voice AI usage meter. A surge in one fast growing workload, like voice agents or medical transcription, can sit inside a broader revenue base that also includes slower moving enterprise deals, committed contracts, and deployment models such as cloud, on premises, and on device. That mix makes workload growth a weak proxy for company wide revenue growth.1 2 3 4

• Speechmatics has grown much more gradually than newer API first peers. Estimated revenue moved from $14.5M in 2023 to $16.1M in 2024 and $18.7M in 2025, while Deepgram reached a $100M annualized run rate in August 2026 and AssemblyAI reached $76M annualized revenue in October 2026. That gap shows how fast application usage can scale at rivals without mapping cleanly to Speechmatics overall.1 2 3
• Different workloads monetize differently. AssemblyAI charges separately for pre recorded transcription, streaming, voice agent minutes, LLM usage, and add ons like medical transcription and redaction. Deepgram likewise sells speech to text, text to speech, and a unified Voice Agent API. In practice, one minute of audio is not one unit of revenue across products or customers.2 3
• Speechmatics also sells deployments that behave less like self serve API spend and more like enterprise software. Its product footprint includes cloud APIs plus on premises and on device transcription. That means revenue can be shaped by contract timing, committed spend, and deployment mix, not just raw usage in whichever application is growing fastest.1 4

Going forward, the important question is not whether a single Speechmatics workload is growing quickly, but whether high growth use cases become a large enough share of the mix to pull the whole company onto a faster curve. If voice agents and premium vertical workloads keep compounding, they can gradually shift Speechmatics from steady enterprise growth toward a more usage driven profile.1 2 3 4