ElevenLabs, valued at $6.6bn and profitable, says voice models will commoditize over the long term, so its edge must come from more than models alone.
Mati Staniszewski says ElevenLabs will not train the compute-intensive models rivals build, focusing instead on human-computer interaction and conversational voice agents.
Staniszewski says ElevenLabs can trace generated content to its creator from release and is close to its $300m ARR target.
ElevenLabs bets on agents and products as voice models commoditize
ElevenLabs, a profitable voice AI company valued at $6.6bn, expects its core product to become a commodity. Staniszewski's answer is to own the layer around the model, pairing its in-house audio expertise with products, conversational agents and fused multimodal systems. He argues that focus on how people talk to technology, plus trust tools like traceability, lets a specialist stay independent against generalist giants.
Takeaways
The CEO expects voice models to commoditize. Staniszewski says voice models will become commodities over the long term. In the next year or two he expects fused, multimodal approaches, citing Google's Veo 3 as a strong example of combining audio and video models.
Model plus product is the defensible position. ElevenLabs builds its own models, while OpenAI, Google and Adobe often partner. In audio, architecture matters more than scale. Staniszewski compares fusing AI with product to Apple pairing hardware and software, though he won't train rivals' compute-heavy models.
Conversational agents are now the main focus. The company began as a tweet-length text box for downloading speech, then added music and speech-to-text. Voice agents have been its biggest focus for two years. Hippocratic uses them to schedule appointments and send medicine reminders.
Focus, not size, separates it from giants. OpenAI and Google already have creator tooling and distribution. Staniszewski says ElevenLabs differs by concentrating on human-computer interaction, and that the big labs take far more risk with recent releases than ElevenLabs would.
Revenue is near the $300m target. ElevenLabs passed $200m in ARR and aimed for $300m by the end of 2025. Staniszewski says it is close and on track. Partners include Cisco and Deutsche Telekom, and it works pro bono with Ukraine's government.
A voice marketplace pays human contributors. About 10,000 shared voices have earned the community $10m, and voices work across 70 languages. Staniszewski argues human value lies in story and emotion, and that cheaper tools let more people create rather than fewer.
How ElevenLabs plans to handle deepfakes
Staniszewski predicts AI-generated content will outnumber human content, with AI music already approaching half of all music.
His framework has three layers. Detect and certify human content, watermark authenticated AI voice agents, and treat everything else as potentially fake.
ElevenLabs mostly operates on the second and third layers today.
Its content can be traced to whoever generated it from the moment of release, and it builds detection classifiers with university and industry partners including Reality Defender.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
It relies on moderation and pre-use restrictions, and warns that open-source voice tools offer no traceability or detection.