ElevenLabs is reporting $600 million in annual recurring revenue as it pushes its voice technology deeper into business use. In an interview with TechCrunch, co-founder and chief executive Mati Staniszewski said more than 55 percent of that revenue comes from conventional enterprise customers. Much of the remainder, he said, comes from smaller businesses, developers and creators.
The company makes speech models used for customer service, dubbing, audiobooks and other audio work. Its customer list, as described by TechCrunch, includes Klarna, Deutsche Telekom, Cisco and Adobe. A reported $22 billion valuation reflects investor expectations, but the article does not provide audited revenue, profitability or gross-margin figures. Those distinctions matter when a fast-growing model provider begins competing with some of the companies that have built on its tools.
Enterprise demand broadens the voice business
Staniszewski described the company’s goal as moving beyond convincing speech synthesis toward conversational systems that can respond to context and emotion. He said today’s models can still differ meaningfully in quality, even if those differences narrow over the next three to five years. Passing a conversational Turing test remains an ambition, not a milestone he claims ElevenLabs has reached.
The enterprise share suggests that large deployments are now central to its business. Staniszewski discussed public-sector examples and customer support uses, including an appointment-reminder deployment in Poland. He said such systems can combine different underlying language models with ElevenLabs’ voices depending on a customer’s requirements, including data residency. The interview does not quantify the effectiveness of the Polish deployment or disclose how much each named client spends.
He also distinguished simpler informational calls from transactions that require authentication or action. In his view, open-weight models can serve some knowledge-base interactions, while frontier models remain preferable for more demanding financial-service requests. That is his product judgment, not a comparative study in the article.
Competition reaches across the platform boundary
TechCrunch points out that Decagon, a conversational AI company, trained a voice product on ElevenLabs and now competes with it. Staniszewski argued that the borders between model provider, platform and application maker are becoming less clear. For ElevenLabs, that means a customer relationship can also become a competitive relationship as businesses gain the ability to build more of their own voice stack.
The chief executive did not give a detailed gross-margin figure. He said he is willing to accept lower margins if it helps expand market share and pass savings to customers. That frames the company’s current priority, but without costs or profitability disclosed, readers cannot calculate how sustainable that trade-off is.
Staniszewski also described the labor behind training the models. He said thousands of contract workers help annotate not just words but timing, emotion and accent, with voice coaches brought in to improve accent identification. The account complicates a purely automated picture of voice AI development: the model output depends in part on extensive human labeling. The interview does not state the total cost or working conditions of that annotation effort.
Disclosure and safety remain open questions
Asked whether callers should be told when they are speaking to an AI agent, Staniszewski favored disclosure at present. He expects social expectations could change as people become more accustomed to agents, but that is a prediction. He suggested offering customers a choice between waiting for a person and speaking to an agent; the article does not independently verify his claim that most choose the agent when offered that option.
On security, he said ElevenLabs does not deploy agents that can create more agents and that customers pass identity checks. He acknowledged wider cybersecurity risk while arguing the company’s architecture limits exposure. Those are company assurances, not a published security audit.
The interview therefore presents a business with substantial claimed recurring revenue and a strategy built around enterprise deployment, even as prices, margins and safety evidence remain partly opaque. Its reported valuation may be large, but the clearest concrete disclosures here are the revenue mix, the company’s willingness to trade margin for adoption and its leader’s view that customers deserve to know when a voice is artificial.
