ElevenLabs, the company that builds voice technology for artificial intelligence, says it is now pacing at $600 million in annual recurring revenue. Its CEO shared new details about the business in an interview with TechCrunch.
The four-year-old company is reportedly valued at $22 billion by its backers. Co-founder and CEO Mati Staniszewski spoke at Nrth, an entrepreneurship conference in Toronto formerly known as Elevate.
ElevenLabs makes models that turn text into speech that sounds human. Many people hear its technology when they call customer service, often without knowing it.
Klarna uses it to run first-line phone support for 35 million U.S. customers. Deutsche Telekom, Cisco, Adobe and several governments are also customers.
Enterprise Makes Up Most of the Business
Staniszewski said more than 55% of revenue comes from large enterprise customers. Much of the remaining 45% comes from small and medium businesses, developers and creators.
Creators use the platform for audiobooks, dubbing and music. Businesses mostly use it for customer service calls.
The company also faces competition from some of its own customers. Decagon, a conversational AI platform, trained its voice product on ElevenLabs and now competes with it.
Staniszewski said the lines between model, platform and application companies have become "much more blurry." He pointed to Anthropic as an example of a model company that now offers a wide set of applications.
Customers can choose which AI model powers the "reasoning layer" of their agents. He said open-source models work well for simple information calls, while frontier models still lead in areas like financial services.
Government deployments vary by country. In Poland, AI agents call patients to remind them of health appointments, since 18% of patients never show up.
Margins, Disclosure and IPO Plans
Staniszewski declined to share gross margin figures in detail. He said the company does not mind margins going lower if it helps customers and grows market share.
He said businesses should tell customers when they are talking to an AI agent. "The common pattern is you don't want to feel cheated on that call," he said.
He expects that view to change within five years, as more people use their own AI agents. He suggested companies offer a choice, such as an agent instead of a 30-minute wait for a human.
On training, he said the main work has been annotating data rather than collecting large volumes of it. Thousands of contractors help label what was said, how it was said and what emotions were used.
He also said ElevenLabs does not let agents create more agents. Every customer goes through know-your-customer checks.
Last year, Staniszewski predicted audio models would become commoditized within a couple of years. He now says quality gaps between models will likely shrink over the next three to five years.
Reports have said the company is looking at a 2028 initial public offering. Staniszewski did not confirm that timeline, saying ElevenLabs is "preparing the foundation" to go public "in the next years," depending on the time and place.
