ElevenLabs is pacing at $600 million in annual recurring revenue and is now reportedly valued at $22 billion, according to CEO Mati Staniszewski, who laid out the numbers at the Nrth conference in Toronto. The four-year-old voice AI company sells the text-to-speech models behind Klarna's first-line phone support for 35 million U.S. customers, along with deployments at Deutsche Telekom, Cisco, Adobe, and a growing list of governments. Enterprise now accounts for more than 55% of revenue, with small and medium businesses, developers, builders and creators making up the remaining 45%.
Staniszewski also acknowledged the reported 2028 IPO timeline without confirming it, saying the company is preparing the foundation to go public in the next years but that timing will depend on conditions. The valuation puts ElevenLabs among the most richly priced private AI companies relative to age, at roughly 37 times run-rate revenue if the $22 billion figure holds.
On models, Staniszewski walked back — slightly — a prediction he made at TechCrunch Disrupt last year that audio models would be commoditized within a couple of years. He now says the quality gap at the model level is still significant, though he expects it to narrow over the next three to five years. His stated goal is to be the first to pass the Turing test for conversational AI, which he argues requires combining raw intelligence with emotional intelligence — a model that can detect when to slow down or speak up based on the caller's tone.
“There is still a lot of work to be done, and the quality delta you can achieve just on the model level is still significant.”— Mati Staniszewski, ElevenLabs co-founder and CEO
Key facts
- 01ElevenLabs is pacing at $600M in annual recurring revenue and is reportedly valued at $22 billion by its backers, four years after founding.
- 02Enterprise accounts for 55%-plus of revenue; SMBs, developers, builders and creators make up the remaining 45%.
- 03Klarna runs first-line phone support for 35 million U.S. customers on ElevenLabs; Deutsche Telekom, Cisco and Adobe are also customers.
- 04CEO Mati Staniszewski says the company is preparing the foundation for an IPO reportedly targeted for 2028, but timing is not fixed.
- 05In Poland, ElevenLabs-powered agents call patients to cut an 18% no-show rate in public healthcare appointments.
The competitive picture is getting messier. Decagon, a conversational AI platform, trained its voice product on ElevenLabs and now competes with it. Staniszewski framed this as the new normal, drawing a parallel to Anthropic, which he described as a model company that has become a platform and increasingly a wide set of applications. The clean splits between model, platform and application layers are gone.
Customers pick the reasoning layer from a menu at ElevenLabs, and Staniszewski says the choice between frontier models and open-weight alternatives is use-case dependent. Informational customer service calls, where the knowledge base defines a good experience, can run on open-source models. Financial services calls involving authentication, transaction details or refunds still go to frontier models because, as he put it, there's no room for error. Some of the open-weight options are Chinese, which shapes conversations with U.S. and European government customers — each deployment gets a different model mix depending on residency and procurement requirements.
One government case study: Poland's public health system has an 18% patient no-show rate for appointments. ElevenLabs deployed reminder-call agents built on models tuned to the Polish government's knowledge base, with data residency preserved. The company also works with the Brazilian government under similar terms.
“But if we can pass on any savings to the customer, we do that.”— Mati Staniszewski, ElevenLabs co-founder and CEO
On gross margins, Staniszewski was deliberately vague but explicit about direction: he does not mind them going lower if it means winning market share. He said the company's research capability lets it fine-tune and constrain models efficiently, and that any savings get passed to customers rather than banked. The bet is that the value created over the next five years will be worth compressed near-term margins.
Training data is another differentiator. ElevenLabs handles millions of hours of customer service calls but says the volume of data is less important than the annotation. Thousands of contractors help label not just what was said but how — when people spoke, what emotions were present, how phrases were delivered. Voice coaches were brought in to detect accents accurately. In some enterprise deals, ElevenLabs and the customer built the model together.
On the ethics of AI voice agents, Staniszewski says businesses should disclose when a caller is speaking to an agent rather than a human, at least for now. He argues that will shift in about five years, once everyone has their own agent and expects an agent on the other end of the line. His preferred pattern: when there's a 30-minute wait for a human, offer the customer a choice. He claims almost all callers pick the agent and are surprised by the quality.
Asked whether ElevenLabs could face the kind of scrutiny that has hit Hugging Face over model hosting, Staniszewski drew a line: ElevenLabs does not train the text models or the intelligence layer at the core of the frontier-lab safety debate, and its technology does not let agents create more agents. Every customer goes through KYC. He acknowledged cybersecurity risk exists but said the guardrails are in place.
The $22 billion price tag is the interesting tension in this story. At $600 million ARR, ElevenLabs is at the revenue scale where a public-market listing becomes credible, and a 2028 window gives the company time to expand enterprise share while margins compress in a fight with Decagon and other rivals building on top of its own stack. Voice is one of the few AI categories where a specialist can plausibly hold ground against the frontier labs — the annotation moat and the emotional-intelligence problem are both real. Whether that's worth 37 times revenue is the question the 2028 IPO would answer.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
