Hustler Words – At the vanguard of artificial intelligence, ElevenLabs is rapidly establishing itself as the definitive architect of AI’s auditory dimension. The company specializes in crafting advanced models that convert text into speech so uncannily human-like, many users interact with it daily without even realizing. This groundbreaking technology underpins the first-line phone support for an astounding 35 million U.S. customers of financial giant Klarna, and its expanding roster of enterprise clients includes titans like Deutsche Telekom, Cisco, and Adobe, alongside a growing number of governmental entities. Beyond corporate applications, ElevenLabs also empowers creators, providing the voice for audiobooks, dubbing projects, and even musical compositions.
Despite its rapid ascent, the landscape for ElevenLabs is not without its competitive challenges. The firm increasingly finds itself in a dynamic market where former collaborators, such as conversational AI platform Decagon—which initially leveraged ElevenLabs’ voice technology—are now evolving into direct competitors. Yet, this intensifying rivalry appears to do little to dampen investor enthusiasm. Just four years since its inception, ElevenLabs reportedly commands an impressive $600 million in annual recurring revenue (ARR) and is now valued by its backers at a staggering $22 billion.
To delve deeper into the company’s trajectory and strategic vision, Hustler Words had the opportunity to interview Mati Staniszewski, co-founder and CEO of ElevenLabs, at the Nrth entrepreneurship conference in Toronto. The conversation spanned critical industry topics, from the ethical imperative of disclosing AI interaction to the company’s long-term financial outlook and its role in shaping the future of AI.
Staniszewski, while unable to disclose specifics regarding gross margins, conveyed a clear strategic stance: he is prepared to see margins tighten if it translates into significant market share expansion. On the crucial question of transparency, the CEO firmly believes businesses should currently inform customers when they are interacting with an AI, acknowledging the present societal expectation. However, he anticipates a paradigm shift within the next five years, where the ubiquity of personal AI agents will lead to an expectation of AI interaction, altering the need for explicit disclosure. He suggests offering customers a choice between waiting for a human or engaging with an AI agent, noting that the latter often leads to surprisingly positive experiences.
Reflecting on his past prediction at a hustlerwords.com event that audio models would be commoditized within a couple of years, Staniszewski acknowledged that while significant progress has been made, a substantial "quality delta" still exists at the model level. He projects that these differences will diminish over the next three to five years. The ultimate aspiration for ElevenLabs, he revealed, is to be the first to achieve the Turing test for conversational AI, a feat requiring not only advanced intelligence but also sophisticated emotional understanding to modulate speech and pace appropriately.
ElevenLabs’ business model demonstrates a robust diversification, with over 55% of its $600 million ARR derived from classic enterprise clients. The remaining 45% is attributed to a vibrant ecosystem of small and medium businesses, independent developers, and content creators. Staniszewski noted the increasing fluidity between model companies, platform providers, and application developers, citing Anthropic as an example of this evolving landscape where lines are becoming "much more blurry."
Regarding the choice between "frontier lab models" and "open-weight" options for the "reasoning layer" of their AI, Staniszewski explained that the decision is highly contextual. For purely informational customer service interactions, open-source models can suffice, with the knowledge base being the primary determinant of experience quality. However, for high-stakes scenarios like financial services, where authentication and transactional accuracy are paramount, "frontier models will still lead" due to the zero-tolerance for error. This nuanced approach extends to their government clients, where deployments are tailored to specific national requirements, potentially integrating open-weight, closed-source, or custom fine-tuned models while ensuring data residency, as exemplified by a Polish healthcare project aimed at reducing missed appointments.
When pressed on the company’s gross margins, particularly given the costs associated with models and inference, Staniszewski offered a strategic rather than a numerical answer. He emphasized ElevenLabs’ internal research capabilities, which allow them to "fine-tune and constrain models in extremely smart ways." Ultimately, he reiterated a willingness to prioritize customer value and market expansion, even if it means accepting lower margins in the short term to foster long-term growth.
The CEO also shed light on ElevenLabs’ rigorous training methodology. Rather than solely relying on vast volumes of data, the company has invested heavily in meticulous annotation. Thousands of contractors, including professional voice coaches, are employed to precisely label not only what is said but also how it’s said, including speech patterns, emotional nuances, and accurate accent detection. In some instances, ElevenLabs co-creates bespoke models directly with enterprise partners for specific use cases.
Addressing reports of a potential IPO target around 2028, Staniszewski remained strategically vague, stating the company is "preparing the foundation to be able to do it in the next years," with the ultimate timing dependent on market conditions.
On the broader industry debate surrounding the pace of frontier AI development, Staniszewski affirmed that "everybody is aligned to work together on finding a way to pace" and that "we should all take the right precautions as we deploy the technology." He clarified that ElevenLabs’ core focus is on voice synthesis, not the development of the underlying text models or the "intelligence side of models," which is central to the ethical and safety discussions. He also distinguished ElevenLabs from potential vulnerabilities seen in other AI platforms, asserting that their technology does not facilitate self-replicating agents or allow agents to create more agents, with robust KYC protocols and cybersecurity measures in place to mitigate risks.
Leave a Comment