The people shaping ElevenLabs’ $3.3B voice AI platform

The scale of ElevenLabs

Mati Staniszewski leads ElevenLabs, a research company that reports $600 million in annual recurring revenue. Backers reportedly value the company at $22 billion. The company serves a diverse range of clients, from creators to massive corporations. Klarna uses the platform to handle first-line phone support for 35 million U.S. customers. Other major clients include Deutsche Telekom, Cisco, Adobe, and Revolut. The business split includes 55 percent enterprise customers and 45 percent small and medium businesses, developers, builders, and creators. While the company maintains a 55 percent enterprise customer base, the remaining 45 percent includes small and medium businesses, developers, builders, and creators who use the platform for audiobooks, dubbing, and music. Backing for the company includes $80 million from an early 2024 round led by Andreessen Horowitz and Sequoia Capital, plus a $500 million Series D involving BlackRock, Wellington, D.E. Shaw, Nvidia, Salesforce Ventures, and celebrities like Jamie Foxx and Eva Longoria. Staniszewski focuses on a mission to make content universal, ensuring anyone can access information in a voice that resonates in their own language, accent, and tone. In Poland, the technology helps the public health system by using agents to call patients and remind them of appointments, which addresses an 18 percent no-show rate. The company also prepares the foundation for a potential IPO, with several major investment banks in discussions to underwrite the offering.

Emotional intelligence and competition

Staniszewski works to pass a conversational Turing test by building models with emotional intelligence. These models must understand the user to adjust their speed or tone during a conversation. The company relies on thousands of internal contractors to annotate data, which captures how people speak and the emotions they use. Voice coaches help the models detect accents accurately. This focus on quality helps the company compete against Decagon, a platform that trained its own voice product on ElevenLabs and now runs its own queries. Staniszewski expects the quality gap between audio models to shrink in three to five years, but the current difference remains significant. He previously predicted that audio models would become commoditized within a couple of years. To maintain market share, the company is willing to accept lower gross margins to prove value to its customers.

The telephony infrastructure gap

OpenAI’s 2026 API cycle changes how developers build voice agents. The Assistants API shut down in August 2026, forcing a migration to the Responses and Conversations APIs. The hosted Agent Builder retires at the end of November 2026, with the Agents SDK serving as the path forward. The Realtime API moved to general availability in August 2026 with revised per-minute pricing. These changes only affect the model pipeline. A model vendor’s voice API provides audio in and audio out, but it does not reach the public telephone network. You probably already know that a model alone cannot make a phone call. The OpenAI service lacks SIP termination, carrier-grade routing, STIR/SHAKEN caller-ID attestation, A2P messaging rules, and E911 address registration. These requirements belong to the telephony layer. ElevenLabs focuses on the voice layer that integrates into existing communication workflows. For developers who need a bot to place a real phone call, the model pipeline is only one component. The OpenAI Realtime API remains useless for companies that require a phone number a customer can call. Will OpenAI eventually bridge the gap between model intelligence and telephony infrastructure?