ElevenLabs Secondary Trading Valuation Rises to $22 Billion
London-based AI voice company ElevenLabs has seen its valuation rise to $22 billion after completing a $300 million equity sale, doubling its valuation from the last funding round in February this year, placing it among the most expensive AI labs outside of China.
Existing investors and employees sold shares in a tender offer led by Wellington Management and Franklin Templeton. New buyers include EQT, Goldman Sachs, Singapore's sovereign fund GIC, Ontario Teachers' Pension Plan, Sapphire Ventures, and BDT & MSD; original shareholders such as Andreessen Horowitz, Lightspeed, Iconiq, and D.E. Shaw continue to participate. In February, the Series D round raised $500 million from Sequoia, Andreessen Horowitz, and Iconiq, valuing the company at $11 billion. In September 2025, the company had a $6.6 billion valuation with $100 million in employee cash-out.
CEO Mati Staniszewski told the media that the secondary transaction aims to bring in long-term institutional investors and support the recruitment battle for researchers. He hopes the company will be ready for an IPO within two and a half years. The company was founded in London in 2022 by him and Piotr Dabkowski, expanding its products from text-to-speech to audible, speakable, and translatable agents, covering over 90 languages. ElevenAgents handles over 15 million conversations weekly, about three times the volume in February, managing refunds, renewals, appointments, and more. Enterprise clients account for 55% of revenue, and the technology has been adopted by five of the top ten tech companies, five of the top ten insurance companies, and four of the top ten telecoms globally. Clients such as Stripe, Deutsche Telekom, DoorDash's SevenRooms, and the governments of Ukraine and Greece have already deployed the technology.
The valuation level is close to France's Mistral, which completed a €3 billion financing this month at a valuation of €21 billion (approximately $24 billion), led by Samsung, marking the largest private tech equity financing in Europe. The ElevenLabs transaction involves the transfer of existing shares and does not necessarily add cash to the company's balance sheet; Mistral's financing was for new share issuance to expand computing power. Meanwhile, Meta is pushing its personal agent Muse into the small business voice market.
Funds are shifting from early employees and venture capital accounts to pension funds, sovereign funds, and public giants. Buyers seek recurring revenue from enterprise voice agents, while sellers desire liquidity before an IPO. The event is driven by secondary pricing rather than a new round of burning computing power. Beneficiaries are employees and early shareholders who can cash out at the $22 billion price, as well as institutions receiving long-term shares; those under pressure are later entrants pricing voice models based on application layer multiples and general large model labs that must use cash to expand training. The company did not provide the latest overall recurring revenue in the announcement; conversation volume and enterprise proportion are the anchors for this round of pricing.
Source: Public Information
ABAB AI Insight
Staniszewski and Dabkowski started the London company from dissatisfaction with voiceover quality, transforming it into a voice generation and then a call agent. The valuation trajectory shows a seed round of about $10 million in 2023, a secondary valuation of $6.6 billion in September 2025, a Series D valuation of $11 billion in February 2026, and a secondary valuation of $22 billion in September, jumping three orders of magnitude in three years. The company does not rely on the $300 million to supplement training budgets but allows employees to cash out before an IPO, changing the shareholder roster from Silicon Valley venture capital to Wellington, Franklin Templeton, GIC, and teacher pensions. This is a shift in identity for a European application layer company funded by American institutions.
Capital has shifted from 'burning another round for the model' to 'pricing the already launched customer service agents.' With enterprise revenue accounting for 55% and 15 million weekly conversations, it indicates that pricing power lies in deployment rather than parameter scale. Money goes into the pockets of early shareholders, while the company gains a shareholder structure that allows for a two-and-a-half-year IPO window. The motivation is to retain researchers: options that can be cashed out are more tangible than paper wealth that only appreciates in valuation. In contrast to OpenAI and Anthropic, which are still issuing billions to buy computing power, ElevenLabs chooses to turn computing power needs into customer-side call minutes, selling voice interfaces.
In comparison to Mistral's €3 billion purchase of European sovereign computing power and open-source model positioning, ElevenLabs exchanges $300 million of existing shares for institutional endorsement, presenting two European paths: one focused on training clusters and the other on industry voice agents. It also contrasts with Microsoft voice, Google and Amazon contact centers, and Meta's deployment of Muse into small businesses. The industry is shifting from demonstration-level human-like voices to daily connections for insurance companies, telecoms, and governments, expanding in minutes while controlling vertical models and integration.
The essence is the transfer of pricing power. General large models are financed based on computing power, while voice agents are financed based on call and renewal rates. The mechanism is: once refunds, renewals, and appointments can run stably on agents, investors are willing to price 'ears and mouths' at 40-50 times revenue, without questioning whether it has the largest base model. The secondary transaction writes this pricing into institutional holdings, triggering the IPO clock. Voice first becomes enterprise infrastructure, and then the valuation will detach from the lab narrative.
ABAB News · Cognitive Law
- The secondary sale is to the shareholder structure, not computing power.
- Call minutes are closer to revenue than parameter scale.
- Options that can be cashed out are what retain researchers.