← All posts / Industry

From $9 Million to $22 Billion in Four Years: ElevenLabs Doubles Its Valuation on the Strength of Voice Agents

ElevenLabs closed a $300M employee tender at a $22B valuation, 2x its February Series D, as its ElevenAgents platform hits 15M weekly conversations and enterprise climbs to 55% of revenue.

From $9 Million to $22 Billion in Four Years: ElevenLabs Doubles Its Valuation on the Strength of Voice Agents

Four years ago, ElevenLabs raised its first funding round at a $9 million valuation, announced alongside Eleven v1 — a text-to-speech model its founders described as the first to cross the “uncanny valley” of synthetic speech. This week, the company closed a $300 million employee tender offer that values ElevenLabs at $22 billion, double the $11 billion valuation from its February 2026 Series D. In an AI market where eye-popping valuations have become routine, the ElevenLabs story stands out for one reason: the number is being pulled up by a concrete, measurable enterprise business, not by model-release hype.

What actually happened

The transaction is a tender offer, not a traditional primary round. Existing investors and employees sold $300 million worth of stock, with the round led by Wellington and T. Rowe Price — two long-term institutional investors that typically write the checks reserved for companies with credible paths to public markets. As part of the tender, EQT, Goldman Sachs, GIC, OTPP, Sapphire Ventures, and BDT & MSD invested in ElevenLabs for the first time, alongside existing backers including Andreessen Horowitz, Lightspeed, ICONIQ, D.E. Shaw, DISRUPTIVE, and Alkeon.

For employees, the tender is a liquidity event pure and simple: a chance to sell shares in a company that remains private. For the company, it is a price-discovery exercise — and the price it discovered, $22 billion, is a 2,400x multiple on that first 2022 round and a clean doubling in roughly seven months.

The engine: voice agents went from demo to daily operations

The core driver, according to the company’s announcement, is enterprise demand for conversational agents. Enterprise now accounts for 55% of ElevenLabs’ revenue, a striking inversion for a company that started as a consumer-flavored developer tool for synthetic voices. Its technology is now used in daily operations at five of the world’s ten largest tech companies, five of the ten largest insurers, and four of the ten largest telecoms.

The product at the center of this shift is ElevenAgents, the conversational agents platform ElevenLabs built after its February Series D. Since that round, organizations including Stripe, Deutsche Telekom, DoorDash’s SevenRooms, Admiral, Customers Bank, Cadence, and the governments of Ukraine and Greece have deployed ElevenAgents across sales, support, and business operations.

The usage numbers are the most telling part. ElevenLabs’ agents now handle more than 15 million conversations every week — up 3x since February — resolving tasks like handling refunds and exchanges, renewing insurance policies, upgrading phone plans, booking healthcare appointments, and accessing public services. Over the same period, ElevenAgents’ annualized revenue has grown more than 3x. The company closed 2025 with over $330 million in ARR, so tripling the agents business on top of that base explains how a $22 billion price becomes defensible to institutional investors.

There is also a counterintuitive data point worth flagging: ElevenLabs says its own customer analysis shows voice agents resolve issues 31% faster than chat agents on average. The intuition that voice is a slower, more cumbersome channel than text does not survive contact with real deployments — people volunteer more context when speaking, and context is what lets an agent close a ticket in one turn.

The stack argument

Why is a speech company winning enterprise agent deployments rather than the frontier labs? ElevenLabs’ argument is structural: it owns the full stack. It builds the underlying models — speech synthesis, speech recognition, and translation covering more than 90 languages spoken natively by over 5.5 billion people — and then builds the application layer itself, tuning models and products together for expressiveness, latency, and reliability. Improvements in the models reach customers immediately, and feedback from real deployments flows straight back into training.

The model cadence supports the story. In late September the company shipped Eleven v4, its most expressive text-to-speech model, ranked #1 by Artificial Analysis, alongside Eleven v4 Turbo, a low-latency variant with a median inference latency of roughly 100 milliseconds built for real-time agent use. A healthcare-optimized transcription model followed the same pattern: general capability first, then verticalized versions tuned for industry-specific vocabulary and compliance. Forward-deployed engineers work with customers to orchestrate models and agents for specific industries and geographies — a playbook that will look familiar to anyone who has watched Palantir or OpenAI’s own enterprise efforts.

T. Rowe Price’s Emma Norchet framed the bet in exactly those terms: “As ElevenLabs has grown its conversational agent business, the combination of research, product, and deployment has proved an important structural advantage. The team is make improvements at every layer of the stack, allowing them to execute at high velocity.”

Why the price still raises eyebrows

A $22 billion valuation against a business that exited 2025 at $330 million-plus in ARR implies a rich revenue multiple, even assuming substantial growth in 2026. The bulls would argue that the multiple compresses quickly if ElevenAgents continues tripling, that enterprise voice is a winner-take-most layer because integration depth creates switching costs, and that Wellington and T. Rowe Price’s involvement signals IPO-grade diligence rather than momentum-chasing. The bears would note that every AI infrastructure company’s valuation now assumes near-perfect execution, that the frontier labs are pushing hard into voice themselves — OpenAI’s voice mode and real-time API compete directly for the same enterprise budgets — and that tender-led marks are softer prices than open-market ones.

Both things can be true. What the tender undeniably establishes is that the voice-agent category has moved from experiment to line item. Companies are not piloting conversational AI anymore; they are wiring it into refunds, insurance renewals, telecom plan changes, and government service delivery, and they are picking specialists where the specialist is best.

The road from here

ElevenLabs now employs more than 800 people, and the company explicitly framed the regular tender offers as a way to let the team “participate in the value they are creating” — language that usually accompanies a company preparing for a public listing. With Anthropic’s monster IPO expected to dominate October’s markets, the arrival of institutional secondary buyers at the voice layer of the AI stack is one more sign that the gap between “AI-native company” and “public company” is closing fast.

For an industry still arguing about whether agents are real, 15 million weekly conversations — refunded, renewed, booked, and resolved — are a reasonable rebuttal.