The news
ElevenLabs, the voice AI company based in New York and London, said on September 30, 2026 that a $300 million employee tender offer values it at $22 billion. That doubles the ElevenLabs valuation of $11 billion set in its February 2026 Series D, and the company tied the jump to enterprise demand for conversational agents.
Wellington and T. Rowe Price led the tender, according to ElevenLabs. New investors include BDT & MSD, EQT, GIC, Goldman Sachs, OTPP and Sapphire Ventures, alongside existing backers such as Andreessen Horowitz, ICONIQ and Lightspeed. In a tender offer, existing shareholders such as employees sell shares to investors. TechCrunch reported that ElevenLabs ran a $100 million tender at a $6.6 billion valuation in September 2025.
The company shared growth multiples rather than total revenue. ElevenLabs said its agents now handle 15 million conversations a week, three times the February level, and that annual recurring revenue for ElevenAgents, its platform for building voice agents, has also tripled since February. Enterprise customers account for 55% of revenue, it said, and headcount is above 800. Named customers include Stripe, Deutsche Telekom, Admiral, Customers Bank and the governments of Ukraine and Greece. ElevenLabs also says its voice agents resolve issues 31% faster on average than chat agents; that is the company's own figure.
Rivals moved in the same week. On October 1, Microsoft (MSFT) released MAI-Transcribe-2-Streaming, its first streaming transcription model, priced at $0.54 per audio hour, SiliconANGLE reported. The model returns first transcript results in 320 milliseconds on average and supports more than 60 languages, according to the report. Microsoft also released MAI-Voice-2.1 and MAI-Voice-2.1-Flash speech models at $22 and $15 per million characters.
Also on October 1, Suno, best known for AI music generation, opened a public beta of Speech. Suno describes it as the first audio model that creates voice and music together in one track: users type text and describe the voice and music style they want. The announcement did not describe limits on imitating real people's voices.
On the data side, BleepingComputer reported on October 4 that Anthropic is asking Claude users to let it use their voice conversations to train its models. According to the report, the setting is off by default, users must choose to allow it, and they can switch it off or delete the data in Claude's privacy settings. It is separate from Claude's existing training controls for chat and coding sessions.
The numbers
- ElevenLabs valuation
- $22 billion (September 30, 2026)
- Employee tender size
- $300 million
- Prior valuation (Feb. 2026 Series D)
- $11 billion
- Weekly agent conversations
- 15 million, up 3x since February (company figure)
- Microsoft streaming transcription price
- $0.54 per audio hour
Why CEOs should care
For customer service and operations leaders, the price of testing voice automation keeps falling, and vendors are now quoting results. Treat those results as claims to check. ElevenLabs' 31% faster resolution is an average across its own customers; ask any vendor to measure resolution, transfer-to-human rates and error rates on your own call types in a time-limited pilot, and to show how calls are handed to a person when the agent fails.
For CFOs, compare per-hour and per-character list prices against your fully loaded cost per call, then add telephony, integration, monitoring and human review. Note that speed costs more: SiliconANGLE reported that Microsoft's earlier, non-streaming MAI-Transcribe-2 is priced at $0.10 per audio hour, against $0.54 for the new real-time model. Budget for the real-time tier only where callers are waiting on the line.
For CISOs, privacy officers and boards, voice recordings are sensitive personal data, and Anthropic's opt-in setting shows the direction of travel: explicit, reversible consent. Ask vendors whether they train on your customers' call audio, how long they keep it and whether your contract forbids it. Check what staff share through consumer voice features. And with realistic synthetic voices widely available, review any process that still uses a caller's voice as proof of identity.
The bigger picture
Voice is becoming a front door for software, and both specialists and platforms want it. ElevenLabs' valuation doubled in about eight months, and TechCrunch noted that employee tenders have become a common way for AI startups to keep staff. Microsoft building its own transcription and speech models adds a large platform to a market where ElevenLabs sells directly to enterprises. For buyers, more competition could mean lower prices, but also more vendors to vet on data handling.
What’s next
Watch whether ElevenLabs discloses total revenue rather than growth multiples, where Microsoft puts its new voice models to work in its own products, whether Suno adds rules on voice imitation as Speech leaves beta, and whether Anthropic publishes retention terms for the voice data users choose to share.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
