Generative media · United Kingdom (Polish founders)

ElevenLabs

A company that never left audio grew a product family through viral output, a developer API and a single credit wallet, reaching a half-billion-dollar annual revenue claim in four years; margin and retention are still undisclosed.

Verdict: Gaps remain

Updated: 2026-10-06

Profile

What it does: Text to speech, voice cloning, dubbing, sound effects, music and conversational agents (ElevenAgents). API, self-serve interface and enterprise sales draw on the same credit pool [5].
Founded: 2022, London. Operations now span fourteen cities: "Operations span 14 cities including London, New York, San Francisco, Warsaw" [1]
Founders: Mati Staniszewski (CEO) and Piotr Dąbkowski (CTO) [1]
Pricing model: Shared credit pool: Free 10k credits; Pro $99/600k credits; Scale $299/1.8M credits; Business $990/6M credits; Enterprise custom. "For V2 Multilingual models, 1 text character equals 1 credit." A dubbing minute costs 2,000 credits, a music minute 900 [5].
Funding: January 2024: $80M Series B (a16z, Nat Friedman, Daniel Gross); team from 5 to 40 [3]
30 January 2025: $180M Series C at $3.3B (a16z + ICONIQ) [2]
4 February 2026: $500M Series D at $11B (Sequoia); $781M total across five rounds [1]
May 2026: $100M secondary tender offer, the second in under a year [4]

Growth timeline

2024-01-22
$80M Series B, unicorn
"generated over 100 years of audio"; "41% of Fortune 500 companies"
[3]
2025-01-30
$180M Series C at $3.3B
"more than 250,000 conversational AI agents"; Voice Library "$2 million in payouts"
[2]
2025-12-10
CEO: voice models will commoditize within a couple of years
"voice models will be commoditized in just a couple of years"
[7]
2026-02-04
$500M Series D at $11B; 2025 closed above $330M ARR
"ElevenLabs closed 2025 with over $330 million in ARR"
[1]
2026-05-05
$500M ARR passed; 530 employees
"We ended 2025 with $350 million in ARR, and in the first four months of 2026, we have already surpassed $500 million ARR."
[4]

Growth loop

1Free-tier voice generation
2Shared output: video, audiobook
3New users and developers
4Enterprise agent deployment
5More credits, bigger Voice Library
Leak: commoditization risk: voice quality alone will not hold

The loop ran on voice quality; the company is now moving the same loop onto agent workflows, because its own founder expects voice models to commoditize.

The drawing shows the shape of the loop; each node is explained in the list below. The red arrow marks value leaking out of the loop.

Mechanism

Acquisition channels: Four channels stacked. Prosumer viral output: early users found unexpected uses in audiobooks and video; the founder on Sequoia's podcast: "we had one of those book authors copy paste his entire book inside this box, download it" [8]; Sequoia's summary: "from beta users copying entire books into their tiny text box to create audiobooks" [9]. Developer channel: the Series C announcement says developers had built more than 250,000 conversational agents [2]. Enterprise sales: the 2026 statement attributes growth to "enterprises deploying voice agents across their businesses"; Deutsche Telekom, Square, Revolut and the Ukrainian government are listed as customers [1]. Strategic investors: Deutsche Telekom, Salesforce, NVIDIA, Santander and KPN double as distribution doors [4]. (opinion) The investor list here is a legitimacy and channel list.
Loop: A creator or developer generates audio on the free tier, the output spreads online, new users arrive and some move to the API and enterprise use. People who upload voices to the Voice Library get paid; $2M had been paid out by January 2025 [2]. As the library grows, the product gets more valuable. From 2025 a second loop runs on enterprise agent deployment and conversation volume, which means credit consumption [4].
Activation: First generation with 10k free credits; one character, one credit [5]. (opinion) Tying the credit definition to characters removes price anxiety at first use.
Retention evidence: No official retention, churn or cohort figure. Sacra estimates the mix at roughly half self-serve, half enterprise (secondary data) [10]. Margin is not disclosed. No retention evidence.
Role of pricing: One credit pool ties every new product (music, dubbing, speech to text) to the same wallet; product expansion means more credits from the same customer. First-month discounts sit on the pricing page as a self-serve conversion lever [5]. (opinion) This model does expansion selling by itself as the product family grows.
Distribution moves: Voice Library marketplace (January 2024) [3], Dubbing Studio, ElevenAgents, Eleven Music and offices in fourteen cities [1]. The CEO describes the focus as "By staying laser-focused on audio" [9].
Team: Headcount went from 5 to 40 (2023) [3], to 120 (January 2025) [2] and to 530 (May 2026) [4]. (opinion) On the May 2026 figures that is roughly $0.9M of annual revenue per employee; a ratio computed from two official numbers, not a company claim.

Real reading

The five questions of the Growth Claim Check, applied to this company's public claims.

01
What was measured, and by whom?
yes
ARR, user counts and years of audio are company statements; no third-party audit.
02
Perception or outcome?
yes
ARR is an outcome metric; "1,000 years of audio" is volume, not value.
03
Is there a stability or quality metric?
no
Gross margin, NRR and churn are undisclosed; the CEO declines to answer the margin question.
04
Is there a counterfactual or cohort?
no
No control group or cohort disclosure.
05
Nominal or real?
unclear
Credit-based revenue is exposed to usage swings; two different official figures (330 and 350 M$) exist for the 2025 close.

Verdict

Gaps remain

The ARR cadence is consistent and comes from the official blog, but without margin and retention it cannot be called real. How much of the credit revenue recurs is unknown.

Test your own claim with the same five questions: Growth Claim Check →

Shadow side

The CEO's own forecast: voice models will be "commoditized in just a couple of years"; that is why the company is shifting to an agent platform [7]. (opinion) The engine that produced the growth and the future moat are different things.
The 2025 closing ARR differs across two official texts: "over $330 million" in February [1], "$350 million" in May [4].
Margin is not disclosed in any official text.

What it teaches a founder in Türkiye

This section is opinion; a reading, not a source.

01
(opinion) One domain, many outputs. The product family grew without leaving audio; in Turkey a narrow but deep vertical, say a Turkish-speaking agent, can be built on the same logic.
02
(opinion) Output is advertising. Viral videos and audiobooks became a channel without a marketing budget; making the output shareable is a design decision.
03
(opinion) A single credit wallet does expansion selling on its own. For a founder pricing in lira, credits can also be the one place where currency risk is managed.
04
(opinion) The investor list is a distribution list. Telecom, bank and software-giant investors were the first enterprise customers; corporate VC arms in Turkey can play the same role.
05
(opinion) Declare commoditization yourself. The founder saying "voice will commoditize" is the discipline of moving the moat on time. Test question: is today's edge still an edge in two years?

A question for you: remove this company's logo and drop the loop onto your own product. Which node stays empty?

Sources

Numbers are the references in the text. An 'opened' source was read by the researcher; 'secondary' ones are verified against their primary source before publication.

[1]
Series D
ElevenLabs blog · 2026-02-04 · opened
“ElevenLabs closed 2025 with over $330 million in ARR”
[2]
Series C
ElevenLabs blog · 2025-01-30 · opened
“developers to build more than 250,000 conversational AI agents”
[3]
Series B
ElevenLabs blog · 2024-01-22 · opened
“generated over 100 years of audio”
[4]
ElevenLabs crosses $500M ARR and welcomes new investors
ElevenLabs blog · 2026-05-05 · opened
“We ended 2025 with $350 million in ARR, and in the first four months of 2026, we have already surpassed $500 million ARR.”
[5]
Pricing
ElevenLabs · accessed 2026-10-05 · opened
“For V2 Multilingual models, 1 text character equals 1 credit.”
[6]
Mati Staniszewski: The Untold Story (episode page)
20VC · 2025 · opened
“ElevenLabs took 20 months to hit $100M ARR. 10 months to hit $200M ARR.”
[7]
ElevenLabs just hit a $6.6B valuation; its CEO says the real money isn't in voice anymore
TechCrunch podcast · 2025-12-10 · opened
“voice models will be commoditized in just a couple of years”
[8]
Training Data: Mati Staniszewski (episode page, summary)
Sequoia Capital · 2025-07-01 · opened
“And we had one of those book authors copy paste his entire book inside this box, download it.”
[9]
ElevenLabs' Mati Staniszewski: Why…
Inference by Sequoia · 2025 · opened
“from beta users copying entire books into their tiny text box to create audiobooks · By staying laser-focused on audio”
[10]
ElevenLabs profile (estimates)
Sacra · 2026 · opened
“a mix of self-serve and enterprise customers approaching a 50/50 split”

Open questions

The real URL of the September 2026 TechCrunch piece (returned 404 on research night).
Gross margin, NRR, churn: no official source at all.
2025 closing ARR: 330 or 350; it cannot be called "restated" without asking the company.
Current Voice Library payout total (the $2M figure is dated January 2025).
Deepfake and misuse incidents and the company's response; source not opened.
Verbatim sentences from the 20VC and Sequoia podcast transcripts.