How ElevenLabs Turned AI Voice Into a Product

The short version

ElevenLabs is now a $11 billion voice AI company with $500 million in annual recurring revenue, 530 teammates across 50+ countries, and a voice catalog used by the likes of NVIDIA, Salesforce, Deutsche Telekom, Klarna, Revolut, Meta, Epic Games and 41% of the Fortune 500. The company is best understood as a specialist audio platform, not a general AI assistant: text-to-speech, speech-to-text, voice cloning, dubbing, music, and now conversational agents all share one research base. If you need realistic, multilingual, production-grade voice, ElevenLabs is the most credible option on the market as of July 2026.

Why this launch story matters

I keep coming back to ElevenLabs because it is one of the few AI-native companies that turned a research demo into a real product business without losing the research edge. Most voice AI tools I have profiled either got stuck at the demo stage, defaulted to a single use case, or got absorbed into a big-tech assistant. ElevenLabs did the opposite: it expanded the surface area (TTS, STT, music, dubbing, agents), kept model quality high, and built both an enterprise platform and a creator economy on top.

If you are evaluating voice AI in mid-2026, this is the company to benchmark against. Every serious alternative—OpenAI Voice, PlayHT, Hume, Resemble, Cartesia—is measured against what ElevenLabs ships.

The founding story, in one paragraph

ElevenLabs was founded in 2022 by Mati Staniszewski (CEO) and Piotr Dąbkowski (CTO), two Polish childhood friends who left Palantir and Google, respectively, to build a better text-to-speech engine (Wikipedia). The company name is a wink at Polish Independence Day on 11 November. Their motivation, Staniszewski told Forbes, was the uniquely Polish horror of badly dubbed American films and the cold monotone of the lektor narration style (Forbes). What started as an experiment in expressive Polish dubbing turned into a global voice platform within four years.

Funding timeline at a glance

ElevenLabs has raised $781 million across five disclosed rounds, with a sixth private employee tender offer in 2025. Here is the verified sequence:

Round	Date	Amount	Lead investors	Post-money valuation
Pre-seed	Jan 2023	$2M	Credo Ventures, Concept Ventures	undisclosed (Wikipedia)
Series A	Jun 2023	$19M	a16z, Nat Friedman, Daniel Gross	~$100M (VentureBeat)
Series B	Jan 2024	$80M	a16z, Nat Friedman, Daniel Gross, Sequoia	$1.1B (ElevenLabs)
Series C	Jan 2025	$180M	a16z, ICONIQ Growth	$3.3B (Tech.eu)
Tender offer	Sep 2025	employee secondary	n/a	$6.6B (Bloomberg)
Series D	Feb 2026	$500M	Sequoia (lead), a16z (4x), ICONIQ (3x), Lightspeed, Bond, Evantic	$11B (Sifted, CNBC)

A third Series D close in May 2026 added BlackRock, Wellington, D.E. Shaw, Schroders, NVIDIA (NVentures), Santander, Deutsche Telekom (T.Capital), Jamie Foxx, Eva Longoria, and Hwang Dong-hyuk, alongside a $100 million employee tender (ElevenLabs blog). Forbes estimated the cofounders were both worth roughly $1 billion each as of late 2025 (Forbes).

Revenue and traction in 2026

ElevenLabs ended 2025 with over $330 million in ARR. By May 2026, that figure had crossed $500 million ARR (ElevenLabs blog; also confirmed by CNBC at the Series D). Two things drive that growth:

1. ElevenAgents for enterprise customer operations. Revolut reported 8x faster ticket resolution and over 99.7% successful call completion after deploying ElevenLabs Agents (Revolut case study). Klarna reported 10x faster resolution for its 35 million US customers, with plans to expand to its full 114 million base (Klarna case study).
2. Voice creator payouts in the Voice Library marketplace, which doubled from $11M to $22M earned by 10,400+ creators in six months (Voice creators post). The Voice Library now spans 32 languages.

By Forbes’s late-2025 profile, the revenue split was roughly 50/50 enterprise vs. creator (Forbes). Cofounder Staniszewski told Sifted the company expects that mix to shift to 60/40 by end of 2026 and 70/30 by end of 2027. Forbes also estimated ElevenLabs was profitable at roughly a 60% margin in 2025, unusual for an AI startup of that size.

What the platform actually is in 2026

ElevenLabs now sells three product surfaces that all share the same in-house research models (ElevenLabs):

• ElevenCreative — studio for creators, marketers, and publishers. TTS, music, sound effects, voice design, dubbing, image, and video in one editor.
• ElevenAgents — enterprise platform for voice and chat agents with testing, analytics, guardrails, workflows, and PCI/HIPAA-ready compliance.
• ElevenAPI — low-latency developer APIs for TTS, STT, music, and dubbing used by Meta, Epic Games, Salesforce, MasterClass, Harvey, and others to power experiences reaching over one billion end users (Series D post).

The model lineup as of mid-2026 reads like a research roadmap:

1. Eleven v3 (GA Feb 2026) — most expressive model, 70+ languages, audio tags like [whispers] and [sarcastically], multi-speaker dialogue (Eleven v3).
2. Eleven Multilingual v2 — most consistent lifelike speech across 29+ languages.
3. Eleven Flash / Turbo v2.5 — sub-100ms latency models for real-time agents.
4. Scribe v2 (Jan 2026) — batch STT with keyterm prompting and entity detection.
5. Scribe v2 Realtime (Nov 2025) — most accurate real-time transcription (Artificial Analysis ranked it #2 by WER at 2.2% in July 2026).
6. Music v1 (Aug 2025) and Music v2 (May 2026) — licensed-data music generator with vocals (Music v2).
7. Dubbing v2 (May 2026) — emotion-preserving translation across 90+ languages (Dubbing v2).
8. ElevenAgents Expressive Mode (Feb 2026) — emotional turn-taking for support agents (Expressive Mode).

“We are one of the very few companies that are ahead of OpenAI—not only on speech, but speech-to-text and music. That’s hard.” — Mati Staniszewski, co-founder and CEO (Forbes)

ElevenLabs pricing tiers (July 2026)

ElevenLabs sells a shared credit pool that powers every product. Below is the verified July 2026 plan ladder (ElevenLabs pricing):

Plan	Monthly price	Credits	Notes
Free	$0	10,000	TTS, STT, Sound Effects, Voice Design, Music, Productions, Image; 3 Studio projects
Starter	$5–$6 (annual: $5/mo)	30,000	Commercial license, Instant Voice Cloning, Music commercial use, Dubbing Studio
Creator	$11 first month, then $22 ($18.33/mo annual)	121,000	Professional Voice Cloning, additional credits
Pro	$99 ($82.50/mo annual)	600,000	44.1kHz PCM API output, 192kbps audio
Scale	$299 ($249.17/mo annual)	1.8M	3 seats, team collaboration, 3 Professional Voice Clones
Business	$990 ($825/mo annual)	6M	10 seats, 10 Professional Voice Clones, low-latency TTS from 5¢/min
Enterprise	Custom	Custom	DPAs/SLAs, BAAs for HIPAA, SSO, elevated concurrency, fully managed dubbing

A few practical notes I keep in mind:

• Unused paid credits roll over for up to 2 months (cap at 3x monthly quota), but the Free plan does not roll over.
• 1 TTS character ≈ 1 credit on V2 Multilingual; Flash and Turbo models are discounted to 0.5–1 credit per character.
• STT is 330 credits/minute, Music is 900 credits/minute, Dubbing is 2,000–10,000 credits/minute depending on watermark and Studio vs. automatic.
• Annual billing = 2 months free.

For heavy creators, the math rarely works on Starter; most podcasters and audiobook producers I see move to Creator or Pro.

Voice cloning policies, ethics, and consent

This is the part I will not gloss over. ElevenLabs treats voice cloning as a regulated capability, not a novelty. From the Safety page and the elections post, here are the safeguards in place as of July 2026:

1. No Go Voices (NGV). Uploaded samples are screened against a proprietary list of high-risk voices (politicians, candidates, celebrities, widely recognized figures). Cloning attempts are blocked.
2. Voice CAPTCHA. Uploaded samples are compared against a live recording of the user reading a dynamically generated prompt, ensuring the person cloning is the person speaking.
3. Professional Voice Cloning is gated behind identity and consent verification, with seven full-time human moderators plus automated systems scanning for child safety, hate speech, scams, and fraud (Forbes).
4. AI Speech Classifier. Free public detector since 2023 that returns a probability score for whether a clip was generated on ElevenLabs.
5. SynthID partnership. As of June 2026, ElevenLabs content is watermarked with Google DeepMind’s SynthID, so AI audio is identifiable downstream even after re-encoding.
6. Content Authenticity standards. ElevenLabs is a member of the C2PA and Content Authenticity Initiative.
7. Voice Library marketplace lets verified creators set their own licensing terms, with revocation controls and a notice period of up to two years (Voice creators).

The company also disclosed it received only 5 government information requests globally in 2025 and operates a designated single point of contact for EU authorities under the Digital Services Act (Safety).

If you want my opinion: do not treat voice cloning as a toggle. If a voice resembles a real person, get consent in writing, retain proof, and disclose AI use where your audience might be misled. Technical capability is not permission.

The Biden robocall and what changed

The most-cited misuse incident was the January 2024 New Hampshire Democratic primary, where an AI-generated robocall impersonating President Biden told voters to skip the primary. Pindrop Security analyzed the audio and concluded with over 99% confidence that it was made with ElevenLabs’ technology; UC Berkeley’s Hany Farid independently reached the same conclusion (Wired, Pindrop).

The fallout pushed ElevenLabs to formalize the No Go Voices list, expand Voice CAPTCHA, and grow its moderation team. By July 2026, the company has formalized MoUs with Brazil’s Superior Electoral Court for the October 2026 election, joined California’s Countering Tech-Enabled Fraud Task Force, and partnered with deepfake detection firm Reality Defender (Elections post).

ElevenLabs vs OpenAI Voice, PlayHT, Hume, Resemble, Cartesia

I get asked about this constantly, so here is the head-to-head I would defend in July 2026.

Capability	ElevenLabs	OpenAI Voice	PlayHT	Hume	Resemble	Cartesia
Realistic TTS quality	Best-in-class for expressive, multilingual	Strong, general-purpose	Good for English-first	Optimized for emotional prosody	Strong for enterprise cloning	Strong for ultra-low latency
Languages	70+ (TTS), 90+ (Dubbing v2)	~50 (model dependent)	140+ voices, fewer true multilingual models	Limited catalog	60+ languages	English-first
Cloning controls	Voice CAPTCHA, NGV, marketplace licensing	Restricted, account-gated	Per-account consent	Limited	Strong enterprise IP controls	API-level controls
Speech-to-text	Scribe v2 at 2.2% WER (top 3 globally)	GPT-4o Transcribe ~4.0% WER	Limited STT	Limited STT	Yes	Yes
Conversational agents	ElevenAgents + Expressive Mode	Realtime API in ChatGPT	Yes	EVI (emotional voice interface)	Yes	Sonic (sub-100ms)
Music / SFX	Music v2, SFX	Limited	No	No	No	No
Pricing (TTS)	$0.17/min on Pro, lower at scale	Lower base price	Mid-range	Mid-range	Enterprise quote	Enterprise quote
Notable customers	NVIDIA, Salesforce, Deutsche Telekom, Klarna, Revolut, Meta, Epic	ChatGPT users	SMB creators	Research and CX apps	Enterprise call centers	Robotics, gaming

Where ElevenLabs wins, in my view:

• Expressive quality for narration, audiobooks, and dubbing is still the benchmark. Forbes reports the company charges up to 3x what US rivals do because quality holds up (Forbes).
• Full audio stack under one roof: TTS, STT, music, SFX, dubbing, agents. Few competitors cover that surface.
• Voice Library is a real creator economy with $22M paid out and verified consent built in.

Where ElevenLabs loses or compromises:

• Latency for pure real-time agents is still best handled by Flash/Turbo variants or competitors like Cartesia Sonic.
• Price is higher than OpenAI Voice or open-weight models for raw minutes.
• TTS-only shops may prefer PlayHT for cost or Resemble for IP controls.

Notable customers and 2026 partnerships

ElevenLabs publishes a long customer list on its homepage; I pulled the ones that show up in primary reporting (ElevenLabs, Series D post, Revolut, Klarna):

• Enterprise agents: Revolut (UK + Europe rollout, 8x faster resolution), Klarna (US phone support, 10x faster), Deutsche Telekom (Europe’s largest telco, customer support + in-network AI agent), Square, Cars24 (India), Meesho, Deliveroo, City of Midland, Freedom Forever, TELUS Digital.
• Public sector: Ukrainian government (“first agentic government”), Brazil’s Superior Electoral Court (October 2026 election voice assistant), UK Government MoU, Poland (state investment + EU presidency multilingual diplomacy), Greece, California Countering Tech-Enabled Fraud Task Force.
• Tech platform deals: NVIDIA (multilingual marketing + ACE at Computex), Salesforce (developer voice), Meta, Epic Games (Darth Vader in Fortnite, with James Earl Jones’ estate consent), MasterClass (AI instructors), Harvey (legal multilingual), Cisco (Webex), Twilio (Conversation Relay), Chess.com (iconic voices).
• Media and publishers: TIME, Bertelsmann, HarperCollins, The Washington Post, Storytel, Reuters, Curio.
• Creative icons and investors: Sir Michael Caine (Iconic Marketplace), Matthew McConaughey (Lyrics of Livin’ + investor), Jamie Foxx, Eva Longoria, Hwang Dong-hyuk, Liza Minnelli, Art Garfunkel (all confirmed as investors in the May 2026 Series D close, ElevenLabs blog).

Founders’ stated playbook

Both cofounders stay unusually visible for an audio AI company. A few quotes I keep coming back to:

• On discipline: “Having a ton of compute can be a curse because you don’t think how to solve it in a smart way.” — Piotr Dąbkowski (Forbes).
• On category ambition: “Every major enterprise will communicate with its customers and audiences through AI agents. ElevenLabs has built the technical leadership and commercial traction to define the category.” — Rob Mazzoni, Wellington Management (ARR post).
• On safety: “AI safety is inseparable from innovation at ElevenLabs. Ensuring our systems are developed, deployed, and used safely remains at the core of our strategy.” — Mati Staniszewski (Safety).
• On trajectory: “We started by building a voice that could sound human — and we did. Yet we stay hungry, knowing how early this space still is, as we build toward IPO and beyond.” — Mati Staniszewski, Series D announcement (Series D post).

The “build toward IPO” line is the one to watch. A company that already runs profitable, raised at $11B, and is doubling ARR every year has a credible public-market path.

Recent 2026 product launches (timeline)

I compiled this from ElevenLabs’ blog index and the homepage research timeline. Every entry is verified:

1. Jan 9, 2026 — Scribe v2 launches with lowest WER on industry benchmarks.
2. Jan 14, 2026 — Deutsche Telekom partnership announced.
3. Jan 28, 2026 — Revolut selects ElevenAgents for voice support in UK + Europe.
4. Feb 2, 2026 — Eleven v3 moves from alpha to generally available.
5. Feb 4, 2026 — Series D closed at $11B valuation, $500M led by Sequoia.
6. Feb 10, 2026 — Expressive Mode for ElevenAgents ships with Eleven v3 Conversational and a new turn-taking system.
7. Feb 11, 2026 — Klarna case study: 10x faster US phone resolution.
8. Feb 11, 2026 — ElevenLabs for Government launches.
9. Mar 19, 2026 — Music Marketplace launches.
10. Mar 31, 2026 — Polish EU Council presidency multilingual diplomacy deployment.
11. May 5, 2026 — Series D third close: $500M ARR, BlackRock, NVIDIA, Foxx, Longoria.
12. May 26, 2026 — Music v2 launches with better vocals and instrumentation.
13. May 28, 2026 — Dubbing v2 launches with emotion-preserving translation across 90+ languages.
14. Jun 8, 2026 — UK Government MoU and London HQ expansion.
15. Jun 18, 2026 — Poland invests in ElevenLabs.
16. Jun 22, 2026 — California expansion: 173 high-paying jobs.
17. Jun 25, 2026 — SynthID watermark partnership with Google DeepMind.
18. Jun 30, 2026 — Procedures in ElevenAgents (structured agentic workflows).
19. Jul 7, 2026 — ElevenLabs launches in Canada.
20. Jul 24, 2026 — “Strengthening and protecting elections” policy paper.

Roadmap signals for the rest of 2026

Reading between the lines of the Series D post, the ARR announcement, and the elections paper, here is where the company is heading:

1. ElevenAgents as the lead product. The $500M Series D post explicitly says the round is “doubling down on ElevenAgents and conversational voice models” (Series D post).
2. ElevenCreative converging audio + image + video. Image and Video launched in November 2025 and is now woven into the creative editor.
3. More government and civic deployments. Brazil’s TSE, Ukraine, Poland, UK, Greece, and US state partners point to a sustained push on civic AI.
4. Creator economy expansion. Music Marketplace follows the Voice Library playbook; expect ElevenLabs to keep doubling creator payouts.
5. Latency and reliability improvements. Expressive Mode plus Scribe v2 Realtime plus Eleven v3 Conversational are the building blocks of agent-grade voice.
6. IPO readiness. Staniszewski has now used the phrase “build toward IPO” in the Series D and ARR posts.

Limitations and where it still struggles

I want to flag the rough edges so you don’t buy on hype alone:

• Credit math is opaque. Heavy producers on the Creator plan can burn through 121,000 credits faster than expected once they start dubbing, music generation, and STT together.
• PVC quality vs. v3 — Professional Voice Clones are not yet fully optimized for Eleven v3 (Eleven v3 docs). Use Instant Voice Clones or designed voices when you need v3 expressiveness.
• Free tier commercial restrictions — verify plan terms before publishing monetized content.
• Audio still needs editorial review. ElevenLabs voices are excellent, but proper nouns, brand names, and pacing still need a human pass.
• Audiobook lawsuit — ElevenLabs settled an audiobook narrator training-data lawsuit (Vacker and Boyett) out of court in November 2025 (Forbes). Training-data provenance will remain a watch-item across the industry.

Who ElevenLabs is best for in 2026

Based on the product surface, customer mix, and pricing:

1. Enterprise CX and support teams that need multilingual, low-latency voice agents with PCI/HIPAA-ready compliance.
2. Audiobook narrators and publishers producing backlist titles at scale, especially multilingual.
3. YouTubers and creators localizing video content via Dubbing v2 across 90+ languages.
4. Studios and broadcasters using ElevenProductions for high-end dubbing with human translators.
5. Game studios that need character voices across archetypes (Epic Games uses it for Fortnite).
6. Hobbyists and authors who want an audiobook under $100 in credits via the Voice Library marketplace.

If you only need a single English voice for short clips, the value proposition is thinner — OpenAI Voice or a fine-tuned open-weight model may be cheaper.

Related on AIUncovers

• ElevenLabs profile
• ElevenLabs alternatives
• Voice content production workflow
• AI voice cloning policy guide
• Conversational AI agents buyer guide

Responsibility note

AIUncovers does not treat voice cloning as a novelty feature. If a voice resembles a real person, rights and consent are mandatory. Technical capability is not permission. ElevenLabs’ safety stack (No Go Voices, Voice CAPTCHA, SynthID watermarking, AI Speech Classifier) is meaningful, but the legal and ethical responsibility still sits with whoever publishes the audio.