$11B ElevenLabs Triples Valuation, Opens First Government Voice Sales Desk
ElevenLabs Puts a Closer in D.C.
ElevenLabs launched a government-specific voice AI initiative and opened its first dedicated government sales roles. At the same time, telecom partner Deutsche Telekom embedded its voice agents into network infrastructure, pulling the company’s synthetic voice stack out of consumer demos and into public agencies and carrier networks.
On February 11, 2026, the company opened ElevenLabs for Government, its first dedicated push to sell voice and chat agents to ministries, agencies, and local governments. The agents aim to replace rigid phone menus and long hold times. Zero G Talent’s board data shows the new role — Account Executive Lead, North America, Government — based in Washington, D.C., among 38 ElevenLabs jobs posted in the past seven days. The listing sits beside other recent ElevenLabs hires, proving the go-to-market motion is live, not a slide-deck promise.
Money backs the seriousness. Adwaitx reported that the company closed a $500 million Series D led by Sequoia Capital in February 2026 at an $11 billion valuation, up from $3.3 billion a year earlier. That tripling in thirteen months funded a Washington sales desk.
| Round | Date | Valuation | Raise |
|---|---|---|---|
| Series C | Jan 2025 | $3.3B | $180M |
| Series D | Feb 2026 | $11B | $500M |
The company did not enter the public sector cold. Its blog says it already serves the Governments of Ukraine and the Czech Republic, plus U.S. localities like Midland, Texas. Ukraine signed a memo to weave ElevenLabs agents through ministries of education, health, and economic development. Danylo Tsvok, Ukraine’s minister of AI, showed those initiatives at ElevenLabs’ London summit in February 2026. Mati Staniszewski, CEO and co-founder, demoed a fictional government workflow there.
The product meets public-sector compliance: SOC 2 Type 2, GDPR, CPRA, HIPAA. Forward Deployed Engineering teams embed with agencies during setup. The conversational agents listen, read, and reply like humans across phone, chat, email, and WhatsApp. Expressive Mode, running on the Eleven v3 model, reads tone, pacing, stress, and intent, not just words. Beyond FAQs, a “digital twin” lets leaders broadcast official messages in dozens of languages while keeping their own voice (aib.vote/en/news, 2026-04-01).
In Midland, the city replaced voicemail and deep IVR trees with a voice-first concierge named “Jacky” that speaks English and Spanish. In the Czech Republic, the deployment answers about 5,000 calls a day and resolves six in seven without a human. Those are operating numbers, not pilot metrics.
Why now? Citizen frustration with unanswered calls pushes the timing. A January 2026 Google Public Sector study found nine in ten U.S. federal agencies already use some AI. A global index from the same year showed three in four public servants adopt AI, yet fewer than one in five believe their government uses it effectively. ElevenLabs pitches its voice agents as the missing layer that makes those adoptions feel responsive.
The new AE lead owns the North America government GTM across federal, state, and public agencies. The role also defines long-term public-sector strategy and represents ElevenLabs at government events. That is a builder job, not a quota-taker.
Voice synthesis alone won’t win agencies. The hire signals ElevenLabs expects procurement cycles, not app-store downloads, to drive its next growth curve. The sales lead walks into offices where pilots already answer thousands of calls a day; the product shipped before the salesman did.
When Does a Voice Model Become Infrastructure?
Deutsche Telekom removed the app from AI voice assistance. At Mobile World Congress 2026, the carrier unveiled the Magenta AI Call Assistant, a ElevenLabs-powered agent embedded at the network level that answers “Hey Magenta” on any calling device, legacy landlines included. The integration puts voice agents inside the telephone network rather than on a screen — a structural shift that answers the same question the government buildout raised: when does a voice model stop being a consumer toy and start being infrastructure?
The telecom move predates the Washington hire. Deutsche Telekom backed ElevenLabs in its earlier funding round and started building voice agents for customer service months before a prior deployment this year. What changed at MWC 2026 is depth: system integrator Radisys moved the assistant from a customer-service overlay to a network-native function inside the call path.
A user on a Deutsche Telekom line in Germany says “Hey Magenta” mid-call; the agent can translate live, pull calendar slots, or query a map, as Wired reported from the demo. The carrier says first services include live translation, call summaries, and open-domain questions. Support for up to 50 languages is planned over the next year. Because the feature lives in the network, it needs no download, no smartphone, and no setup. Abdu Mudesir, Board Member for Product & Technology at Deutsche Telekom, said at the event: “We are bringing it directly into our networks, products, and services – scalable, secure, and with clear added value for people and the economy.” He called the move a step toward an “AI-first telco that sets the standard.”
That standard is carrier-grade by definition. A telecom operator controlling the voice agent at the switch can enforce opt-in consent, EU data rules, and reach that an app developer cannot match. Deutsche Telekom says both parties on a call must agree to usage, voice recordings aren’t saved, and the design complies with GDPR. The carrier characterized complexity, exclusivity, and device dependency as the main barriers to AI adoption, now removed. The announcement places Deutsche Telekom at the center of a race among operators to embed intelligence into communication infrastructure instead of confining it to standalone apps.
The embedding also surfaces hard problems. Avijit Ghosh, a technical AI policy researcher at Hugging Face, told Wired he worries about injecting an always-listening assistant into non-encrypted phone calls and cited accent bias in ElevenLabs’ synthetic voices. Spain’s data protection authority published a 71-page guide in February 2026 on agentic AI privacy risks; the Dutch authority issued a warning that same month against open-source agents with full communications access. Deutsche Telekom’s consent model could become a reference for regulation, but the tension between network-wide reach and individual privacy remains.
The proof point for ElevenLabs’ voice agents as infrastructure is not the government sales role in Washington, but the agent already riding the German telephone network. Over the next year, as language count climbs to 50 and mid-call restaurant bookings go live, the call itself becomes a compute surface. The carrier that owns the switch owns the agent.
The Price War for Speech
OpenAI slashed real-time voice API pricing by roughly two-thirds to nearly nine in ten in late 2024, resetting the cost floor while ElevenLabs secured its billion-dollar raise. The drops didn’t stall the voice specialist; they validated voice agents as a distinct market. Hyperscalers and open-source labs now treat carrier-grade voice as a price battleground, not just a quality contest.
The first salvo came December 17, 2024, when OpenAI launched GPT-4o and GPT-4o Mini on its Realtime API. DeepNewz reported the new structure put Mini at about $0.90 per hour, a tenfold decrease, and cut audio token pricing by 60% to $2.50 per million tokens. That made advanced voice accessible to developers who previously balked at cost. A commenter on the release summed the shift: “Finally we might see some actual apps since its not prohibitively expensive.”
OpenAI followed with general availability of the gpt-realtime model. SoftReviewed documented per-call pricing from $0.11 for a one-minute chat to $2.68 for ten minutes, with Mini at $0.16–$0.33 per minute depending on system prompts. Later meters broke audio and text tokens into fractions of a cent to cents per minute, as OpenAI’s docs state (https://developers.openai.com/api/docs/guides/realtime-costs).
The table below compares published voice API cost structures as of late 2025:
| Provider | API / Model | Cost basis | Price | Note |
|---|---|---|---|---|
| OpenAI | Realtime Mini | per minute audio | $0.16–$0.33 (Aug 2025) | system prompt doubles cost |
| OpenAI | Realtime standard | audio in/out | $0.06 in + $0.24 out / min (Oct 2025) | plus text tokens |
| Live API | token-based | not disclosed | multilingual excellence | |
| Cerebrium + Rime | combined stack | ~60% below OpenAI | variable | better price performance |
| MiniCPM-o | open-source | per minute | $0.01 (Aug 2025) | ultra-low cost option |
| ElevenLabs | TTS / voice | per character | competitive (Oct 2025) | pure synthesis quality |
Google’s Gemini Live API entered with token-based multilingual support. Open-source alternatives pushed floors lower: MiniCPM-o at a penny per minute, Moshi from Kyutai, and Ultravox built on LLaMA. Cerebrium with Rime claimed roughly 60% savings versus OpenAI. The voice AI market is set to expand from about $3 billion in 2024 to roughly $48 billion by 2034, pulling adoption forward — almost all customer service interactions were predicted to involve AI agents by 2025.
Hyperscalers treat voice as one capability among hundreds. Falling API prices usually compress startup raises, yet ElevenLabs’ number climbed. The resolution sits in contract type: government and carrier deals price on reliability and integration, not per-minute token burns.
OpenAI’s board footprint on Zero G Talent shows 573 roles with a median salary of $335k and 71 added in the past week, evidence the rival is staffing to win enterprise voice beyond price. ElevenLabs posted its government lead among the same week’s listings. The hiring parallel confirms both sides view voice as infrastructure, not experiment.
Procurement teams evaluating ElevenLabs against OpenAI should pull live Realtime API rates from OpenAI’s developer page and benchmark against per-character quotes before signing agency contracts. The price war is settled for developers; for agencies, the negotiation starts now.
A Translator Between Buyers and Code
ElevenLabs’ posting for the Account Executive Lead, North America, Government demands eight-plus years selling to government and a record of closing complex deals in regulated environments. That separates it from the company’s consumer voice work. The role, flagged among last week’s postings, is the clearest signal of what the government buildout requires: a seller who can also speak engineering.
The profile is hybrid. The lead must work closely with Product and Engineering to shape solutions meeting security, compliance, and deployment needs for federal, state, and public agencies. A public-sector AE in the Middle East carries the same burden: own complex engagements end-to-end, managing procurement cycles, tenders, RFPs, security and data-residency reviews, and multi-stakeholder approvals alongside Engineering. The salesperson becomes the interface between bureaucratic buyers and the technical staff who ship voice agents.
Experience thresholds are steep. The North America spec asks for eight-plus years in public-sector sales and proven closure in regulated settings. The broader public-sector posting wants ten-plus years in SaaS or enterprise sales, plus demonstrated deals through Middle East procurement. Both require building a vertical from early stages — not a rep stepping into a mature book. ElevenLabs calls it a “builder mindset: establishing our presence in the market while helping shape the long-term strategy for the government vertical.”
Compliance weight makes the profile distinct from consumer roles. The government AE must translate data sovereignty, cloud and cybersecurity frameworks, procurement law, and localization mandates into deal structures. For the Middle East posting, professional Arabic is mandatory. The North America role prefers basing in Washington D.C., remote-first notwithstanding, to sit near procurement officials.
Daily tasks show the hybrid. The lead will represent ElevenLabs at government events, build trusted relationships with senior public leaders, procurement officers, and technical stakeholders, and partner with Legal, Solutions Engineering, and Customer Success to ensure implementation. Missing the engineering half breaks the sale: a voice agent that fails a security review or cannot reside in-country is dead on arrival.
Contrast this with other ElevenLabs listings pulled this week — Account Executive for ElevenCreative in Spain, Enterprise AE for France & Benelux, mid-market roles in Spain. Those target commercial creativity and regional business users. They don’t ask for procurement cycles or data-residency reviews. The government profile is a different species.
The skill set boils down to non-negotiables: deep public-sector expertise with years closing in regulated environments; ability to build a GTM motion from zero; direct collaboration with engineering on security and compliance; fluency in procurement mechanics like RFPs and tenders; and regional grounding, whether in D.C. or Arabic fluency.
ElevenLabs’ career page states the company has raised $781M, but the government hire isn’t about that number. It’s about installing a translator between agency buyers and the voice models the company builds.
The candidate who fills the Washington desk will write the first playbook for selling carrier-grade voice into the public sector. That document, not the launch press release, will determine whether agencies adopt conversational AI or stall in pilot limbo.
Why the Creator Tiers Don’t Matter Here
ElevenLabs’ consumer plans run from free to $990 per month as of July 2026, a ladder built for solo creators and app developers, not federal procurement officers. This section draws a line around our government thesis. The dedicated government launch and the North America AE role sit outside the broad consumer voice synthesis market and apart from generic cloud speech APIs that Google, Microsoft, Amazon, and other hyperscalers sell to any enterprise.
Founded in 2022 and headquartered in New York City, ElevenLabs became a leading provider of hyper-realistic AI voice generation serving creators, developers, and enterprises worldwide, per Costbench’s June 2026 profile. The platform covers text-to-speech, voice cloning, speech-to-text, sound effects, music, dubbing, and conversational agents under one credit system. That breadth is exactly what we fence off. The government initiative targets agencies with voice and chat agents to replace rigid menus — a targeted product line with its own sales lead, not the all-you-can-clone creator toolbox.
Consumer tier mechanics
Consumer usage follows a credit-based model with seven published tiers. Costbench listed monthly prices after June 2026 adjustments:
| Plan | Price (July 2026) | Key limit |
|---|---|---|
| Free | $0 | 10 min audio, no commercial rights |
| Starter | $6 | Min for commercial use, 1 instant voice clone |
| Creator | $22 | 1 pro + 3 instant clones, single user |
| Pro | $99 | Single user |
| Scale | $299 | Multi-seat workspaces |
| Business | $990 | Multi-seat, highest tier cut from $1320 |
| API Pro | $99 (separate) | 100 credits, independent of UI |
Instant Voice Cloning needs one to five minutes of audio on Starter and up; Professional Voice Cloning needs 30 minutes minimum on Creator and above, Costbench wrote. Multi-seat workspaces appear only on Scale and Business. Free and Starter credits expire monthly; rolled-over credits on higher tiers vanish if you cancel, with no refund. As of mid-2026, ElevenLabs support is primarily AI-automated, with users reporting difficulty reaching a human. A Hacker News commenter cited by Costbench called the service “400 times more expensive than the rest” for some uses.
The table shows the consumer face. Conversational AI agents bill at ten cents per minute of conversation on top of text-to-speech credits, and real-world credit burn runs one and a half to two times raw character counts, ElevenLabs Magazine reported in April 2026. Thirteen documented hidden costs sit beyond list price, including opaque credit pricing that can add 20–50% to expected license costs and enterprise API underestimation of 25–75%, per Costbench. A creator on the $22 plan gets roughly one hour of audio — cheaper than a $150–$500 human voice actor hour, but that math means nothing to an agency buyer negotiating HIPAA compliance add-ons at $1,000 per month.
Hyperscaler voice utilities
Hyperscalers take a different tack. Google Cloud TTS charges about $0.016 per 1,000 characters, Azure TTS about $15 per million characters, and Amazon Polly $4 per million for standard voices, $16 for neural, per pricing guides from ElevenLabs Magazine and AIToolPick in May 2026. These are pay-per-use utilities inside broader cloud stacks. Business20channel noted in June 2026 that Google, Microsoft, Nvidia and others pushed streaming APIs and on-device processing, but their voice features remain one capability among hundreds. The carrier-grade embedding with Deutsche Telekom described earlier is a motion no hyperscaler cloud TTS sale replicates.
For enterprise and developer customers, the pricing is becoming more granular — separate model tiers, burst capacity pricing for agents, and API versus UI pricing differentiation are all signals of a platform evolving toward infrastructure-grade pricing.
ElevenLabs Magazine wrote that line in April 2026, and it captures the split we mark. Consumer side consolidates: Studio 3.0 bundles TTS, music, SFX, and video editing under one credit pool. The enterprise and government offering instead prices on reliability and integration, as seen in the carrier and agency deals. The Washington seller does not pitch the $6 Starter tier. He sells infrastructure uptime and compliance, not instant voice clones from a five-minute sample.
Play.ht and Murf undercut ElevenLabs on character limits — $39 for 500k chars, $29 for 480k chars — but rate lower on naturalness, AIToolPick and Scored.tools noted. Those consumer-grade alternatives confirm the market outside government is a price-per-character brawl. The government voice bet escapes that brawl.
A federal agency replacing phone menus does not care that a creator plan yields roughly one hour of audio for $22 versus a $150–$500 human voice actor hour. It cares that the voice agent runs inside a telecom network with measured latency. Our analysis therefore excludes creator plans, hyperscaler APIs, and cheaper clones. The government GTM is a different product with a different buyer, and the sales org reflects that split.
Working in AI? Zero G Talent tracks the openings: see every open OpenAI role, browse AI jobs, openings at ElevenLabs, and the people building the field.