ElevenLabs' $22B Tender Meets OpenAI's Free Mobile Voice — The Voice Interface Is Being Decided Without Consumers Paying for Either

ElevenLabs is in early talks for a tender offer that would value the company at roughly $22 billion, Bloomberg reported on July 2, 2026 — doubling the $11 billion mark set five months earlier in a Sequoia-led $500 million Series D and doubling the $6.6 billion employee tender ElevenLabs ran just nine months before that. The deal is structured as a secondary share sale expected to close by September 2026, so the $22 billion is the price of optionality on existing shares, not new capital into the company. That detail matters: a tender at this level is a verdict from investors on what the next primary round has to clear, not a discounted cash-flow exercise.

The other trade is OpenAI's: give the same capability away for free on the device people already own. GPT-Live, the full-duplex voice model that now powers ChatGPT Voice, rolled out globally on July 8, 2026, with GPT-Live-1 mini as the default for Free users. Then on September 23, 2026, OpenAI extended Voice into Gmail, Google Calendar, and Slack via plugins, with ChatGPT Work event-triggers wired in for Plus and Pro subscribers. The free tier gets the connector surface; the paid tier gets the autonomy. In other words, OpenAI has already shipped the voice interface for the smart speaker — minus the speaker.

The smart speaker OpenAI has been building with Jony Ive — a mobile, screen-free device with a camera, sensors, rechargeable battery, and an advanced GPT-Live voice stack — was reported by Bloomberg and The Information in July 2026 at a $200–$300 price point. It is now slated for a February 2027 release, not late 2026. The slip is partially legal — Apple sued OpenAI on July 10, 2026 alleging trade-secret theft by former iPhone design head Tang Tan, with the complaint naming more than 400 ex-Apple staff and seeking an injunction — and partially product: the company is still working through voice latency, personalization from email, and the camera/sensor stack.

That is the cannibalization the brief is pointing at. The voice interface — the part of the smart speaker that actually mattered — has been unbundled into the app you already have open. When a Plus user can speak \"summarize the Slack thread about the Q3 launch and draft a reply\" and have it executed on a phone that costs nothing extra, the Jony Ive device is competing with its own employer. It still has a halo: a stationary home device with always-on microphones, camera, and a non-phone form factor. But the software wedge is gone.

ElevenLabs' ARR curve tells you why a 36.7x multiple is defensible. End-2024: $26 million. End-2025: $330 million — 12.7x in twelve months. April 2026: $500 million. June 2026: $600 million per Sacra. Q1 2026 alone added $100 million of net new ARR, the company's best quarter to date, driven by enterprises deploying conversational voice agents for customer support, sales, and internal training.

At $22 billion, a 20% revenue CAGR over four years — conservative versus the 175% ElevenLabs just printed — gets you to roughly $1.5 billion of ARR by 2030. Apply a 15x exit multiple and the implied2030 enterprise value is $22 billion today. That is not a thesis on a takeout; it is the math a Series D lead uses when justifying a doubling in five months. The risk is that the OpenAI-style free voice models get good enough to displace ElevenLabs in contact-center RFPs — which is why the $22 billion is a tender, not a primary round.

Voice AI in 2026 is a two-sided market with three distinct supply chains: model training and inference (upstream), enterprise voice agents and telephony plumbing (midstream), and the consumer-facing hardware and assistant layer (downstream). OpenAI's move on September 23 disrupts the downstream layer by pulling the assistant into the phone. ElevenLabs' $22 billion is a bet that midstream — voice agents that enterprises actually pay for — will scale faster than the consumer wedge gets commoditized.

Three near-term catalysts set the tape. First, the ElevenLabs $22 billion tender closing window — expected by September 2026, originally set out by Bloomberg on July 2 — is the read-through to OpenAI's next private valuation step (which would put OpenAI above $500 billion on reported terms) and to listed voice comparables like SoundHound. Second, OpenAI's smart speaker slip from late 2026 to February 2027 — disclosed in Apple's July 10 lawsuit and The Information's separate reporting — pushes the Jony Ive halo to a calendar year when ChatGPT Voice has had nine more months to settle into the phone. Third, Apple Siri AI: now delayed to 2026 in the iOS 26.4–27 window, with the new assistant reportedly running on Google's Gemini models — a public concession that Apple cannot build a voice interface competitive with OpenAI on its own silicon.

The stocks that move first are the ones with the tightest read-through. Sonos, which is launching Beam Ultra ($699) and Era 100 AI ($269) in September 2026 with Sonos 2.7 OS adding AI voice controls, now has to sell AI features against a free ChatGPT Voice that is already in users' pockets. Twilio, sitting at ~$290 and a $44.6 billion market cap, has a more direct lever: ElevenLabs revenue compounds Twilio voice minute volumes. SoundHound, with Q2 2026 revenue of $61.9 million up 45% year over year, is the listed proxy for standalone voice AI but is priced like the loser of this contest — shares collapsed from a 52-week high of $22.17 to roughly $6 by late September 2026.

If the OpenAI + ElevenLabs arc plays out, voice in 2030 is a layer of the stack the same way SQL was by 2005. ElevenLabs' IPO target is 2028 per CEO Mati Staniszewski (Bloomberg, March 2026), which means a public-market print on the standalone voice platform is the next big reference point. By then, three things will be true or won't: GPT-Live will either have eaten the consumer voice interface entirely or will have been commoditized by open-source clones (Hume, Kyutai, Cartesia are already shipping); the Jony Ive speaker will either be a category-creator for ambient AI devices or a footnote next to a free phone; and ElevenLabs will either be the Twilio of voice or the SoundHound of voice — that is, a margin-rich enterprise utility or a struggling horizontal.

The structural risk for OpenAI's bet is that voice is the interface everyone — Apple, Google, Amazon, Anthropic, and a dozen open-source contenders — is racing to own. Claude shipped voice mode to Opus and Sonnet on July 23, 2026, two weeks after GPT-Live. Gemini is now embedded in Siri. The model layer is commoditizing faster than the agent layer is monetizing. Whoever owns the workflow — inbox, calendar, customer support queue — owns the voice interface that wins.

Read together, the $22 billion ElevenLabs tender and the September 23 ChatGPT Voice upgrade are not contradictory — they are the two halves of one bet. The private market is paying36.7x ARR for a voice API that enterprises will license because it is the best product in its category and because voice is the new text box for customer support, sales, and internal operations. The consumer market is being given away free because the platform owners — Alphabet, Amazon, Apple, Microsoft, OpenAI — have decided that owning the voice interface to the inbox, calendar, and operating system is worth more than selling it. ElevenLabs wins the midstream. OpenAI wins the interface. Both can be true. Neither guarantees the listed voice names survive.