OpenAI's Free Mobile Voice Agent Just Cannibalized Its Own $300 Smart Speaker — And Apple's $1B Gemini Bet

OpenAI's Sep 23 release lets ChatGPT Voice invoke any plugin or connected app the user's account supports, follow the spoken answer in on-screen text, and finish the task in writing if the voice call ends mid-workflow. Inside the \"Work\" tab, the same voice surface can draft documents, build presentations, push tasks into connected apps, and resume in text when the conversation stops — features that, until today, only ran on web and desktop (OpenAI release notes, Sep 23; Business Insider, Sep 23). Plus and Pro subscribers get the full agentic surface; Free and Go users get plugin-and-connected-app voice inside standard chats.

Bloomberg's Aug 6 report on OpenAI's speaker framed it as the company's first consumer device — a battery-powered, donut-shaped home companion priced between $300 and $400, expected in 2027, designed by Jony Ive's LoveFrom and built by Foxconn (Bloomberg via TechCrunch, Aug 6; OpenAI-Foxconn collaboration, Nov 20 2025). That price band sits on top of a global smart-speaker market worth roughly $19 billion in 2026, where Amazon Echo holds ~23–30% of shipments, Alphabet Nest 20–25%, and Apple HomePod just 10–15% — a category already defined by hardware margin compression and a thin software moat.

The smart speaker is a2027 product. The mobile agent is a 2026 product with a 900-million-user installed base and no new BOM to absorb. Both come from the same shelf of ChatGPT capabilities, but only one of them has to clear a $300 price tag, a custom Foxconn supply chain, and a Jony Ive industrial-design budget.

Apple launched Siri AI at WWDC on Jun 8, 2026 — a rebuilt assistant running on a custom 1.2-trillion-parameter Gemini model, paid for by Apple at a reported ~$1B/year on top of the ~$20B/year Alphabet already pays Apple to be the iPhone's default search engine (Apple newsroom, Jun 8 2026; Bloomberg / Reuters, 2026). The thesis behind that spend: ship a Siri that finally feels agentic, then monetize it through Services attach rates and a possible Siri-AI tier inside iCloud plans.

Three months later, the Siri AI bet collides with two free rivals. Alphabet is sunsetting the original Assistant on Android on Sept 4 and pushing Gemini as the system assistant across phones, cars, watches and TVs. And ChatGPT Voice is now a voice-agent surface on iOS and Android that already has 150 million weekly users — bigger than Siri's daily active base on most published counts. Both arrive without a hardware bill; both arrive at the moment Apple is paying Google a nine-figure annual check just to make Siri competitive.

Because the central event is software, the supply-chain map is split: the voice-agent mobile launch costs nothing in components, but the OpenAI speaker it's about to undercut still anchors a real hardware BOM, and the China/Taiwan ODMs that lose that build face an awkward read-through.

Read through the other way, Apple's HomePod story gets structurally weaker, not stronger. Apple delayed HomePod mini 2 into fall 2026, the same window as iOS 27, and the category leader — Amazon Echo with 23–30% share — has been bleeding Alexa subscribers as Alexa+ rolls out free. A category that was already shrinking in relevance now has two free, agentic mobile competitors. The HomePod hardware refresh is now Apple's marketing positioning.

In the next 30 days, the catalyst calendar is unusually dense. OpenAI DevDay 2026 runs Sep 29 in San Francisco, six days after the mobile voice launch — where managed agents and ChatGPT Work expansion are the rumored headline (OpenAI DevDay announcement). Expect OpenAI to lean further into voice-as-UI on existing devices rather than into the puck. That keeps the spotlight on app-distribution economics and away from the 2027 hardware story.

Over a 1–3 year horizon, three things have to be true for the thesis to hold. First, Alphabet has to convert Gemini into the default mobile assistant across a billion Android devices despite the ChatGPT Voice app sitting one tap away. Second, Apple has to either bundle Siri AI into a paid iCloud tier or monetize it indirectly through Search-default retention — a binary tied to the DOJ Google search remedy. Third, OpenAI has to either materially differentiate the 2027 speaker (always-on, room-far-field, camera-equipped AI companion) or quietly cut its consumer hardware scope. None of those is a sure thing.