OpenAI is building a phone. Or rather, it is building something that may end up being a phone, despite months of insisting it would be something else entirely.
When Sam Altman acquired Jony Ive's hardware company io Products for $6.5 billion in May 2025, the pitch was explicitly anti-smartphone.
Altman and Ive released a promotional video in which they dismissed phones and laptops as "legacy devices" and promised to create something fundamentally new: a screenless, voice-first, always-on AI companion, reportedly codenamed Sweetpea, that would interact with the world through sensors and conversation rather than a touchscreen.
Recent reports suggest the reality is converging on something that looks a lot more like a phone.
The strategic logic is sound. ChatGPT has nearly a billion weekly active users, but every one of them accesses it through hardware and software platforms controlled by Apple or Google. Apple takes a 30% cut of App Store subscriptions. Google controls default search and assistant integrations on Android. Neither company has any incentive to let OpenAI become the primary interface through which users interact with their devices.
If OpenAI wants to build the kind of always-on, context-aware AI experience that Altman describes, it needs a device that can constantly capture ambient information, spin up local software on the fly, and route tasks through AI agents without permission from a platform gatekeeper. Apple's iOS does not permit that. Google's Android is more open but still imposes constraints. The only way to build a truly AI-native operating environment is to control the hardware.
The problem is that this argument has been made before, by companies with far deeper hardware expertise, and it has never worked. Microsoft spent billions on Windows Phone and Nokia.
Amazon's Fire Phone was discontinued within a year. Meta poured tens of billions into the metaverse in part to escape Apple's platform control. The Humane AI Pin, which promised a screenless, voice-first AI experience virtually identical to what OpenAI is now describing, was a commercial disaster and was sold to HP within months of launch.
The iPhone's dominance is not a function of hardware quality alone. It rests on an ecosystem of applications, services and integrations that users depend on daily: banking, messaging, ride-hailing, social media, health tracking and dozens of other categories. Any new device that lacks those applications, or that relies on AI agents to interact with them on the user's behalf, immediately runs into a distribution problem that no amount of industrial design can solve.
Uber chief executive Dara Khosrowshahi has made clear that his company wants its own interface in front of users, not an AI intermediary. Instagram will not willingly hand its engagement metrics to a chatbot. Banks will not route transactions through an unregulated voice assistant. The applications are the service, and without them, a device is a novelty.
OpenAI has set a target of shipping 100 million devices faster than any company in history, with an unveiling expected in the second half of 2026. The ambition is characteristic of Altman's approach: set a target so audacious that achieving even a fraction of it would be transformative.
The more likely outcome is that Apple's decision to open iOS 27 to third-party AI models, allowing users to route Siri queries through ChatGPT or Claude, will quietly undercut the case for a dedicated OpenAI device before it ever ships.