A $300 metal doughnut is a ridiculous object on which to wager the future of computing. That’s why Silicon Valley may be serious about it.
OpenAI’s first consumer device, developed with former Apple design chief Jony Ive, will reportedly resemble a doughnut-shaped hockey puck small enough to carry around the house in one hand. It will have microphones, speakers, lights, cameras, environmental sensors and mechanical components that move as the device listens and responds.
There will be no display. The hardware will instead offer a more capable version of ChatGPT’s voice experience, learning about its owner over time and eventually completing tasks on their behalf. OpenAI is considering a price between $300 and $400, with an unveiling expected before sales begin in 2027, according to Bloomberg’s Mark Gurman. OpenAI has not confirmed those specifications, so they remain advanced product plans rather than finished facts.
On the surface, this is an expensive smart speaker entering a category Amazon invented more than a decade ago. Underneath, it is OpenAI’s attempt to seize control of the interface through which people communicate with artificial intelligence. The company does not want ChatGPT to remain an app trapped inside hardware, operating systems and distribution channels controlled by Apple, Google and Microsoft.
NEW: OpenAI’s device – a human-like smart speaker –
will look like a doughnut and be the size of a hockey puck. It meant to be held. It has a camera, speakers, microphones, lights and, most notably, moving parts to show interactivity. It’ll be $300-400. https://t.co/gDxQU6FB8E— Mark Gurman (@markgurman) August 6, 2026
What OpenAI appears to be building
The reported device is battery-powered and designed to work in several positions: held in the hand, placed on a kitchen counter or left beside a bed. A ring-shaped metal body will contain speaker grilles and microphones, while lights indicate when the machine is listening.
Cameras and additional sensors will allow it to perceive its surroundings. Its moving components are intended to signal attention, response and personality, making the object feel less like a stationary appliance and more like a small physical presence.
That last detail is where this stops sounding like a HomePod and starts sounding distinctly strange. OpenAI is apparently trying to give the device a form of body language. A tilt, rotation or mechanical movement could show that it has heard somebody, is considering a request or is directing its attention toward a person in the room.
Precisely how these components work is unknown. The ring could accommodate a 360-degree microphone array, directional speakers or cameras capable of observing the room from several angles. Movement might orient those sensors, improve acoustic performance or simply create emotional expressiveness. Those possibilities are informed speculation, not disclosed features.
What is clear is that OpenAI wants the machine to feel alive without making it look human. That is a difficult design brief. Too little personality and the product is another Alexa. Too much and it becomes the haunted doughnut from a low-budget science-fiction film.
The $6.5 billion design hire
OpenAI did not recruit Ive to improve the speaker grille. In 2025, it acquired io, the hardware startup founded by Ive and a group of former Apple designers and engineers, in an all-stock transaction valued at approximately $6.5 billion. OpenAI already owned about 23% of io, meaning it paid roughly $5 billion for the remainder. Around 55 product designers, hardware engineers and manufacturing specialists joined OpenAI, while Ive’s independent LoveFrom studio assumed broad creative responsibilities across the company.
The deal brought together an unusually accomplished collection of Apple alumni, including former industrial-design leader Evans Hankey, product executive Tang Tan and hardware specialist Scott Cannon.
Ive’s collaboration with Steve Jobs helped produce the iMac, iPod, iPhone, iPad and Apple Watch, while his influence extended to Apple’s software, packaging, retail environments and headquarters. When Ive left Apple in 2019, Tim Cook said his contribution to the company’s revival “cannot be overstated.”
The partnership with OpenAI is Ive’s first opportunity since leaving Apple to design an entirely new computing platform rather than a car, building, logo or luxury object.
“I have a growing sense that everything I have learned over the last 30 years has led me to this moment,” Ive said when the io deal was announced. OpenAI’s accompanying letter described existing computers as interfaces inherited from an earlier technological era—machines that have not been redesigned around systems capable of seeing, reasoning and understanding.
That is the central thesis of the project. Artificial intelligence has changed dramatically, but we still access it through keyboards, glass rectangles and operating systems whose basic metaphors were created decades ago. OpenAI and Ive believe the interface is now the bottleneck.
Ive’s war against the smartphone he helped create
There is a pleasing historical symmetry here. Ive helped turn the smartphone into the most important consumer object of the past 20 years. He is now helping to construct the argument that we spend too much time looking at it.
The device he and Altman describe is supposed to be calmer than a phone. It will not bombard users with grids of apps, banners, notifications and red circles demanding attention. Instead, the AI will supposedly understand context, decide what matters and interrupt only when necessary.
At an Emerson Collective event, Altman contrasted the intended experience with the sensory chaos of Times Square, comparing the device instead to sitting peacefully beside a mountain lake. Ive said he prefers solutions that “teeter on appearing almost naive in their simplicity”—objects that are sophisticated internally but feel unintimidating and almost thoughtless to use.
That sounds attractive because modern smartphones have become casinos we pretend are filing cabinets.
But removing the screen does not remove complexity. It hides it.
A screen allows users to inspect information, compare choices, correct errors and understand what a machine is doing. A voice-first device must communicate all of that through conversation, sound, light and movement. It must know when to speak, when to remain silent and when to admit that it has no idea what is happening.
Those are not minor user-interface details. They determine whether the product feels magical or infuriating.
Smart speakers promised this once before
OpenAI is not the first company to promise a computer that disappears into the room.
Amazon launched the first Echo in 2014, creating the modern smart-speaker category. Google followed with Google Home in 2016, pitching a conversational assistant capable of answering questions, managing household tasks and controlling connected devices.
The products sold well, but their ambitions contracted. Alexa and Google Assistant became useful for timers, music, weather reports and turning lights off after somebody had already climbed into bed.
They never became general-purpose computers because they could not reliably understand ambiguous requests, retain meaningful context or complete complicated tasks. The underlying assistants were command systems pretending to be conversational partners.
Large language models change that equation. ChatGPT can interpret imprecise instructions, ask follow-up questions and maintain far richer conversations than the original Alexa. Multimodal models can also interpret images, voices and environmental information, while agent systems are beginning to operate software and execute longer sequences of actions.
OpenAI is betting that the smart speaker was not a bad idea. It merely arrived before the intelligence required to make it useful.
That may be correct. It may also be an elegant excuse for building Alexa again, this time in anodized aluminum.
Why OpenAI needs hardware
Frontier models are becoming harder to defend as standalone products. Competitors replicate features, open models improve and benchmark leads expire within months.
As Gilded Age has argued in its examination of the AI compute wars, the durable advantage may no longer be the model itself. It may be distribution, infrastructure, proprietary context and control over the system surrounding the model.
Hardware gives OpenAI all four.
A successful device would establish ChatGPT as the default assistant without requiring users to open an app. It would generate unique sensory and behavioral data unavailable to rival laboratories. It could create a new subscription business while reducing OpenAI’s dependence on Apple and Google’s operating systems.
It would also give OpenAI control over the entire experience: microphones, processors, memory, sensors, privacy controls and the physical signals through which the AI communicates.
Apple understood this model better than anyone. Its dominance came not from possessing the best individual component but from controlling the interaction between hardware, software, services and distribution.
OpenAI wants the same vertical integration for artificial intelligence.
The increasingly hostile relationship between the two companies makes the strategy explicit. Apple has sued OpenAI and several former employees, alleging that confidential hardware information and manufacturing techniques were misappropriated. OpenAI denies the accusations and says it is developing something fundamentally different from Apple’s products.
Whatever the courts decide, the broader conflict is obvious. Apple owns the most valuable consumer-computing platform in history. OpenAI believes that platform was designed for an era that is ending.
What remains unknown
For all the excitement, the most important facts about the device remain missing. OpenAI has not revealed its name, final appearance, battery life, processor, launch countries or subscription model. It has not explained how much computation occurs locally, whether it will function without an internet connection or which outside services it can control.
The absence of a display also raises practical questions. How will the device present maps, documents, photographs or detailed comparisons? Will it send visual information to a phone, television or computer? Can several family members use it without their conversations and memories bleeding together?
The likely architecture is a hybrid one. Basic wake-word detection, sensor management and privacy filtering could happen on the device, while demanding reasoning tasks are sent to OpenAI’s cloud. Running frontier-level models entirely inside a hockey puck would be difficult given battery, heat and memory constraints.
A recurring subscription also appears plausible because complex voice conversations and agentic tasks consume ongoing inference capacity. OpenAI has said nothing publicly about such a fee, however, and the $300-to-$400 price may be intended to reduce resistance to an additional service charge later.
The moving components are another mystery. They may prove functional, delightful or irritating. Consumer-electronics history contains plenty of motorized innovations that looked enchanting in a demonstration and became expensive repair tickets two years later.
The smallest railroad of the AI Gilded Age
The industrialists of the original Gilded Age fought to control railways, oil pipelines and electrical grids. The new barons are fighting over the infrastructure connecting human intention to machine action.
Google has Search, Android and Workspace. Apple owns the iPhone and its operating system. Amazon has Alexa and the commercial machinery behind it. Meta is placing AI inside glasses. Microsoft controls the workplace desktop.
OpenAI currently has intelligence and attention, but it does not own the physical doorway.
The doughnut is its attempt to build one.
The first device probably will not replace the smartphone. Altman has acknowledged as much, just as the smartphone did not eliminate the laptop. A more plausible outcome is that it makes the phone slightly less central: something retrieved for visual work rather than consulted compulsively throughout the day.
That would still be a profound shift. The company controlling the assistant people speak to before touching a screen would occupy an extraordinarily powerful position between desire and action.
OpenAI is therefore not really preparing to sell a smart speaker. It is asking consumers for permission to place ChatGPT in the room, give it eyes and ears, teach it their routines and allow it to act on their behalf.
For $300, the metal doughnut may look small. The claim it intends to stake is enormous.
