Bloomberg says the debut hardware carries a camera, sensors, a rechargeable battery, and mechanical parts built to make it read as alive. It ships in 2027, assuming Apple's lawyers don't reach it first.
OpenAI's first consumer hardware product is a portable, screenless smart speaker built to live in the home as a companion, according to a Bloomberg report published Tuesday. The machine carries a camera and environmental sensors, draws on the owner's email to build a picture of who they are, and contains mechanical elements that move under their own power so it reads as something alive rather than as an appliance waiting on a wake word.
Mark Gurman's report, sourced to people who asked not to be identified because the project has not been announced, describes a product OpenAI refuses to call a speaker internally. The framing inside the company is a new class of home computer built for the AI era. The pitch, per Bloomberg's sources: a "humanlike AI companion that lives in the home."
Reuters and CNBC matched the report within hours. OpenAI has not commented on it.
The reveal is expected before the end of this year, with the product shipping in 2027. It arrives four days after Apple filed a federal trade secrets suit that names OpenAI's chief hardware officer as a defendant and asks a judge to bar the company from touching anything it allegedly took.
What the machine actually does
Strip out the language about personality and the spec sheet is legible enough. The device controls smart home appliances, plays media, answers questions, and handles messages, all of it routed through ChatGPT. A rechargeable battery lets an owner carry it from the laundry room to the kitchen to a bedroom across the course of a day, though it works plugged into one spot if that is what the buyer wants.
The camera is what separates it from an Echo.
Bloomberg says the sensors exist to give the device context about its surroundings, so it can respond to what it sees and not only to what it hears. The Information reports a likely price between $200 and $300 and says the camera can register details like objects on a nearby table or conversations happening within earshot.
Voice runs on GPT-Live, the model family OpenAI shipped on July 8. That system uses a full-duplex architecture, meaning it listens and speaks at once instead of waiting for a clean pause, and it hands hard questions to a frontier model in the background while keeping the conversation going. OpenAI said at launch that more than 150 million people already use ChatGPT's voice features weekly.
Personalization is the stated differentiator. OpenAI wants the device to anticipate needs and surface things before it is asked, getting sharper as it accumulates a picture of the household.
The earbuds got bumped
Here is the detail most of the coverage skipped. This is not the device the supply chain has been pointing at for six months.
Since January, leaks under the codename "Sweetpea" have described OpenAI's first product as a screenless audio wearable, a behind-the-ear unit closer to a premium hearing aid than to AirPods. The leaker Smart Pikachu put the design at metal and egg-shaped, with pill-shaped components resting behind the ear. Reporting out of Taiwan's Economic Daily pegged a September 2026 launch at first-year volume between 40 and 50 million units, manufactured by Foxconn in Vietnam after a switch away from Luxshare.
Chris Lehane, OpenAI's chief global affairs officer, told Axios at Davos in January that the company was on track to unveil a device in the second half of 2026. The same supply chain leak said Foxconn had been told to prepare five products by the fourth quarter of 2028, with a home device and a stylus under consideration further down the road.
The home device just moved to the front of the line, and the ship date moved to 2027.
There was a signal, for anyone reading it. Court filings in iyO's trademark suit against OpenAI and io Products indicated that the first device under development was neither a wearable nor an in-ear product. That reads as a clean fit for a countertop machine and a poor one for earbuds. Either the Sweetpea-first leaks were wrong from the start or the roadmap got reordered somewhere between January and July. OpenAI has confirmed neither.
Bloomberg says roughly five products are in development. Beyond the speaker, that list includes a portable device meant to replace the phone. A wearable pendant and home robotics are also in the mix.
Apple's suit is aimed at the program, not the product
Apple filed in the Northern District of California on July 10 against OpenAI Foundation, OpenAI Group PBC, io Products, and two individuals.
The individuals are the story. Tang Tan spent 24 years at Apple, rising to vice president of product design for iPhone and Apple Watch, before leaving in 2024 to co-found io Products. He is now OpenAI's chief hardware officer. Chang Liu spent eight years at Apple as a senior systems electrical engineer.
Apple's account of Tan's conduct is that he ran recruiting as an intelligence operation. The complaint says he used Apple's confidential project codenames on candidates who still worked there, pressed them for updates on those codenamed projects, asked for details on unannounced products, and directed them to bring physical components to interviews for show-and-tell sessions where his team could draw out more. It also alleges he circulated an internal Apple offboarding document that taught new hires how to slip past the company's exit security checks.
Liu's alleged conduct is blunter. Apple says he kept a work-issued laptop after leaving and found a bug that let him reach the company's network file storage from outside. He then downloaded dozens of confidential files covering unannounced technologies and unreleased technical specifications, many of them marked confidential. Axios reported the message he allegedly sent to a former colleague still inside Apple, telling them he could still get into the storage and that he found it "so funny."
The complaint claims more than 400 former Apple employees now work at OpenAI. Apple wrote to OpenAI in February to raise concerns and says it never received a reply. The filing calls what it has found so far the "tip of the iceberg" and argues that Apple has no visibility into what happens behind closed doors at its rival.
OpenAI's public response was a statement on X: "We have no interest in other companies' trade secrets."
Then there is the defense OpenAI is floating through anonymous sources. Bloomberg's people say the speaker veers from anything Apple currently sells and is unlikely to violate any trade secrets. Read that scoping carefully. Apple's complaint is not about what Apple sells. It concerns downloaded files on unannounced products and specifications for things that do not exist yet.
Apple's own home roadmap, reported by the same journalist, includes a tabletop device with a nine-inch display mounted on a motorized arm, slated for 2027. That is the same year OpenAI plans to ship a speaker with parts that move on their own.
Neither is on the market. Both move. Apple is asking for an injunction.
A category with a body count
Screenless, personality-forward home companions with moving parts are not a new idea. They are a graveyard.
Jibo shipped in 2017 at $899, promising a social robot with a swiveling body and a personality, and its servers were switched off in 2019. Mayfield Robotics killed Kuri before it reached buyers. Anki's Vector sold for roughly the price The Information now reports for OpenAI's device, and Anki collapsed in 2019. Humane's AI Pin lasted about ten months at $699 plus a monthly fee before HP bought the wreckage for $116 million. Amazon spent north of $1 billion building Astro, a $1,600 home robot that rolled after you from room to room, and the Journal's reporting files it among the devices division's notable write-offs.
The argument OpenAI is making with its capital is that the model was the missing ingredient. Jibo could pivot and blink but could not hold a conversation. GPT-Live can hold a conversation.
Jibo's failure had little to do with conversation quality. The device cost $899 and did nothing a $50 speaker could not do, and the personality wore thin by about week three. Novelty depreciates. OpenAI is betting that a model good enough to be useful bends that curve into something else.
The arithmetic at $250
Amazon's devices division lost more than $25 billion between 2017 and 2021, according to internal documents obtained by the Wall Street Journal, most of it on Alexa, running inference that was cheap and mostly dumb. One former senior employee gave the Journal the epitaph: "We worried we've hired 10,000 people and we've built a smart timer."
Amazon sold Echo units below cost on the theory that owners would shop through them. They set timers instead.
OpenAI is proposing a device at well under Apple's rumored $350 home hub price, carrying a camera, a battery, motors, sensors, and a continuously listening frontier voice model behind all of it. Full-duplex audio processed without pause is not a cheap workload. A speaker that proactively surfaces things has to be doing work when nobody is talking to it, which is the expensive default.
That math needs a subscription behind it or it does not close.
The pressure here is not abstract. OpenAI filed confidentially for an IPO and was valued at $852 billion in a March round. Reporting in late June said the company is leaning toward 2027 rather than a late-2026 listing, with CFO Sarah Friar pushing for more runway and Sam Altman treating anything under a trillion as a non-starter. Projected cash burn runs to roughly $27 billion this year and about $63 billion next.
A hardware line that loses money on every unit is a difficult paragraph to write into an S-1.
The word OpenAI avoided six days earlier
On July 8, at the press briefing for GPT-Live, OpenAI told reporters it was not trying to build an AI companion. TechCrunch reported that emphasis directly.
On July 14, Bloomberg reported that OpenAI is building a device pitched internally as a humanlike AI companion, running on GPT-Live, whose defining feature will be its personality and its ability to connect with people on a human level.
Six days.
The GPT-Live safety work is not incidental to this. OpenAI built audio-native evaluations covering self-harm, psychosis and mania, emotional reliance, violence, and sexual content, alongside real-time safeguards that can redirect a response mid-sentence. The company also said it is running post-launch measurement focused specifically on emotional reliance.
Then it put that model inside a machine engineered to feel alive, watch the room, read the owner's email, and speak first.
Those two efforts pull in opposite directions, and the device is the one with the budget.
Privacy is the other unsolved half. A camera on a kitchen counter that logs objects and overhears conversations has a consent problem the second another person walks in, and that person did not buy anything.
Everyone else is building one too
The capital chasing this category is not waiting for a product to exist.
Hark, the AI lab Brett Adcock founded in late 2025 with $100 million of his own money, raised more than $700 million in a Series A at a $6 billion post-money valuation in May. Parkway Venture Capital led. Nvidia, AMD Ventures, Intel Capital, and Qualcomm Ventures all wrote checks into the same round, which is an unusual signal about a hardware roadmap that has never been shown to anyone. Hark has disclosed no form factor. Its design director, Abidur Chowdhury, came from Apple, where he fronted the iPhone Air announcement.
Apple's answer arrives on a slower clock. Gurman has reported a $350 home hub with a seven-inch display and facial recognition, plus an indoor security camera targeted at late 2026. The tabletop robot follows in 2027, with final assembly handled by BYD in Vietnam. The hub has slipped repeatedly while Apple rebuilds Siri, and the revamped Siri now runs on Google's Gemini rather than OpenAI's models, which tells you where that partnership currently sits.
Amazon has Alexa+ broadly deployed and refreshed hardware already on shelves. Google has a new Home Speaker.
OpenAI is walking into a room where every competitor has distribution and it has none.
What happens next
The calendar decides most of this. OpenAI plans to unveil the device before the end of the year. Apple's injunction request gets heard on a court's schedule rather than OpenAI's, and a ruling that restricts what Tan's organization can work on would damage the 2027 date more than any engineering problem will. OpenAI's public S-1 will eventually have to describe this hardware program as a risk factor, in language the company does not get to write alone.
Discovery is the wildcard. Apple has said in filings that it expects more to surface, and it holds subpoena power over an organization that has spent two years hiring the people who built the iPhone.
Comments
Join the discussion and share your perspective.