OpenAI’s bid to build a palm-sized AI companion with Jony Ive is running into compute limits — and the harsh realities of hardware
Florian Gaertner/Photothek via Getty Images
Sam Altman wanted a revolution you could hold in your hand. What he has instead is a problem that keeps overheating. According to new reporting from the Financial Times, OpenAI’s top-secret hardware project — reportedly a palm-sized, screenless personal assistant that is being designed by former Apple $AAPL design chief Jony Ive — is mired in technical snags that could delay its launch beyond next year.
The device, which has started to sound a lot like an Apple product (made with former Apple employees), would sit on your desk or travel in your pocket, quietly absorbing context and answering questions like a ChatGPT-powered companion. Altman has pitched the device as something closer to a new computing category than a gadget. The goal is reportedly to create an “AI companion” that coexists with users and responds naturally, freeing people from their phones rather than replacing them.
OpenAI had reportedly been aiming to ship the first version in 2026, with long-term ambitions to put as many as 100 million of the devices into circulation. To get there, Altman and Ive are betting that a generation trained to distrust social media and crave “less screen time” will welcome a subtler, more ambient form of AI in their lives — even if it means the AI is literally in the room.
But the same qualities that make OpenAI’s device sound futuristic also make it nearly impossible to build. People familiar with the project told the FT that OpenAI and Ive’s team are still wrestling with fundamental issues around power, privacy, and performance — starting with how to run enormous AI models on a tiny, battery-driven gadget.
The computing problem looms largest.
One source familiar with the project told the FT that Amazon $AMZN and Google $GOOGL have the advantage here, because they currently have the vast cloud capacity to keep their respective Alexa and Nest devices running smoothly. OpenAI, by contrast, is already straining to feed ChatGPT’s demand. Adding a consumer device to that load without burning cash or melting chips will require a breakthrough in efficiency — which doesn’t exist yet.
Then, there’s the question of personality. Engineers are reportedly testing ways to make the assistant sound helpful but not servile, conversational but not clingy — an algorithmic Goldilocks act that has eluded even the most polished voice platforms. The OpenAI product’s “always on” design only heightens the stakes. Privacy advocates have already dubbed the device a potential surveillance cube, and even casual observers have wondered how a microphone that never sleeps will fit into a world already suspicious of data leaks and eavesdropping.
OpenAI’s $6.5 billion acquisition of Ive’s studio, io, was supposed to signal confidence — proof that the company behind ChatGPT could evolve from code to craftsmanship. The company has also reportedly partnered with Luxshare, one of Apple’s key manufacturing contractors, to produce early units — a nod to just how closely OpenAI’s product mirrors Cupertino’s playbook. But the sheen of design gloss can’t hide the physics underneath. The device’s sensors, processors, and compute costs collide with the limits of power, bandwidth, and public trust. And as Humane’s AI pin and other “always-on” devices have shown, grand ideas can still flop when the real world refuses to turn them on.
This hardware bet is as symbolic as it is strategic for Altman, who just pushed OpenAI’s valuation to roughly $500 billion — a way to turn the abstract magic of software into something you can actually buy. But as delays pile up and costs balloon, the device is sounding less like a category-defining, iPhone moment and more like Apple's Newton era: a neat idea that arrived years too soon.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
