OpenAI's $6.5B Hardware Bet: Jony Ive's Screenless AI Device Coming H2 2026
OpenAI is attempting something riskier than launching GPT-5.5: building hardware.
In 2025, OpenAI acquired Jony Ive’s hardware startup io Products for approximately $6.5 billion in an all-equity deal. This wasn’t merely an acquisition—it was a strategic pivot from pure software toward integrated consumer technology.
At the January 2026 Davos forum, OpenAI’s Chief Global Affairs Officer Chris Lehane confirmed the project remains on track, with the first device expected to debut in the second half of 2026.
This is not an accessory. Internally, Sam Altman positions it as a “third core device”—alongside smartphones and laptops, but with fundamentally different logic.
Product Form: Screenless, Pocket-Sized, Always-On
Multiple sources have clarified the device’s core characteristics:
| Feature | Description |
|---|---|
| Screen | None |
| Interaction | Voice-first + ambient awareness |
| Size | Pocket-grade |
| Chip | Custom 2nm processor |
| Manufacturer | Foxconn |
| Target Volume | 40–50 million units (first year) |
| Design Philosophy | ”Calm computing”—reduce screen dependency without replacing phones |
Jony Ive’s design language is explicit: minimal, restrained, unobtrusive. After viewing prototypes in November 2025, Altman stated: “I can’t believe how jaw-droppingly good the work is.”
Why Screenless
OpenAI’s assessment: smartphones were built for the App era, not the AI era.
Touchscreens, notification centers, app grids—these structures inherently fragment attention. AI’s core value lies in “working in the background.”
This device experiments with ambient intelligence:
- It listens, but doesn’t interrupt constantly
- It sees (via sensors), but doesn’t record everything
- It responds, without requiring unlocks, app launches, or prompt typing
The interaction design bar is far higher than ChatGPT. No screen means:
- All output must be delivered through voice or extremely restrained physical signals
- Users cannot “see” feature boundaries—the device must teach itself through usage patterns
- Error costs are severe—a misspoken response is harder to tolerate than a mis-tapped button
Competitors and Reference Points
| Product | Outcome | Lesson |
|---|---|---|
| Humane AI Pin | Poor sales, company sold | Immature hardware + unclear use cases |
| Rabbit R1 | Reputation collapsed | Overpromised software, underdelivered |
| Meta Ray-Ban Smart Glasses | ~10M annual shipments, 75–80% market share | Don’t replace phones—be a good accessory first |
OpenAI’s approach hews closer to Meta’s strategy: don’t replace the phone, escape the screen first. But the ambition is larger—if ambient intelligence is validated, this becomes a new category, not just a new product.
Timeline and Key Milestones
| Date | Event |
|---|---|
| May 2025 | OpenAI acquires io Products for $6.5B |
| November 2025 | Altman confirms first prototypes complete |
| January 2026 | Lehane at Davos confirms H2 2026 launch |
| September 2026 | Expected reveal window |
| Q4 2028 | Product portfolio planned to expand to 5 devices |
Foxconn is confirmed as the manufacturing partner, with production expected in the US or Vietnam. This suggests OpenAI is learning from Apple’s supply chain playbook—but skipping Luxshare Precision to leverage Foxconn’s established systems directly.
Risks and Unknowns
Technical:
- Latency. Cloud-inference voice loops must compress below 1 second, or the experience collapses
- Battery. Always-on multimodal sensors + 2nm chip—power management is a hard constraint
- Context switching. Without a screen, users struggle to confirm what the device “is currently doing”
Commercial:
- Pricing unknown. If positioned as premium consumer electronics, what’s the early-adopter conversion rate?
- Privacy. Always-listening devices face far higher trust barriers than mobile apps
- Legal. OpenAI’s trademark/copyright dispute with Iyo may slow marketing timelines
Conclusion
OpenAI is building hardware not because software isn’t profitable enough. Because Altman and Ive both believe AI’s ultimate form is not an app, but a way of being.
Whether this device succeeds or fails, we’ll know by late 2026. But regardless of outcome, it tests a more fundamental question: are humans willing to delegate part of their daily decision-making to an invisible, unobtrusive, always-on intelligence?
That matters more than GPT-5.5’s SWE-Bench score.