OpenAI’s First Phone Targets 1H 2027 as Jony Ive Vision Shrinks

Ming-Chi Kuo, the TF International Securities analyst whose supply chain notes have priced every Apple launch cycle for a decade, said this month that OpenAI is now targeting the first half of 2027 to mass-produce its first phone, pulling the schedule forward by a full year. The chip is a customized MediaTek Dimensity, the assembler is Luxshare, and Kuo estimates the device can ship roughly 30 million units across 2027 and 2028.

That is the same Sam Altman who, eighteen months ago, paid $6.5 billion for Jony Ive’s hardware startup and teased a screen-free pocket gadget that would render the smartphone obsolete. The retreat from that pitch to a rectangle of glass with a SIM tray is the actual news.

Kuo’s 1H27 Target, in Numbers

Kuo’s industry check posted on May 6 reads more like an IPO-readiness memo than a product leak. He cites two drivers behind the accelerated calendar: supporting a year-end public offering narrative for OpenAI, and a fast-closing window in the AI agent phone race before Chinese OEMs ship their own. The combination explains why a device that did not exist on a roadmap eighteen months ago is suddenly tooling up for volume.

The silicon side reads premium-Android, not moonshot. The custom Dimensity sits on TSMC’s N2P process, the same node Apple will move to for its 2027 iPhones. Dual neural processing units (NPUs, the on-chip blocks that run AI models locally) handle inference, an enhanced image signal processor reads the world in real time, and LPDDR6 RAM with UFS 5.0 storage flank the package.

How that compares against what is shipping today:

Why the Pendant Promise Collapsed

The original vision from io, the design studio OpenAI bought in May 2025, was a screen-free, voice-first ambient gadget that fit in the palm. Altman and Ive described it as the post-app device. Early concepts pointed to wearables, a desk-bound smart speaker with a camera, and a pocket pendant. Reviewers compared it, generously, to the operating-system companion in the film Her.

Then the category dropped dead in 2025.

Humane shut down its AI Pin business on February 28, selling residual assets to HP for $116 million; existing pins lost cloud service the same day, bricking the hardware on customers’ lapels. Rabbit’s R1, the orange brick that drew the loudest CES 2024 noise, sold roughly 100,000 units and held about 5,000 daily active users five months later, a 95% abandonment rate. Both products promised what Altman and Ive were teasing: agents that replace apps. Both proved that without a working phone in your other pocket, the agent has nothing to act on.

For winning in the AI era, it is imperative for OpenAI to move beyond software and AI models and focus on owning the end-user interface and the hardware ecosystem itself.

Prabhu Ram, vice president of the Industry Research Group at CyberMedia Research, made that argument to Financial Express Online this month. The framing inverts the io pitch. Owning the interface no longer means inventing a new device category; it means controlling the device category that already sits in 6.8 billion pockets.

MediaTek and Luxshare Take the Build

The reported supplier shortlist tells you what kind of phone this is. None of these companies build experimental gadgets at scale; all of them build flagship-class hardware for established brands.

• MediaTek, the Taiwanese fabless designer that overtook Qualcomm in global smartphone SoC share in 2024, is providing the application processor. The Dimensity 9600-class part is rumored to be customized for OpenAI’s agent workloads with dual NPUs.
• TSMC fabricates the chip on the N2P process, the 2-nanometer node that enters volume production in late 2026. Apple has booked early N2P capacity for the 2027 iPhone; OpenAI is sharing the same supply line.
• Luxshare, the Chinese contract manufacturer that assembles AirPods and a growing share of iPhones, handles co-design and final assembly. The pick signals OpenAI wants Apple-grade fit and finish at iPhone unit cost.
• Security stack includes protected Kernel-based Virtual Machine (pKVM), the same isolation primitive Google uses on Pixel for sensitive workloads.

Notably absent: anything radical on the bill of materials. No retinal projector, no laser ink display, no batteryless ambient compute. The hardware is conservative, deliberately. The bet sits one layer up.

The OS Is the Wedge, Not the Glass

Where the OpenAI phone diverges from every other Android handset is the operating system. Multiple supply chain reports indicate OpenAI is building its own OS, not skinning Android. The home screen is not a grid of icons. The primary interface is a conversational agent that calls tools, drafts replies, books transport, and handles multi-step tasks through natural language.

That architecture is already being rehearsed inside ChatGPT. In December 2025, OpenAI opened a ChatGPT app directory built on the new Apps SDK, letting third-party services like Booking.com, Spotify, Canva, and Zillow run as in-chat experiences rather than tabs you open separately. The mechanism for replacing apps with agent-callable tools is already shipping on the web and mobile clients. The phone is the same idea, given silicon.

Scale gives Altman cover. ChatGPT crossed 900 million weekly active users in February 2026, more than double the figure from a year earlier, and OpenAI now reports 50 million paying subscribers and an annualized revenue run-rate above $25 billion. That is a distribution moat any new phone OEM would kill for. The hard part is converting attention on someone else’s hardware into purchases of your hardware.

What Apple and Google Stand to Lose

Apple finished the first quarter of 2026 with 21 percent of global smartphone shipments per Statista’s quarterly tracker, topping the worldwide market in any Q1 for the first time. The iPhone 17 family swept the top three best-selling models. Android, led by Samsung at 21 percent and a long tail of Chinese OEMs, takes the rest. Between them, Apple and Google control 100 percent of the licensed app distribution to billions of phones, and collect a cut on every paid transaction that flows through it.

An agent-first phone is a structural attack on that pipe. If a user asks the device to book a flight and the OS executes the booking through a server-side API call rather than launching the Expedia app, the app store toll does not apply. Apple and Google have spent the past three years rate-limiting how much of iOS and Android their on-device assistants can touch, partly because their own agentic plans rely on the same plumbing.

That is the rule Altman is trying to step around by leaving the building. The question is whether enough developers and consumers follow him out.

The Bet Riding on 30 Million Units

Thirty million units across two years is a stretch goal even for a flagship Android maker. For context, Google’s Pixel line, after a decade of marketing, shipped roughly 10 million units globally in 2024. OnePlus, a more aggressive Android brand, ships in the same band. To hit Kuo’s number, OpenAI has to convert ChatGPT habit into a hardware purchase at a rate no first-time phone maker has achieved this century.

Three variables decide whether the math closes. Developer adoption of the Apps SDK has to deepen past the launch directory so agents have real tools to call when they leave Wi-Fi. Agent latency on a battery-constrained device has to feel snappier than opening an app the user already knows. And the price has to land somewhere a ChatGPT Plus subscriber considers reasonable, which probably means the high three figures rather than four.

If those break right, OpenAI does what no software company has done since Microsoft tried with Windows Phone: ship hardware that matters in the same room as Cupertino and Mountain View. If they do not, the device joins Humane and Rabbit on the shelf marked tried that, and the $6.5 billion paid for Jony Ive becomes a very expensive line on next year’s prospectus.

Kuo will publish the next update sometime around the 1H 2027 ramp. Until then, the wedge is the operating system; the volume is the test of whether anyone wanted the wedge.