Analyst Ming Chi Kuo reports OpenAI is working with MediaTek and Qualcomm on a smartphone processor, with Luxshare as manufacturer, targeting 2028 production. The core thesis: only by fully controlling OS and hardware can AI agents achieve comprehensive real time inference, as…
When Ming-Chi Kuo publishes research about a tech company's supply chain, the industry pays attention. The TF International Securities analyst has a two-decade record forecasting component flows and production timelines with enough accuracy to move Apple shares. His newest finding, published this week via Sina Finance: OpenAI is working with MediaTek and Qualcomm to develop a smartphone processor, with Luxshare as the exclusive manufacturer, targeting 2028 production. The wire ran it as a hardware story.
But the more important passage in the research was an architectural argument. Only by completely controlling the operating system and hardware can you provide comprehensive AI agent services, Kuo wrote. Only the phone has all of the user's current state, which is the most important input for real-time AI agent inference. The logic: for an AI agent to act on your behalf in the moment, it needs continuous access to what you're actually doing right now, not just what you asked. Who you're messaging. Where you are. What you just booked. What you owe. What you missed. Your phone is the only device that has all of that. A server-based model or a smart speaker sees a query. The phone sees the moment.
OpenAI declined comment. What the company has confirmed, in a sworn court filing from February, is something smaller. Peter Welinder, OpenAI's vice president and general manager, testified that the first hardware device will not ship before the end of February 2027, Business Insider reported. Most widely reported as a smart speaker with a camera, priced between $200 and $300, according to people familiar with the plans, MacRumors noted. The smartphone Kuo describes is a separate project with no confirmation it exists.
The gap between the confirmed first device, eight months away, and the unconfirmed second device, production not expected until 2028, is where the real timeline risk lives. The processor spec and vendor decisions are expected at the end of 2026 or early 2027, according to Kuo's research. If that milestone passes without a visible design win, the 2028 target is already in trouble.
There is also the name problem. OpenAI acquired io Products, Jony Ive's startup, for $6.5 billion in May 2025, NPR reported. Nine months later, it told a court it would not use the name io or any variation of it, after losing a trademark dispute with a hearing-aid company called iYo, Business Insider wrote. The company's largest hardware bet came with a name it could not legally use.
The architectural argument matters beyond this specific device. Today's AI assistants, in phones and browsers and smart speakers, see a query, not a context. They don't know what you were doing before you asked, what your calendar looks like, what messages you haven't answered. An agent with continuous access to your phone's state does not need you to narrate the situation; it already knows. The question is whether OpenAI can build the full stack: the hardware, the chip, and the developer ecosystem required to make it real. The spec decision expected before the end of 2026 is the first concrete test.
Hardware is where software companies go to fail. Microsoft wrote off $7.6 billion on Nokia and spent a decade making Surface competitive. Google has tried and largely abandoned multiple hardware form factors. Amazon's Fire Phone became a textbook example. OpenAI has the brand, the models, and the capital. It does not have the supply-chain relationships, the manufacturing discipline, or the after-sales infrastructure Apple and Samsung spent decades building. And the OS question is not small: OpenAI does not have a mobile operating system. Building or licensing one, and convincing developers to create AI-native experiences for it, is a multi-year effort with no guarantee of adoption.
The smart speaker arriving in February 2027 is the first evidence of whether OpenAI can ship hardware at all. The smartphone is the evidence of whether it can own the relationship.
Research for this article drew on Ming-Chi Kuo's April 27 industry research published by Sina Finance; OpenAI's February 2026 court filing in the iYo trademark matter, reported by Business Insider; and supply-chain reporting from MacRumors, NPR, and TechCrunch.