OpenAI developing AI agent smartphone with Qualcomm and MediaTek, targeting 300-400M annual shipments by 2028

OpenAI is developing a smartphone built around AI agents rather than apps, with Qualcomm and MediaTek jointly designing the custom processor and Luxshare Precision Industry co-designing and exclusively manufacturing the device, according to Ming-Chi Kuo, the TF International Securities analyst whose Apple supply-chain intelligence has made him the most closely followed hardware analyst in the industry. Kuo projects 300 to 400 million annual shipments if the device succeeds, a figure that would exceed Apple’s iPhone volumes and place the phone in direct competition with the two companies that control roughly 40% of the global smartphone market. Specifications and the supplier list are expected to be finalised by late 2026 or the first quarter of 2027, with mass production targeted for 2028. Qualcomm’s shares surged as much as 13% in premarket trading on the report. None of the three companies, Qualcomm, OpenAI, or MediaTek, confirmed the partnership. This is an analyst report, not an announcement, but the supply chain Kuo describes is not speculative. It is the supply chain that already builds the devices you own.

The concept

The phone Kuo describes is not a smartphone with an AI assistant. It is a device where the AI agent is the interface and the app is obsolete. Instead of downloading applications and navigating screens, users would interact with agents that handle tasks directly: ordering transport, booking restaurants, managing email, conducting research, writing messages. The architecture would process lighter tasks on-device, including context awareness, memory management, and smaller AI models, while offloading complex inference to the cloud. The device would maintain what Kuo calls “full real-time state,” continuously capturing a user’s location, activity, communication, and environmental context to feed the agents. This is the vision Qualcomm CEO Cristiano Amon has been articulating throughout 2026: that AI agents will replace the mobile operating system and apps as the primary interaction layer, and that the hardware must be designed from scratch to support continuous, power-efficient AI inference rather than retrofitting existing chipsets with neural processing units bolted on.

The concept is separate from OpenAI’s other hardware project with Jony Ive, the former Apple design chief whose company io is developing a non-phone device, reportedly a smart speaker with a camera first, then glasses, a lamp, and earbuds, with the first product expected in early 2027. OpenAI is pursuing two parallel hardware strategies: a device that reimagines what a personal computer looks like without a screen, and a device that keeps the phone form factor but replaces everything that runs on it. Apple is testing AI smart glasses with a custom chip, cameras, and Siri powered by a Gemini model, targeting 2027. The question of whether AI lives in your phone, on your face, or in a speaker on your counter is being answered simultaneously by every major technology company, each with a different bet. OpenAI is betting on all of them at once.

The supply chain

The credibility of the report rests on the supply chain, not the concept. Luxshare Precision Industry is a major Apple supplier that assembles AirPods, Apple Watch components, and an increasing share of iPhones. Qualcomm’s Snapdragon 8 Elite Gen 5 powers 75% of Samsung’s Galaxy S26 series and has, for the first time, overtaken Apple in raw multi-core and GPU performance. MediaTek’s Dimensity 9500 matches Qualcomm and Apple in CPU performance at lower cost with better efficiency. These are not the suppliers of a concept phone. They are the suppliers of phones that ship in the hundreds of millions. Qualcomm’s acquisition of Edge Impulse, an edge AI developer platform, in 2025 signalled the company’s strategic commitment to on-device AI inference across device categories. The Snapdragon 8 Elite Gen 5’s Hexagon NPU delivers 37% faster AI processing than its predecessor, supports agentic AI that learns from user behaviour, and includes a personal knowledge graph and continuous context awareness through an upgraded sensing hub. Qualcomm is also reportedly building custom 3D DRAM specifically optimised for AI workloads on mobile devices. The silicon for the phone Kuo describes does not need to be invented. The components exist. The question is whether the software paradigm works.

The financial context matters. Qualcomm’s stock was trading at $149.84 before the report, down from a 52-week high of $205.95, with earnings growth declining 46.9% and gross margins down to 55.1%. The company reports earnings on April 29, two days after the Kuo report. In February, Bloomberg reported that Qualcomm gave a “tepid forecast in sign of shaky phone market.” An OpenAI partnership would represent a new revenue stream in a market where Qualcomm’s traditional business, supplying modems and processors to phone manufacturers, is under pressure from Apple’s efforts to develop its own modem chips and MediaTek’s encroachment on the premium Android segment. Qualcomm would be helping build a device designed to challenge the iPhone while continuing to supply Apple with modem chips through at least 2027, a business relationship that embodies the contradictions of the semiconductor supply chain.

The graveyard

The AI device category has produced more failures than products. The Humane AI Pin, a $699 wearable with a laser projector that beamed information onto the user’s palm, was permanently bricked on February 28, 2025, when HP acquired Humane’s remnants for $116 million and shut down the servers. The Rabbit R1, a $199 “large action model” device, attracted 100,000 pre-orders but retained only 5,000 active users after five months, a 95% abandonment rate. Its founder admitted the device launched too early. Both failed for the same reason: they created new form factors that solved no problem the smartphone did not already solve, at price points that demanded the user carry a second device. The OpenAI phone takes a fundamentally different approach. It is not an additional device. It is a replacement for the device 4.7 billion people already carry, in the same form factor, with the same basic capabilities, but with a radically different interaction model. Whether that is enough to avoid the graveyard depends on whether agents can do what apps do, better, faster, and without the friction of learning a new paradigm.

AI is already reshaping the mobile app ecosystem, with “vibe-coded” applications flooding the App Store in such volume that Apple has had to crack down on submissions. The EU is preparing to force Google to open Android to rival AI assistants including ChatGPT and Claude under the Digital Markets Act, requiring equal system-level access for voice activation and deep integration. The smartphone’s software layer is already in flux. Samsung’s Galaxy S26 runs a triple AI engine with Gemini, Perplexity, and Bixby. Google’s Pixel 10 hands off multi-step tasks to background AI agents. Apple Intelligence processes queries on-device with an emphasis on privacy. Every major phone manufacturer is moving toward AI-first experiences, but all of them are constrained by backward compatibility with billions of existing apps and the operating systems that run them. OpenAI’s advantage, if the phone materialises, is that it has no legacy. It can design a clean-slate interaction model without worrying about whether Instagram’s notification system works or whether the banking app renders correctly. The disadvantage is that users may not want a clean slate. They may want their apps and an AI assistant that works around them, which is what Samsung, Google, and Apple already offer.

The question

Kuo’s projection of 300 to 400 million annual shipments would make the OpenAI phone one of the most successful consumer electronics products in history. For context, Apple ships roughly 230 million iPhones per year. Samsung ships approximately 220 million Galaxy phones. A new entrant reaching those volumes has no precedent in the smartphone era. The projection reflects the scale of OpenAI’s ambition, not a reasonable base case for a first-generation device from a company that has never manufactured hardware, sold through carriers, managed warranty claims, or operated a supply chain at consumer scale. The Jony Ive device carries the same risk: a company whose expertise is in large language models attempting to become a consumer electronics manufacturer, a transition that requires competencies in industrial design, supply chain management, retail distribution, and after-sales service that OpenAI does not have and cannot acquire by hiring one designer, however talented.

The 2028 timeline gives OpenAI two years to finalise specifications, secure component supply, build manufacturing capacity, develop the agent-first software platform, negotiate carrier partnerships, establish retail distribution, and convince hundreds of millions of consumers to abandon their iPhones and Galaxy phones for a device built by a company that has never shipped hardware. The Humane AI Pin took longer than that and shipped a device that lasted nine months before being permanently disabled. The ambition is extraordinary. The supply chain is credible. The concept addresses a genuine architectural limitation of current smartphones, which were designed around apps in 2007 and have not fundamentally changed since. But the distance between a credible supply chain report and a shipping product that displaces the iPhone is the distance between a thesis and a business, and every company in the AI device graveyard had a thesis too.