OpenAI, Qualcomm, MediaTek to Build AI-Native Phone by 2028

OpenAI has been the world's most prominent AI software company for the past three years. Now it wants a piece of your pocket. Reports surfaced Monday that the company is working with Qualcomm and MediaTek to develop custom smartphone processors, with Luxshare — the contract manufacturer best known for assembling AirPods and Apple Watch — tapped to co-design and build the device itself. According to Ming-Chi Kuo of TF International Securities, whose supply chain intelligence has a strong track record on Apple products, specifications will be locked by end of 2026 or early Q1 2027, with mass production scheduled for 2028. The move signals that OpenAI believes the only way to control the end-to-end AI experience is to own the hardware layer — and that the smartphone form factor is not dead so much as waiting for an OS-level replacement.

From Earbuds to a Full Phone: Why OpenAI Changed the Hardware Roadmap

When Sam Altman acquired Jony Ive's design startup io in a $6.4 billion equity deal last year, the prevailing assumption was that the first product would be an ambient wearable — perhaps AI earbuds that could listen, answer, and act without requiring a screen unlock. Carl Pei, the founder of Nothing and a name that surfaced in early io discussions, pointed in the same direction: a screenless or screen-minimal device that lived at the edge of your peripheral awareness. OpenAI's Chief Global Affairs Officer Chris Lehane pushed back on the earbuds thesis in April, telling reporters the first hardware product would be announced in the second half of 2026 and would be "different from a smartphone" in terms of paradigm — not form factor.

The Kuo report clarifies what "different from a smartphone" means in practice. OpenAI is not building a conventional Android handset with ChatGPT as a launcher widget. The architecture removes the app layer entirely. Instead of downloading discrete applications — a maps app, a banking app, a travel booking app — the device deploys AI agents that receive natural language instructions and complete tasks by calling APIs directly. No App Store browse-install-update cycle, no push notification fatigue from a dozen competing apps, no UI fragmentation. The agent mediates everything. That design choice is why OpenAI needed custom silicon rather than an off-the-shelf Snapdragon: agents doing persistent background reasoning, real-time multimodal sensing, and API orchestration impose a different compute and power profile than a conventional smartphone workload mix of camera, social media, and streaming.

Qualcomm's Silicon Play and the Margin Problem It Must Solve

Qualcomm's stock rose 7% on the Kuo note, a reaction that tells you how hungry the company is for a credible AI revenue vector beyond Android handset royalties. The structural problem is visible in the company's own financials: profit margins have compressed from 25.8% to 12% over the past several quarters as the Android premium segment stagnates and Samsung increasingly designs in-house chips for its flagship lines. The licensing-and-silicon hybrid model that made Qualcomm dominant in the 4G era is under strain in a world where Apple, Samsung, and now Amazon have vertical silicon ambitions.

An OpenAI partnership on custom AI silicon — co-designed with MediaTek, which dominates the mid-range Android market and has been aggressively expanding into on-device AI processing — gives Qualcomm something it has lacked: a credible design win in a post-app paradigm. The chips in question are reportedly optimized for on-device AI inference, which requires different silicon trade-offs than traditional mobile processors. Heavy matrix multiplication for transformer inference, low-power NPU clusters that can sustain agent background tasks without draining a battery in four hours, and memory bandwidth architecture suited to loading and partially caching large model weights — these are design choices Qualcomm's Snapdragon X series has been moving toward, but that a co-development with OpenAI would accelerate with actual model-workload profiling.

MediaTek's inclusion is equally strategic. It brings manufacturing process flexibility — MediaTek has relationships with TSMC's leading nodes and has demonstrated competitive NPU designs in its Dimensity series — and geographic diversification of chip supply away from pure Qualcomm dependency. For OpenAI, dual-sourcing silicon from two independent fabs also reduces the kind of supply-chain concentration risk that left smartphone makers scrambling during the 2021 chip shortage.

Apple, Samsung, and the Competitive Stakes of an App-Free OS

The app paradigm is not incidental to Apple's business model — it is the business model. The App Store generated an estimated $85 billion in gross payments in 2025, with Apple taking a 15-30% cut on most transactions. A device that eliminates the app install layer in favor of agent-mediated API calls does not route revenue through an App Store. If OpenAI's device captures even a single-digit percentage of the global premium smartphone market, the structural hit to Apple's Services segment is measurable.

Samsung faces a different exposure. It competes on hardware differentiation while running Google's Android OS — which is itself an app-centric paradigm. A credible agent-OS device would pressure Samsung to either adopt a competing AI platform or accelerate its own One UI AI ambitions, which as of early 2026 remain incremental improvements layered on top of the Android app model rather than replacements of it. Google, for its part, has been building Gemini deeply into Android and has moved aggressively with the Galaxy S26 integration, but even Gemini-powered Android retains the app install as the primary software delivery mechanism. None of the incumbent platforms are structurally positioned to move to an app-free model without destroying their own developer ecosystems.

Apple is the more dangerous incumbent to challenge. Jony Ive spent thirty years shaping Apple's product language. His departure did not burn bridges — it bought optionality. Apple reportedly approached io before OpenAI closed the acquisition. If the Jony Ive-designed OpenAI device is polished enough to win in the premium segment, it is doing so on Apple's own aesthetic and brand territory, not positioning against Samsung's Android mainstream.

Luxshare, Contract Manufacturing, and the Supply Chain Shift

The choice of Luxshare as co-designer and manufacturer deserves more attention than it typically gets in the semiconductor-focused coverage of this story. Luxshare is not a passive assembler. Over the past decade it has moved from cables and accessories into complex electromechanical assembly, winning progressive Apple volumes on AirPods, Watch, and MagSafe accessories. It now has active negotiations to take on iPhone assembly capacity that has historically sat with Foxconn and Pegatron.

Co-designing the OpenAI device with Luxshare — rather than handing Foxconn a finished hardware spec — is a signal about manufacturing co-dependency. OpenAI and Luxshare would share design IP, joint tooling investment, and potentially exclusivity on certain manufacturing processes. That relationship structure is closer to the Apple-TSMC model than to a standard contract manufacturer arrangement. For the broader supply chain, it raises the question of whether Luxshare's growing strategic role across AI hardware clients (it also has emerging relationships with Meta's VR manufacturing and several robotics programs) is positioning it as the dominant contract manufacturer for the post-smartphone device generation.

Component-level, an AI-native phone with persistent on-device inference creates elevated demand for high-bandwidth memory stacked close to the NPU, advanced packaging substrates that can co-locate logic and memory on a dense interposer, and efficient power management ICs that manage NPU duty cycles without the kind of thermal spikes that throttle performance on today's flagship handsets. Samsung Foundry and SK Hynix — already the primary memory suppliers for AI accelerators — stand to benefit from a hardware design that treats on-device model caching as a first-class architectural constraint.

The 2028 Timeline as Strategic Positioning, Not Product Launch

It is worth being precise about what Kuo's 2028 mass production estimate means. It is not a product launch announcement. It means that if specs lock by early 2027, procurement lead times, tooling, and pilot production consume roughly twelve to eighteen months before volume units begin rolling off the line. An H2 2026 announcement of a first hardware product — as Lehane telegraphed — is almost certainly a preview or conceptual reveal for developer and enterprise audiences, not a consumer availability date. The actual first consumer units reaching market would land in late 2028 at the earliest, possibly early 2029.

That timeline is, counterintuitively, a feature rather than a bug for OpenAI. A 2028 hardware ship date buys twenty-four months of model improvement, agent infrastructure maturation, and enterprise API ecosystem development. OpenAI's nearly one billion weekly users are already accustomed to interacting with agents through the ChatGPT interface. Eighteen months of incremental agent capability improvement — better tool use, longer context, more reliable multi-step task execution — means that by the time hardware ships, the software experience may be compelling enough to justify a hardware-layer switch. The device is not the bet; the agent paradigm is the bet. The device is proof-of-concept.

What OpenAI is really announcing with this supply chain configuration is that it intends to control the platform that runs agents, the chips that power them, the device that surfaces them, and the manufacturing relationship that scales them. That is a vertical integration thesis that the smartphone industry has not seen attempted at this scale since the original iPhone in 2007 — and even then, Apple launched with existing GPRS modems and a standard ARM application processor. OpenAI is starting, by contrast, with custom silicon built around its own model workload profiles and a device architecture that treats apps as legacy software.

The incumbent platforms will adapt. They always do. But the window for adaptation is shorter than the 2028 date makes it look. By the time OpenAI devices ship in volume, every major handset OEM will need to have an answer for agent-first OS design or risk a category shift that looks, in retrospect, as inevitable as the touch-screen transition looked when the first iPhone was announced.

The most consequential technology transitions are always announced too early to seem urgent and arrive too late to seem surprising. OpenAI just made its announcement.