OpenAI's App-Free AI Smartphone: Qualcomm and MediaTek Are Building the Chip

Sam Altman has spent the past year telling anyone who will listen that the smartphone era is ending. On April 27, supply-chain analyst Ming-Chi Kuo gave that claim hardware weight: OpenAI is reportedly co-designing a custom smartphone chip with Qualcomm and MediaTek, with contract manufacturer Luxshare handling system integration. The target is a fully AI-native handset shipping in volume by 2028, aiming at 300 to 400 million annual units — a figure that would surpass Apple’s iPhone shipments if achieved.

Neither OpenAI, Qualcomm, nor MediaTek confirmed the partnership as of publication. Kuo’s note, published through TF International Securities, is based on supply-chain intelligence rather than official disclosure. That caveat aside, Qualcomm’s stock surged as much as 12 percent intraday on the report, signalling that markets took the analyst seriously.

Two Hardware Bets, One Mission

The Qualcomm/MediaTek smartphone sits alongside — not instead of — OpenAI’s existing hardware programme. In May 2025, OpenAI acquired Jony Ive’s device startup io for approximately $6.5 billion, its largest acquisition to date. That project, developed under Ive’s LoveFrom creative collective, is positioned as an ambient AI “companion” device — reportedly screenless or near-screenless — targeting 100 million units and expected to debut in the second half of 2026. Altman has said the io device is “different from smartphones,” not a replacement for them.

The Qualcomm/MediaTek project is a different proposition: a full smartphone designed from the silicon up to run AI agents as the primary interface. Where the io device supplements existing devices, the phone wants to replace them. That distinction matters, and the two tracks represent a widening hedge: if ambient computing takes off first, io is ready; if the smartphone form factor persists longer than expected, the agent-native handset is the answer.

The Agent-First Architecture

The device’s central design premise is the elimination of apps. Rather than launching Spotify, Maps, or a banking application, the user would interact with a persistent AI agent that handles all tasks directly — booking tickets, navigating, playing music, managing finances — by calling underlying services invisibly.

Kuo describes the required architecture in detail. Lighter inference tasks — context awareness, memory management, and small on-device models — would run locally on the custom SoC. Complex reasoning and multimodal requests would be offloaded to OpenAI’s cloud infrastructure in real time. The critical differentiator is what Kuo calls “full real-time state”: the device would continuously capture the user’s location, activity, communications, and ambient environment, feeding that context stream to the agent layer. The AI would not need to be asked questions; it would already know the relevant context.

This design philosophy requires full-stack ownership. A software-only AI layer bolted onto an existing Android handset cannot deliver persistent, trusted context capture at that depth. OpenAI needs the hardware to be its own.

Why Two Chip Vendors

The dual-vendor chip strategy — Qualcomm and MediaTek working in parallel — is unusual but not unprecedented. It mirrors how Apple split iPhone chip volume between TSMC nodes to manage supply risk. Kuo suggests a similar logic applies here: splitting the design work between the two largest Android SoC makers accelerates development timelines while hedging against either vendor’s roadmap slipping. Luxshare, which handles final system co-design and manufacturing integration, has grown its share of Apple’s supply chain over the past three years and is an established partner for complex device programs.

Specifications for the chip are expected to be locked down by late 2026 or early 2027, leaving roughly twelve to eighteen months for final device development and supply-chain ramping before mass production in 2028.

The Competitive Landscape

Apple, Google, and Samsung have each been accelerating their own on-device AI capabilities, but all are doing so within the existing smartphone paradigm — AI as a feature layered onto a traditional OS. OpenAI’s reported approach is more radical: discard the paradigm entirely.

Apple shipped Siri-based intelligence in iOS 26.4 with Gemini integration and is investing in its own custom server-side AI inference silicon. Google has embedded Gemini across its Pixel line and Chrome OS. Samsung’s Galaxy AI suite runs on a combination of Qualcomm and its own Exynos chips. None of these companies has publicly committed to an agent-replaces-apps architecture at the device level.

If Kuo’s report is accurate, the strategic bet OpenAI is making is that the current generation of AI-enhanced smartphones is a transitional form — that the real disruption comes when the agent is the OS, not a feature of it. That is a high-conviction view. The incumbent smartphone vendors have billions of dollars in existing app ecosystem lock-in arguing against it.

The Numbers and the Risk

The 300–400 million annual unit projection is ambitious to the point of being disorienting. Apple, after nearly two decades of iPhone, ships roughly 230 million units per year. For a company that has never built consumer hardware before — and whose first device, the io companion, has not yet shipped — projecting volumes that exceed the iPhone at launch is more a market-size estimate than a forecast. The figure reflects the potential total addressable market for a smartphone replacement, not a plausible Day 1 outcome.

The more immediate questions are execution-related: Can OpenAI, a software-and-research organisation, manage a hardware supply chain at scale? Can Qualcomm and MediaTek co-design a chip that meets the on-device inference requirements for persistent AI agents without crushing battery life? And will any of the major mobile carriers and retail channels agree to distribute a device that actively destroys the app economy they depend on?

OpenAI has not answered any of these questions publicly. What the Kuo report does establish is that the ambition is real, the supply-chain conversations have started, and the timeline is concrete enough to move chip vendors.

What Comes Next

With io device launch expected in late 2026, OpenAI will have its first hardware moment before the smartphone project reaches mass production. That device will serve as a proof of concept for the company’s ability to ship physical products, manage manufacturing relationships, and deliver the AI agent experience it has promised.

If the io launch goes well, it builds credibility for everything that follows — including a handset that, if the Kuo analysis is correct, is already in early chip design. If it struggles, the smartphone project faces a harder road, regardless of how good the Qualcomm and MediaTek silicon turns out to be.

Either way, the AI hardware race has a new entrant, and it is the same company that sparked the current AI cycle in the first place.