OpenAI Building AI Agent Smartphone With Qualcomm, MediaTek for 2028

OpenAI is reportedly developing a smartphone built around AI agents instead of apps, with Qualcomm and MediaTek co-designing a custom chip and Luxshare handling manufacturing. Specs are expected to lock in by Q1 2027 and mass production is targeted for 2028, according to a report from veteran Apple analyst Ming-Chi Kuo.

Qualcomm shares jumped about 7 percent on the news, the strongest single-day move for the chipmaker in months.

Why this matters

If the report is accurate, this is the most ambitious consumer hardware project OpenAI has ever announced or been linked to. CEO Sam Altman has been telegraphing for over a year that the company is exploring devices, but a phone with custom silicon and a Qualcomm-MediaTek partnership is several orders of magnitude beyond a smart speaker or wearable.

The pitch — replace the app grid with a single AI assistant that completes tasks for you — is also the most direct attempt yet to challenge the iPhone's interaction model. Every major AI company has discussed this premise. None of them has tried to actually build the device until now.

The other reason this is significant: it's a clear signal about where OpenAI thinks the agent paradigm is heading. If you believe AI agents will eventually replace most of how people interact with software, then the device the agent runs on becomes the most important piece of consumer technology in a decade. Owning that device, rather than shipping software on someone else's, would be enormously strategic.

What was actually reported

The reporting comes from analyst Ming-Chi Kuo, who has a strong track record on Apple supply chain leaks but covers the broader hardware space as well. According to Kuo's report, the partnership structure is roughly:

OpenAI is leading product definition and the AI software stack, including the agent runtime that would handle most user-facing tasks. Qualcomm and MediaTek are jointly designing a custom system-on-chip optimized for AI workloads, with Qualcomm reportedly contributing 3D DRAM specifically tuned for on-device model inference. Luxshare, a Taiwanese manufacturer that builds significant volume for Apple, would handle assembly.

The device timeline calls for finalized specs and component selection by late 2026 or Q1 2027. Mass production is targeted for 2028, with expected annual shipments of 300 to 400 million units once volume ramps — a scale that would put the device on par with mid-tier global smartphone vendors from day one.

None of the companies — OpenAI, Qualcomm, MediaTek, or Luxshare — has officially confirmed the partnership. The information at this point is analyst reporting based on supply chain checks.

How an AI agent phone would actually work

The concept centers on agents replacing apps. Instead of opening a maps app to navigate, opening Uber to book a ride, and opening DoorDash to order food, you'd ask the device to do those things and an agent would execute them — potentially using app-like services in the background, but without the user ever touching an app icon.

That requires solving several hard problems:

On-device model performance. To make agents responsive and private, a meaningful chunk of the work has to run locally rather than in the cloud. The custom Qualcomm-MediaTek silicon is presumably designed to run multibillion-parameter models on battery power without melting the chassis. Qualcomm's reported 3D DRAM design is aimed squarely at this constraint.

Continuous context. The phone needs to remember what you're doing across conversations, locations, and time. That's a different memory architecture than today's chatbots.

Service integrations. Agents are only useful if they can actually do things — book travel, send messages, complete payments. OpenAI would need broad integrations with the long tail of consumer services, either through an app-like SDK or through partnerships with major platforms.

Trust and privacy. The agent will see everything the user does. Building consumer trust in a model that has that level of access is the open social and regulatory question.

Industry impact: the post-app smartphone

If OpenAI ships at the scale Kuo describes — 300 to 400 million units annually — it would be one of the largest new entrants to the smartphone market in a decade. By comparison, Apple sells around 230 million iPhones a year and Samsung sells about 220 million Galaxy phones.

The competitive impact would be felt most in the mid-range Android segment, where Qualcomm is already the dominant chip vendor and MediaTek competes hard. A premium AI-first phone designed for users who want to interact via agent rather than apps could pull customers from both platforms — particularly users who already prefer ChatGPT to native assistants like Siri.

Apple, meanwhile, has been clear that it sees on-device AI as central to its strategy and has been integrating ChatGPT into Apple Intelligence under a partnership signed last year. The new reporting suggests OpenAI is hedging against the partnership by building its own first-party device — a parallel that mirrors how Google built Pixel as a hedge against Android OEM dependence.

What competitors are doing

Several other AI companies have announced or are rumored to be working on hardware. Humane's Ai Pin and the Rabbit R1 launched as early attempts at the post-app device thesis but struggled commercially. Jony Ive's company io, which partnered with OpenAI last year on a device project, may be involved in the new phone effort. Perplexity has been rumored to be exploring an AI-first browser device. Meta has been pushing aggressively into smart glasses with Ray-Ban.

What separates the OpenAI-Qualcomm-MediaTek effort is scale. Production targets in the hundreds of millions of units imply a fundamentally different ambition than a niche enthusiast device.

What industry insiders are saying

Reaction has been mixed but engaged. Hardware analysts have credited OpenAI for assembling a credible supply chain coalition that could realistically execute on these volumes. Skeptics have pointed to the long history of failed attempts to displace the iPhone and Android — the platform inertia is enormous, and customers have repeatedly rejected single-purpose AI devices.

The most common steelman is that even a partial success here would be strategically valuable. If OpenAI ships 50 million units instead of 400 million, that's still a top-ten smartphone vendor and a meaningful first-party distribution channel for ChatGPT and future agent products. The downside cost of trying — particularly given the partnership structure with experienced phone-making suppliers — is bounded.

What's next

Three things to watch over the next nine months. First, official confirmation. None of the companies involved have publicly acknowledged the project, and a deal of this scale would normally require disclosure to public-market investors at some point. Second, the chip details. Qualcomm and MediaTek collaborating on a single SoC is unusual; the technical architecture they choose will tell us a lot about whether the device is genuinely optimized for AI workloads or essentially a rebadged premium Android part. Third, software previews. OpenAI has not described what the agent UI would actually look like. The company will need to demo a working version of the post-app interface well in advance of 2028 to build developer and consumer interest.

For consumers, the device is at least two years away. For investors, the relevant time frame is much shorter — supply chain confirmations and chip details are likely to emerge over the second half of 2026.

The bottom line

OpenAI building a phone is no longer just speculation. The supply chain partners, the chip strategy, the manufacturing footprint, and the timeline are now all on the public record, even if no one has officially confirmed the project. If even part of what Kuo describes ships in 2028, the market for consumer hardware enters its biggest reset since the iPhone launched. The next eighteen months are going to determine whether that's a real shift or another expensive experiment in the long graveyard of post-smartphone devices.