Kuo: OpenAI appears to be fast-tracking its AI agent phone with two NPUs and a custom MediaTek Dimensity 9600 SoC, targeting mass production as early as H1 2027
【Industry Check Update】OpenAI appears to be fast-tracking its first AI agent phone, with mass production targeted as early as 1H27. Potential drivers include supporting a year-end IPO narrative and intensifying competition in AI agent phones. MediaTek currently appears better
Context & Ripple Effects
Related reporting had OpenAI exploring smartphone-chip development with both MediaTek and Qualcomm, while separately building out hardware, design, manufacturing, and supply-chain capabilities. This update narrows that earlier exploration toward a MediaTek-based handset program and advances the indicated production timetable.
The reported dual-NPU design extends MediaTek’s established push to make the NPU a central part of its Dimensity platform. It also fits OpenAI’s broader move from internally deployed custom AI silicon toward consumer-device distribution.
First-order effects
- OpenAI and MediaTek would move from exploratory chip collaboration to a more defined custom-phone program, with the handset’s AI compute architecture centered on two NPUs.
- MediaTek gains a potential design win in a high-profile OpenAI device, while Qualcomm remains relevant to the earlier reported development work but is not named as the selected SoC partner in this update.
Second-order effects
- A two-NPU handset design would put pressure on competing mobile-chip platforms to differentiate not only on peak AI performance, but on how they partition AI workloads across on-device compute.
- OpenAI’s device effort would deepen its need for the hardware and supply-chain capabilities it has reportedly been recruiting for, linking chip design choices more tightly to product design and manufacturing execution.
Third-order effects
- If OpenAI reaches production, leading AI model providers may increasingly treat purpose-built consumer hardware as a distribution layer rather than relying solely on third-party smartphones and apps.
- The effort points to a more heterogeneous mobile-AI stack, where specialized on-device accelerators and model-provider software are co-designed; whether this becomes a durable handset category depends on execution and user adoption.
The trend: AI companies are extending vertically from models and cloud infrastructure into custom silicon and consumer hardware to control the end-user AI experience.