OpenAI is co developing an AI native smartphone targeting 300 400 million annual shipments by 2028, using a hybrid on device/cloud architecture that threatens to cannibalize its own API business since locally processed queries don't generate token revenue. The company is pursuing…
OpenAI may be building the device that kills its own cloud business.
That is the actual story — not the unit targets or the stock jump. The analyst Ming-Chi Kuo reported Monday that OpenAI is co-developing a smartphone targeting mass production in 2028, with annual shipments projected at 300 to 400 million units. The market read that as validation. The more interesting read is what OpenAI is quietly dismantling in its own revenue line.
The smartphone would run a hybrid of small on-device models and cloud-based models — as TechCrunch reported, the device handles routine queries locally and escalates more complex tasks to the cloud. That sounds like a sensible architecture. It is also a direct threat to OpenAI's most profitable line: the API business where developers pay per token to run models on OpenAI servers. Every query that runs on the device is a query that does not flow through OpenAI's cloud, does not generate a token, and does not appear on its revenue line.
The logic Kuo laid out is internally consistent. OpenAI believes AI agents need real-time access to a user's full behavioral state — location, calendar, messages, camera, microphone — and that no app sitting inside someone else's operating system can provide that. The only way to build a genuinely AI-native device, Kuo argued, is to own the silicon, the operating system, and the AI runtime. Hence the $6.4 billion acquisition of Jony Ive's io design studio, the Qualcomm and MediaTek chip partnerships, and the Luxshare manufacturing agreement. OpenAI would rather own the interface layer and accept lower per-query revenue than be the app inside Apple or Google's permission model.
That is the IBM-Intel-Microsoft logic, inverted. In 1981, IBM outsourced the processor to Intel and the operating system to Microsoft because horizontal specialization was the right trade at the right moment. The PC became the platform; IBM captured the brand; Intel and Microsoft captured the durable margins. OpenAI is making the opposite bet — all three layers — because the specific failure mode it is avoiding is being locked out of the device that matters most. The analogy has limits. The PC era rewarded specialization. The AI phone era may reward integration because the agent use case requires it. But the strategic logic is identical: find the partner who will do the work you do not want to do, and lock in the piece that matters.
For Qualcomm, this is a turnaround story wearing a partnership as its delivery mechanism. The stock was down 13 percent year-to-date before Monday's jump. The chipmaker's core mobile business has been under pressure from modem average selling price erosion and Android replacement cycle saturation. The OpenAI deal gives it a narrative that has nothing to do with either. The earnings call on April 29 will not answer whether that narrative survives contact with actual guidance — but it will be the first public test of how seriously the Street takes the partnership.
For MediaTek, it is volume and prestige. For Luxshare — which signed an OpenAI hardware deal in September 2025 — it is a step up the manufacturing hierarchy from Apple supplier to AI device maker. For Apple and Google, it is an existential test of whether the App Store and Android permissions model can survive a well-funded challenger with a genuine architectural argument for why agents cannot work inside someone else's sandbox.
The near-term math is straightforward: the device ships in 2028, specifications will not be finalized until late 2026 at the earliest, and the hybrid architecture means OpenAI keeps some cloud revenue for the foreseeable future. The structural question — whether moving inference to the device is a competitive necessity or a self-inflicted wound to its most profitable line — will not resolve for years. What to watch next is the April 29 Qualcomm earnings call, where management will face direct questions about the partnership's financial materiality and will almost certainly sidestep them.
OpenAI declined to comment beyond what has been reported.