Ming-Chi Kuo says OpenAI is building a phone with custom silicon by 2028, and the strategic story is who controls the agent layer when apps get replaced

Analyst Ming-Chi Kuo posted on X this week that OpenAI is targeting 2028 for mass production of a smartphone, with specs and component suppliers locked in by end of 2026 or Q1 2027. The hardware stack he describes is a custom chip codesigned with MediaTek and Qualcomm, with Luxshare — Apple's largest assembler — acting as the manufacturing partner. Earlier this year OpenAI Chief Global Affairs Officer Chris Lehane indicated the company's first hardware announcement would come in H2 2026, likely earbuds rather than a full phone. Nothing here is officially confirmed: TechCrunch's coverage notes that OpenAI declined to comment, and the entire source chain rests on Kuo's analyst notes rather than internal OpenAI documents. Kuo's track record on Apple supply chain leaks is good; on the strategic intent of companies he does not have inside-track access to, his accuracy is more variable.

The reason the rumor lands as credible rather than speculative is the io acquisition. OpenAI bought Jony Ive's io design firm in 2024 for $6.5 billion in stock — a price that only makes sense if the goal is launching a hardware product line, not buying design consultancy services. Ive's team is the only group that has shipped AI-native consumer hardware at iPhone-tier polish, and the rumored timeline of 2026 earbuds → 2027 specs lock → 2028 phone aligns with what a serious consumer hardware program would look like coming out of an acquisition that closed last year. The custom-chip-with-MediaTek-and-Qualcomm pattern is also coherent: those are the two suppliers a non-Apple, non-Google company would use to skip the long lead time of full-stack vertical integration while still customizing where it matters.

The interesting part for the rest of the AI industry is the "agents replace apps" thesis Kuo and others have been describing. The pitch is that an OpenAI phone would not ship with a Spotify app, an Uber app, and a DoorDash app installed on a launcher. Instead it would ship with an agent layer that calls Spotify's, Uber's, and DoorDash's APIs directly, with the agent handling the UI and the user interacting in natural language. If that vision actually ships, it changes the platform economics dramatically. Today the App Store and Google Play take 30% of consumer software revenue and act as the gatekeepers for distribution. An agent-mediated phone would push that economic and political control up to whoever owns the agent — OpenAI, in this case — while the apps become headless API providers. That is exactly the structural shift that Apple and Google have spent the past two years trying to prevent at the OS layer, and an OpenAI phone would route around it entirely by simply not running iOS or Android.

For builders, the practical reading is that this is a 2028 timeline at the earliest and possibly later, which means it does not change product decisions for 2026 or 2027. But the strategic implications are worth thinking about now: if your product is an app today, the question of whether you have an exposed API that an agent can call without going through your UI is suddenly a strategic question rather than just an engineering one. The companies that have been investing in real APIs — Stripe, Plaid, Twilio, the various commerce backends — are well positioned for an agent-mediated future. The companies whose entire business model is the UI layer (consumer social, mobile games with monetization built into the visual experience, anything that depends on screen attention rather than transaction completion) have a harder positioning problem. The OpenAI phone may or may not ship in 2028 in any meaningful volume; the agent-replacing-apps thesis it embodies is going to play out across multiple platforms regardless of whether the OpenAI hardware does. Plan for that direction; do not bet your product on the specific OpenAI device.