OpenAI spent $6.5 billion to acquire Jony Ive's hardware team and bet on a smartphone where agents replace apps. The correct form factor solves what killed Humane and Rabbit. The $167 billion app economy coordination problem is what might kill OpenAI.
In May 2025, OpenAI paid $6.5 billion in stock to acquire io Products, the hardware company Jony Ive founded after leaving Apple. Fifty-five former Apple engineers came with it. The deal is the most expensive hardware acquisition in AI history and it buys OpenAI two products: a screenless wearable called Sweetpea shipping in the second half of 2026, and an AI-native smartphone with specs finalized in late 2026 and mass production targeted for 2028.
The smartphone is the interesting bet. Sweetpea is another entry in the AI hardware graveyard — a new device category solving a problem nobody has articulated. The phone is different. It is the correct form factor with a new interaction model. That distinction is what separates it from everything that came before. It is also what makes the coordination problem so much harder than OpenAI appears to acknowledge.
The Graveyard
Humane raised $230 million, shipped the AI Pin, sold fewer than 10,000 units, and was acquired by HP for $116 million in early 2025. The devices were permanently bricked on February 28. Rabbit shipped 100,000 R1 units to mass returns and struggled to make payroll through the end of 2025. Both failed for the same reason: they created new device categories that solved problems nobody had. Users do not evaluate a product against its roadmap. They evaluate it against their phone.
OpenAI's Sweetpea wearable — a pill-shaped behind-the-ear device with a 2nm chip and voice-first interface, manufactured by Foxconn in projected volumes of 40 to 50 million units — faces the same structural disadvantage. It asks users to carry an additional object. The phone already does everything a screenless wearable proposes to do, plus everything else.
The smartphone bet reverses the calculus. Instead of asking users to add a device, it asks them to replace one they already carry with something that works differently. The form factor is correct. The interaction model is the variable. Instead of tapping icons to launch applications, the user speaks to an agent that acts on their behalf — booking, buying, messaging, navigating — without ever opening an app.
The Coordination Problem
The phrase agents replace apps sounds like a user experience change. It is a coordination problem disguised as a product decision.
The combined App Store and Google Play generated $167 billion in consumer spending in 2025, growing 11 percent year over year. Apple has paid $550 billion cumulative to developers since 2008. More than 35 million developers participate in the mobile app ecosystem. The 15 to 30 percent commission structure funds not just distribution but verification, payment processing, update delivery, and trust establishment.
When OpenAI says agents will replace apps, it is proposing to replace this entire coordination mechanism. Who builds the agents? How do they monetize? Who verifies that an agent booking a flight is authorized to charge your card? Who handles disputes when an agent purchases the wrong item? Who delivers updates when the agent's capabilities change? Apple solved these problems incrementally over seventeen years. OpenAI needs answers before its phone ships or the hardware arrives with no ecosystem.
The chicken-and-egg problem is severe. Developers will not build agents for a platform with no users. Users will not buy a phone with no agents. Apple faced the same problem in 2007 and solved it by shipping a web browser as the initial app layer, then opening the App Store a year later after proving demand. OpenAI's path to a comparable bootstrap is unclear. ChatGPT is the obvious candidate — a single agent that does everything — but a single generalist agent is not an ecosystem. It is a search engine with a different interface.
The Legal Overhang
On April 23, 2026, a federal judge in the Northern District of California granted iyO Inc. a preliminary injunction barring OpenAI from using the io name. The lawsuit alleges trade secret theft against Tang Yew Tan, OpenAI's Chief Hardware Officer and former Apple VP of Product Design. The complaint claims Tan brought proprietary CAD files and designs from iyO through a former iyO engineer named Dan Sargent. Nine causes of action including misappropriation of trade secrets. Discovery disputes are due May 29.
iyO is seeking a portion of the $6.5 billion acquisition value. The case is in early stages but trade secret injunctions have historically delayed hardware timelines by years when they survive preliminary motions. This one already has. The smartphone's 2028 production target assumes the litigation resolves favorably or does not touch the core hardware design. That assumption is untested.
Apple's Response
Apple is not building a competing AI phone. It is building agent capabilities on top of the existing app ecosystem. The Campos project — an internal chatbot powered by Google's Gemini — adds onscreen awareness and cross-app actions to iOS without replacing the app layer. The strategy preserves the $167 billion coordination mechanism while layering AI on top.
This is the structural response that matters. Apple's bet is that agents augmenting apps is a better product than agents replacing apps, because augmentation preserves the developer ecosystem while replacement destroys it. If Apple ships agent-first features in iOS 27 at WWDC 2026, it undercuts the thesis that apps are the problem before OpenAI's phone exists.
Who Wins Regardless
Qualcomm and MediaTek are supplying chips for the OpenAI phone. They win whether it succeeds or fails — the development contracts are signed. Luxshare has exclusive manufacturing. The supply chain gets paid on production volume, not market adoption.
If the phone succeeds at scale — Ming-Chi Kuo projects 300 to 400 million annual shipments in the bull case, which would exceed iPhone volumes — the losers are Apple and Google. Their combined app store revenue depends on the coordination mechanism the OpenAI phone proposes to eliminate. Every agent that replaces an app is a 30 percent commission that disappears.
If the phone fails, OpenAI absorbs a write-down exceeding $6.5 billion on the acquisition alone plus years of development costs. Ive's reputation takes a second post-Apple hit after io's initial stall.
The iyO lawsuit is the underpriced risk. Trade secret injunctions can delay hardware programs by years. The market is pricing the phone on the 2028 timeline. The litigation is pricing it on the pace of federal discovery.
Three predictions, all falsifiable. First: Sweetpea ships fewer than 5 million units in its first twelve months — a new device category with no app ecosystem faces the same structural disadvantage as Humane and Rabbit. Second: the smartphone does not ship in 2028 as projected. The iyO discovery timeline plus the ecosystem chicken-and-egg problem push mass production to 2029 or later. Third: Apple announces agent-first features at WWDC 2026 that undercut the apps-are-the-problem thesis before OpenAI's phone reaches consumers.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)