The AI-device race was supposed to be about inventing a new interface. Its newest milestone is a lawsuit over an old interface’s engineers.

Apple’s case against OpenAI puts recruiting and confidential-information conduct at the center of a hardware program still reportedly aiming for a 2026 unveiling and a 2027 release. The contradiction resolves once the device stops being treated as a concept and starts being treated as an industrial system. The model and form matter, but so do the people who know how hardware reaches production, the information they may carry, and the incumbent prepared to contest the boundary between expertise and secrets.

The lawsuit does not prove Apple’s allegations. It does show that the contest has changed phases.

The first phase mistook personnel for the product

In December 2023, sources said Tang Tan, then Apple’s vice president for iPhone and Watch product design, planned to leave in February 2024. Later that month, reporting said he would join Jony Ive’s LoveFrom to work on AI hardware. At that point, the program could still be understood through biographies: an experienced Apple designer leaves, joins another celebrated Apple alumnus, and applies familiar product skill to a new computing category.

That framing made sense because the relevant assets were still people and ideas. When the structure is young, talent can look like execution: assemble the right designers, recruit enough engineers, and the object follows. OpenAI’s pay strategy reinforced that logic. Equilar’s 2025 estimate put average stock-based compensation at roughly seven times Google’s pre-IPO level and about 34 times the average among other pre-IPO peers.

OpenAI’s average stock-based compensation per employee in 2025, according to Equilar

That pay system is built to pull scarce capability inward. It fits a phase in which the central question is who can attract the people able to imagine and build something new. But once capability accumulates, recruitment stops being merely an input to the program and becomes a contested boundary around it.

The concept became an organization, and the organization became a target

In May 2025, Jony Ive, Scott Cannon, Evans Hankey, and Tang Tan founded io. The significance was not another cluster of recognizable names. It was the conversion of a loose talent story into an organization with a hardware purpose.

Organizations create continuity, and continuity creates exposure. A concept can leave ownership boundaries untested; a hardware division hiring directly from an incumbent cannot. The closer a program moves toward a product, the harder it becomes to separate general expertise from accumulated process knowledge. Hardware expertise is not only what an engineer can draw or calculate. It is also familiarity with constraints, failure modes, design decisions, and the sequence by which a device becomes repeatable.

Each phase looked complete while it lasted. First the project needed designers. Then it needed an organization. Then the organization needed experienced hardware recruits. That third phase created the conditions for the fourth: an incumbent defending the line between hiring expertise and acquiring protected information.

Apple had already identified the structural weakness

In April, Apple named John Ternus as its next chief executive, with Tim Cook set to become executive chairman on September 1. Reporting around the appointment identified top-talent retention amid historic turnover as one of Ternus’s central challenges. In May, his first comments emphasized “deliberateness and discipline.” By June, sources said he was preparing to reshape Apple’s industrial design team and reassert its influence at the executive level.

Apple had identified the weakness before the lawsuit made it visible: defending its product system depended partly on keeping the people who carried that system’s knowledge. Its response was not simply to design a better device. It was to restore institutional weight to design and tighten the relationship between technical leadership, executive authority, and retention.

It is tempting to reduce the dispute to the reported strain between Tan and his former boss, Ternus. That is the least durable explanation. Personal friction is contingent; the pressure is structural. Apple has an established hardware system to protect. OpenAI has a nascent one to assemble. Both systems place unusual value on the same finite pool of people, and the conflict follows from that overlap even if every individual actor is replaced.

A physical interface cannot remain a software abstraction

“AI hardware” sounds weightless until it requires memory chips, manufacturing geography, experienced engineers, and a release date. Apple’s own operations expose that physical reality. Investment from Apple and Samsung has been flowing into Vietnam’s technology and electronics sector. Apple has also reportedly begun testing CXMT DRAM chips in devices sold in China, putting the supplier and its relationship with Beijing under scrutiny.

These are not side issues to hardware. They are hardware. A device reaches the market through specific components in specific jurisdictions, coordinated by organizations that know how to qualify, integrate, and deliver them. An incumbent’s advantage therefore does not sit inside a single product drawing. It is distributed across personnel, processes, suppliers, and institutional memory.

The AI-device contest began with the assumption that intelligence would reverse the old hardware hierarchy: if the model became the primary interface, then the company with the best model could define the next device. The industrial phase reverses that assumption. Once intelligence needs a body, the old hardware constraints return, and the companies that already control mature product systems gain new ways to defend them.

The lawsuit marks a reversal, not an outcome

Apple alleges that Tang Tan, now OpenAI’s chief hardware officer, directed Apple employees interviewing with OpenAI to share company secrets. Apple also alleges that iPhone engineer Chang Liu carried confidential information with him when he left for OpenAI’s nascent hardware division.

Those are allegations, not findings. The available reporting does not establish that secrets were transferred, and the lawsuit does not establish that OpenAI’s device will be delayed or blocked. OpenAI reportedly still considers the 2026 unveiling and 2027 release achievable.

The schedule is precisely why the dispute matters. If the program were a distant research exercise, the case would sit outside execution. But the program is recruiting hardware engineers against a reported development horizon. That makes the provenance of expertise, the conduct of interviews, and the handling of confidential information part of the path between prototype and release.

The original advantage was mobility: experienced people could leave an incumbent and concentrate their knowledge around a new idea. At sufficient scale, that strategy reaches its treacherous turn. The same hiring that accelerates the challenger gives the incumbent grounds to scrutinize how the capability was assembled. The mechanism that built the program becomes the mechanism through which the program can be contested.

The race was supposed to produce a new interface. By July, its consequential interface was the doorway between two employers.