Apple’s $6.5B OpenAI Fight: One Circuit Could Wreck the Hardware Bet
OpenAI spent $6.5 billion to get into hardware, then Apple alleged its engineers used Apple’s secrets to speed the job up. That is not a talent dispute. It is a product-killing risk.
OpenAI spent $6.5 billion to get into consumer hardware, then Apple alleged one of its engineers used confidential Apple circuit material to accelerate work at OpenAI. That is not a messy HR dispute. It is the sort of allegation that can turn a grand product launch into a very expensive legal waiting room.
Apple has turned a hiring fight into a hardware threat
On August 31, 2026, Apple told a US federal court that former Apple engineer Chang Liu, now at OpenAI, accessed a confidential power-converter circuit schematic while working at OpenAI. Apple alleged Liu used proprietary information in March 2026 to train an AI agent for engineering work.
The evidence, Apple says, came from a MacBook OpenAI provided on August 21 during the legal process. Apple is now pushing for expedited discovery: in plain English, it wants to get inside the evidence quickly, before documents vanish, memories get fuzzy and a disputed product gets further down the production line.
None of this has been proven in court. That matters. Apple has made allegations; OpenAI has asked for the lawsuit to be dismissed and says it is building entirely new products. But operators should not miss the practical point: when a serious company finds forensic evidence it believes supports its case, the commercial problem starts well before a judge reaches a final verdict.
Apple first sued OpenAI, hardware chief Tang Tan, Liu and OpenAI-owned io Products on July 10, 2026. Apple alleged a broader pattern: former employees taking confidential technical material, product information and internal knowledge into OpenAI’s hardware push.
That hardware push is not some side project cooked up by a bloke with a pitch deck. OpenAI paid roughly $6.5 billion last year for io Products, the device company co-founded by former Apple design chief Jony Ive, Tang Tan and others. It was a giant declaration of intent: OpenAI does not merely want to live inside other people’s phones, laptops and browsers. It wants to own a piece of the physical world too.
And now Apple is saying: not with our homework.
Why a power-converter circuit is a much bigger deal than it sounds
Most people hear “power-converter schematic” and immediately tune out. Fair enough. It sounds like the paperwork behind a toaster.
It is not.
In compact consumer electronics, power management is where the fantasy of a beautiful device meets physics, heat, battery life, charging, component limits, safety testing and manufacturing reality. You can have the world’s cleverest AI assistant, but if the device runs hot, has miserable battery life, costs too much to manufacture or fails certification, you have built a very expensive coaster.
That is why Apple’s allegation matters. A circuit design is not just a drawing. It can embody countless engineering decisions: trade-offs tested, failures avoided, components selected, designs rejected and performance limits discovered the hard way. In hardware, those lessons cost years and millions.
Apple alleges Liu accessed a schematic and simulation data, then used the material in electrical-engineering simulation work. Reports on Apple’s filing say the company tied this to an AI agent that could run simulations and help tune parameters. Again, that is Apple’s allegation, not a court finding. But if the allegation is substantiated, the danger for OpenAI is not limited to a cheque for damages.
A court can order evidence preserved. It can restrict use of contested material. It can slow work while the facts are sorted. It can make suppliers, manufacturing partners and future employees very nervous. Hardware runs on calendars. Miss one window and you do not just lose a quarter; you can lose relevance.
Software companies love to pretend delays are harmless because code can be shipped overnight. Hardware does not care about your vibes. Components have lead times. Factories need commitments. Certification takes time. A product that misses its moment can become a product nobody wants.
The $6.5 billion acquisition now looks less like a shortcut
OpenAI bought io Products because it needed something model companies do not automatically possess: taste, industrial-design judgement, supply-chain knowledge and the ability to turn advanced technology into an object normal people will actually buy.
That is hard work. Apple spent decades becoming excellent at it.
The original strategic logic for OpenAI was obvious. If AI changes how people use computers, whoever defines the next major personal device could own a massive new platform. The winner gets distribution, data, customer relationships and potentially a tollbooth on the next generation of computing.
But this lawsuit exposes the rotten little secret in every hardware land grab: hiring a few stars does not transfer an institution’s capability.
You can recruit brilliant people from Apple. You can acquire a celebrated design studio. You can spend $6.5 billion. What you cannot buy cleanly is the right to use another company’s confidential work product.
That distinction is especially sharp when the target is Apple. Apple is not merely protecting files; it is protecting the accumulated advantage that allows it to charge premium prices and ship products at enormous scale. If Apple thinks that advantage has walked out the door, it will not settle for a polite letter and a fruit basket.
For OpenAI, this creates a nasty asymmetry. Apple can afford a long legal fight. Its core business keeps printing money while the lawyers work. OpenAI’s hardware effort, by contrast, is strategically important precisely because it must prove it can become more than a model provider dependent on other companies’ platforms.
The bigger the ambition, the more painful the delay.
The overlooked angle: AI makes trade-secret hygiene more important, not less
Here is the bit too many founders will get wrong: AI does not make confidential information less valuable because it can generate code, copy and concepts cheaply. It makes confidential information more dangerous.
An engineer with a proprietary document and an AI agent may be able to move faster through tasks that previously took a team days. That is the promise. It is also the compliance nightmare.
The old risk was a departing employee emailing a handful of files to a personal account. The new risk is that those files get fed into an AI workflow, copied into prompts, combined with other materials, analysed, simulated, summarised and scattered across devices or systems before anyone asks the obvious question: should this data have been there in the first place?
That is why “we told people not to steal stuff” is not a serious policy. It is corporate wallpaper.
If you run a business using AI, you need to know four things:
1. What data employees can put into AI tools. Not in a vague policy document. In actual permissions, approved tools and technical controls. 2. What happens when you hire from competitors. High-value hires need a clean-room onboarding process: written declarations, restricted access at first, documented reminders and a hard ban on bringing files, prototypes or confidential notes. 3. Where your agents can act. An AI agent that can access cloud storage, source code, engineering tools or customer systems needs tighter boundaries than an intern with a company laptop. 4. Whether you could prove clean conduct later. If litigation arrived tomorrow, would your logs, policies and onboarding records make you look disciplined or reckless?
You do not need to treat every employee like a criminal. But you absolutely need to stop treating data controls like a job for the IT department after the real work is done.
The contrarian view: this may strengthen OpenAI if it forces discipline
There is a lazy take that litigation automatically ruins a company. Nonsense. Plenty of great businesses have survived lawsuits.
The question is whether the business learns the right lesson.
If OpenAI responds by treating the Apple case as a narrow legal nuisance, it risks repeating the mistake. Fast-growing companies often convince themselves that speed excuses sloppiness. It does not. Speed just ensures the sloppiness reaches more people faster.
But if OpenAI builds a genuinely independent hardware operation—with clean provenance for designs, tightly governed recruiting, strong evidence preservation and leaders willing to slow down when necessary—the legal pressure could force a more durable company.
That is the contrarian point: the real asset is not any one circuit. It is the operating discipline required to create a product without borrowing somebody else’s advantage.
Investors should care because hardware is capital-hungry enough without legal uncertainty. A software company can survive a feature setback. A hardware company can burn hundreds of millions before customers touch the product. Add injunction risk, delayed manufacturing and distracted leadership, and suddenly a $6.5 billion acquisition is not a bold bet. It is a balance-sheet lesson.
What this means for you
If you are a founder, do this tomorrow: write a one-page competitor-hire protocol before you make your next senior hire. Make it clear that the new employee must not bring documents, code, designs, supplier lists, customer information, screenshots, prototypes or personal cloud folders from their old employer. Get it signed. Then enforce it.
If you run an AI-enabled team, map every tool that can receive confidential information. If your people are pasting customer data, source code, pricing, roadmaps or designs into public or lightly governed systems, you are not being innovative. You are being careless with better branding.
If you are an investor, stop valuing “ex-Apple,” “ex-Google” or “ex-OpenAI” on the LinkedIn headline alone. Great people matter. But institutional knowledge is not automatically transferable, and it can carry liabilities along with the magic. Ask how the company protects itself from contaminated information before you celebrate the talent density.
And if you are building hardware, remember this: the thing that kills you is rarely a lack of ideas. It is usually poor execution, bad controls and pretending the laws of manufacturing—or the law itself—will make an exception because your product is exciting.
OpenAI wanted to buy its way into the next computing platform. Apple’s latest filing is a reminder that there is no shortcut around building it properly.