“Dramatically Underperforming”: OpenAI's Own Court Filing Admits the Apple Deal Was a Flop

In antitrust litigation, defendants usually fight to keep unflattering numbers hidden. On September 23, 2026, OpenAI did the opposite: in a filing with the Northern District of Texas, it told the court — in its own words — that its flagship partnership with Apple was a commercial disappointment. The ChatGPT integration inside Apple Intelligence, OpenAI conceded, was “dramatically underperforming” by the summer of 2025, and elsewhere in the document, “persistently underperforming.” The admission, first reported by the Financial Times, is the sharpest insider view yet of what was supposed to be the consumer AI deal of the decade — and it is now the centerpiece of OpenAI’s bid to get Elon Musk’s antitrust suit thrown out before trial.

What the filing actually says

The backstory: in August 2025, X Corp and xAI (both now folded into SpaceXAI) sued Apple and OpenAI, alleging that the ChatGPT-Siri integration violated federal competition law by foreclosing rivals like Grok from iPhone users. Musk’s companies dropped their claims against Apple earlier in September 2026, leaving OpenAI as the sole defendant, with trial planned for January 2027.

OpenAI’s new filing asks Judge Mark Pittman for summary judgment — a ruling in its favor without a trial. To get there, OpenAI has to demolish the plaintiffs’ theory of harm: that the Apple deal locked competitors out of a crucial distribution channel. Its chosen weapon is the deal’s own underperformance.

The unredacted portions of the filing sketch a timeline that OpenAI’s partners in Cupertino would probably have preferred stayed sealed:

• December 2024 — Apple ships ChatGPT inside Siri as part of Apple Intelligence. Crucially, the feature is off by default: users face a multi-step opt-in process before ChatGPT appears.
• January 2025 — One month in, the integration “appeared to be off to a slow start,” and OpenAI cut its forecast for incremental logged-in weekly active users expected from the partnership.
• Summer 2025 — By the time Musk’s companies filed suit, “it was clear that Apple’s integration of ChatGPT was dramatically underperforming.” OpenAI had originally expected a “halo effect” from brand association with Apple plus a stream of new subscriptions from Apple Intelligence users. Neither materialized at scale.
• March 2026 — A conversation between Apple and OpenAI about the “persistently underperforming” integration is redacted in the filing, but its existence hints at rising frustration on both sides.
• 2026 — Apple signs with Google; its newest Siri models are Gemini-based. Xcode, meanwhile, expanded to support coding agents from Anthropic, Google, and OpenAI alike.

The legal jiu-jitsu

OpenAI’s argument is a piece of competitive jiu-jitsu: if barely anyone used ChatGPT through Apple Intelligence, then whatever “foreclosure” the deal caused was, in the filing’s words, “indisputably de minimis.” You cannot lock competitors out of a door almost nobody walked through.

To quantify this, OpenAI’s expert, economist Dr. Catherine Tucker, calculated foreclosure shares — the fraction of GenAI consumers who accessed ChatGPT through Apple Intelligence — using the same data and market definition the plaintiffs’ experts relied on. Her numbers are redacted, but the filing says they were “consistent with OpenAI’s internal view that Apple Intelligence saw minimal usage” across all metrics. Notably, the plaintiffs’ own experts declined to calculate foreclosure shares at all.

OpenAI also attacks the exclusivity claim head-on. The agreement, it says, explicitly described itself as non-exclusive and preserved Apple’s right to “integrate products or services that provide the same or similar functionality as [ChatGPT].” Apple told OpenAI it planned to start with one provider and add more — and publicly said as much at launch. Craig Federighi had pointed to Google Gemini as a future option in a 2024 interview. OpenAI did ask Apple for a two-year exclusivity window, according to the filing, and Apple refused.

There is a second, more unusual prong: OpenAI argues that SpaceX’s own IPO registration statement — “replete with disclosures diametrically opposed” to the alleged competitive harm — undercuts the suit. A company telling public-market investors one story about its growth prospects while telling a federal court another, OpenAI implies, cannot have it both ways.

Why the flop happened

The filings also illuminate the mechanics of the disappointment. The integration shipped off by default, buried behind a multi-step opt-in — friction that Apple’s privacy posture made nearly inevitable, and that no amount of “halo effect” could overcome. Distribution inside a billion-device ecosystem is only worth what the funnel actually delivers.

The broader context matters too. Apple Intelligence’s rollout was rocky from the start — delayed, scaled-back, and eventually the subject of a $250 million class-action settlement over the missed Siri launch (the claims website went live September 21, with eligible iPhone owners in line for roughly $25 each). iOS 27’s rebuilt, Gemini-powered Siri shipped just weeks ago behind a waitlist. Apple needed a headline AI partner in 2024; by 2026 it had renegotiated the landscape around itself, pluralizing providers and squeezing OpenAI’s once-treasured placement.

What it means

Three takeaways stand out.

First, distribution is not destiny. The 2024 Apple-OpenAI deal was framed as an unambiguous win for OpenAI — default placement on the iPhone. Two years on, both sides’ own words in court describe a funnel that never opened. Plumbers of AI hype assumed “Apple + ChatGPT” equaled hundreds of millions of engaged users; the reality was an opt-in toggle most users never touched.

Second, the antitrust theory of AI partnerships just got harder. Plaintiffs across the tech sector have been testing the idea that exclusive AI integrations foreclose competition. OpenAI’s filing offers defendants a new playbook: prove the deal flopped, and the harm evaporates. If Judge Pittman buys the de minimis argument, January’s trial may never happen.

Third, leverage has shifted in Cupertino. Apple refused exclusivity, added Google, and now runs a multi-provider stack. OpenAI, meanwhile, is left citing its own underperformance as a legal shield — a striking posture for a company that once described the Apple partnership as a defining consumer milestone.

The irony is thick enough to cut with a knife: the same admission that embarrasses OpenAI’s business development team may be what saves it in court. Sometimes the best defense is confessing that the crime wasn’t worth committing — because nothing was stolen.