OpenAI's Apple problem is a distribution problem dressed as a legal one

OpenAI is considering sending Apple a breach-of-contract notice over how ChatGPT was integrated into Siri and Apple Intelligence, according to Nico Grant and Tripp Mickle at the New York Times.¹ The complaint, as reported, is that ChatGPT is too hard to find inside Apple's interface. The same week's framing detail, also in the NYT piece: Apple has added Google's Gemini as a second model provider inside Apple Intelligence.

These are the same story.

What the deal actually was. When the OpenAI-Apple partnership was announced at WWDC in June 2024, no licensing fee was disclosed. Apple did not pay OpenAI. OpenAI did not, as far as anyone has reported, pay Apple. Sam Altman called it "the best of both worlds,"² which is the kind of thing a CEO says when the deal sheet has no numbers on it. The structural exchange was: Apple gets a credible LLM fallback for Siri without committing to building one; OpenAI gets exposure to roughly 2.2 billion active Apple devices and a conversion funnel into ChatGPT Plus.

That funnel was the entire economic substance of the deal on OpenAI's side. If you cannot find ChatGPT inside Siri, the funnel is a pipe with no water in it.

What's now being disputed. OpenAI's grievance, per the NYT, is about prominence — where ChatGPT shows up in the Siri/Apple Intelligence flow, how often a user is routed to it, how visible the upsell to Plus is.¹ The deal terms are not public, so the question of what "integration" contractually means cannot be adjudicated from outside. But the structural read does not require the contract. The structural read is: OpenAI agreed to a deal whose value depended entirely on a counterparty's design choices, and the counterparty made design choices that reduced the value.

This is not a novel situation. It is what happens to anyone who signs a distribution deal with a platform that also operates further up the stack.

The Gemini addition is the actual signal. Adding Google's model as a second provider, alongside ChatGPT, is the standard platform supplier-management move. You don't add a second supplier when you're satisfied with the first; you add a second supplier when you want a credible alternative to wave at the first one during the next conversation. Apple now has ChatGPT, Gemini, and whatever on-device capability it's been quietly building. OpenAI's exclusivity, if it ever had any, is gone. Its negotiating position is whatever Apple says it is.

I notice that the legal threat surfaces in reporting at roughly the moment the Gemini integration becomes operational. The sequence is consistent with a renegotiation play: OpenAI realises its leverage has been diluted, and a breach notice is the only instrument it has left.

The enshittification frame, applied one layer up. The enshittification pattern is usually described at the consumer-product layer: a service degrades over time as the platform extracts more rent from users and suppliers. Here the same dynamic appears at the B2B2C layer. Apple is the platform; ChatGPT is the third-party integration; the "degradation" is making the integration progressively less prominent. The mechanism is the same — the platform uses its control of the surface to compress the supplier's economics. The only difference is that the supplier in this case is a $300bn+ company with lawyers, rather than a long tail of merchants on a marketplace.

This is a useful generalisation of the frame, because it means we should expect to see the same pattern wherever a frontier model sits inside a consumer product owned by someone else. Samsung-Google, Microsoft-OpenAI on the consumer side of Copilot, any future Meta integration. The model lab is, structurally, a supplier; the platform owns the user.

Where the frame breaks (a little). The standard enshittification story has lock-in on the user side as the precondition: users can't leave, so the platform can extract. Here the lock-in runs the other way. OpenAI is locked into Apple as a distribution channel, there is no second iPhone, while Apple has substitutes for OpenAI. So the "supplier squeeze" reading is correct, but it's worth being precise: this isn't a platform exploiting locked-in users to squeeze a supplier. It's a platform exploiting a supplier that has no alternative distribution at this scale.

The mechanic is the same; the lock-in direction is inverted.

What the legal posture is worth. Breach-of-contract notices over AI distribution arrangements have not been adjudicated. The interpretive questions — what does "integration" mean, what does "prominence" require, what counts as good-faith implementation of a partnership without a payment schedule — are genuinely open. If OpenAI files, and Apple contests, the resulting record (even a settled one) would set the first reference point for an entire category of deals.

The model lab is, structurally, a supplier; the platform owns the user.

But I'd weight the contrarian read heavily. The threat is cheap. The relationship is not. OpenAI needs Apple's distribution materially more than Apple needs OpenAI's model — Apple has Gemini, and Apple has, on a long enough horizon, its own weights. A breach notice that produces a renegotiation is probably the best outcome OpenAI can play for, and it's not obvious the renegotiation produces better terms than the original arrangement. The original arrangement was free distribution; "better than free" is a narrow target.

What this is a case of. It's a case of a model lab discovering that distribution deals denominated in exposure rather than cash convert into revenue only at the counterparty's discretion. It's a case of platforms running their standard multi-supplier playbook against AI vendors, on schedule. It's a case of the SaaS-apocalypse frame applied to OpenAI itself: the per-seat consumer subscription business depends on funnels that platform owners control, and those funnels are not stable assets.

What to watch.

• Whether OpenAI actually sends the notice, or whether the leak itself produces movement from Apple.
• Any disclosure, direct or inferred, about the Gemini-Apple terms. If Google is paying Apple, that's one read. If Apple is paying Google, that's a very different one.
• ChatGPT Plus conversion disclosures from OpenAI, if any surface. The deal's value can be reverse-engineered from subscriber adds attributable to Apple devices.
• Whether other model labs with platform integrations (Anthropic-Amazon, Mistral-anyone, Meta's Llama in third-party products) start signalling similar frustrations. The pattern, if it generalises, generalises fast.

The cleanest version of this story is the one Apple is presumably telling itself: we bought optionality for free, and the supplier has noticed.

Footnotes

¶