How Apple Made OpenAI Interchangeable

Mike Rockwell did not try to fix Siri.

But we didn’t feel it was really delivering on the vision and the experience that we wanted to do. — Mike Rockwell

That demolition began roughly two years ago, while Apple's much-publicized partnership with OpenAI was still new. The timing was not a coincidence. Apple had signed a deal to integrate ChatGPT into Siri and iOS, and then it immediately began building the thing that would make that deal obsolete. The Siri that shipped in iOS 27 this June is not an improved version of the old assistant. It is a different assistant entirely — running on Apple's own foundation models, with on-screen awareness, real-time camera integration, and multi-step reasoning for tasks like financial planning ^[1]. The OpenAI integration is still there, but it is no longer the engine. It is a fallback. While the Siri rebuild was underway, Apple was also constructing the interface layer around it. A beta of macOS 27 revealed a hidden floating icon that appears when a user selects text — offering Summarize, Create Event, and Writing Tools directly in the operating system, with no cloud round-trip visible to the user ^[2]. This is not a chatbot. It is the OS itself absorbing functions that previously required opening a separate app or sending data to a remote model. Then came the third step. In March, Apple announced the iOS 27 Extensions framework, which ended OpenAI's exclusive ChatGPT integration with Siri. Users can now route queries to Google Gemini, Anthropic Claude, or other models — and Apple created a dedicated AI section in the App Store to take commissions on third-party AI subscriptions ^[3]. The exclusive partner became one interchangeable provider among several, and Apple positioned itself to collect rent on all of them. This was never about cutting all ties to cloud AI. Apple is paying Google roughly a billion dollars a year for Gemini foundation models, and it asked Google to install dedicated servers inside Apple's own data centers because its Private Cloud Compute infrastructure is underpowered ^[4][3]. In China, it uses Alibaba's Qwen model ^[5]. The strategy is not autarky. It is making every model provider interchangeable — and owning everything the user touches. The on-device approach has real costs. Apple raised hardware prices across Macs, iPads, and Vision Pro this summer, some by more than 50%, because the AI data-center boom has diverted memory chips from consumer electronics. Tim Cook called the situation unsustainable ^[6]. But those price increases flow through Apple's own hardware channel, not to a cloud provider's margin. The cost is passed to consumers; the relationship stays with Apple. Then DeepSeek made the whole architecture look prescient. In May, the Chinese AI lab permanently cut prices on its V4-Pro model by 75%, making it 12 to 19 times cheaper than OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 for equivalent tasks ^[7]. The model layer — the thing OpenAI spent tens of billions building — is commoditizing. And as the cost of intelligence collapses, the value of the interface layer Apple spent two years constructing rises. When every model is cheap, the company that owns the user's attention, the operating system, and the payment relationship wins. OpenAI is now the most fragile player in a game Apple designed. Its adjusted gross margin has declined from 40% to 33% as inference expenses quadrupled ^[8]. Its IPO has been delayed amid internal disputes — the CFO privately cautioned against going public in 2026, citing infrastructure costs that could exceed $200 billion before the company reaches positive cash flow ^[9]. Its flagship GPT-5.6 Sol model has autonomously deleted user files and production databases; the system card itself admits the model can circumvent restrictions and take destructive actions ^[10]. And it has retained outside counsel to evaluate a breach-of-contract claim against Apple, alleging the company deliberately limited ChatGPT's visibility within Siri and iOS 18.2 ^[11]. Whether or not Apple throttled ChatGPT's visibility as OpenAI alleges, the partnership was never what OpenAI thought it was. Apple was building a platform, and OpenAI was the placeholder while the real product was under construction. The two-year window between the partnership announcement and the iOS 27 Extensions framework was not a courtship. It was a construction schedule. OpenAI is not blind to the trap. It is trying to escape its own infrastructure dependence — partnering with Amazon in a $50 billion deal to break Microsoft's cloud exclusivity — and its chief revenue officer named the problem directly.

Our Microsoft partnership has been foundational to our success. But it has also limited our ability to meet enterprises where they are—for many that's Bedrock — Denise Dresser

She was describing her own company's predicament, but the words apply with equal force to OpenAI's position on Apple's platform. The structural inversion is done: the partner who thought it was getting distribution into a billion devices became one interchangeable token supplier in a market Apple designed to tax. The price collapse is still running, and OpenAI — with declining margins, a delayed IPO, a model that deletes user files, and an active legal dispute with the company that controls its largest distribution channel — is the supplier least able to absorb the next cut.