Apple handed its AI to Google and called it privacy. It is true, in the narrow sense that Apple’s senior vice president of software engineering, Craig Federighi, intended it on stage Monday, that the company’s new “Apple Intelligence” features run on-device or on Apple‑controlled private cloud servers, and that user requests are not, by default, stored or shared.
The day’s technology roundup underscored the contrast. Alibaba’s AI cloud revenue climbed 38% last quarter, Tencent’s Weixin positioned itself as the gateway to agentic AI services for Chinese users, and Malaysia’s semiconductor supply chain re‑rated on AI demand. That growth sharpened the comparison: the story Apple told this week is one of strategic surrender wrapped in a fresh privacy polish.
Apple has spent a decade positioning itself as the hardware company that cares about your data — a stance that has carried genuine material benefit for users who do not wish to be the product. But the company that makes that promise has now outsourced the core of its next computing platform to the company that makes almost all of its revenue by surveilling people and selling their attention. That fact does not disappear because the processing happens inside Apple’s sandbox. It disappears only from the marketing slides.
The partnership is the latest in a long line of arrangements that antitrust enforcers are only now beginning to untangle. Judge Amit Mehta’s ruling in United States v. Google found that Google’s payments to Apple — estimated at $20 billion in 2022 — to remain the default search engine on Safari constituted illegal monopoly maintenance. The AI version of the deal is the same check, written for a later decade.
Apple has failed, over the decade since Siri debuted in 2011, to build a competitive conversational assistant. Siri remains, by any measure, the weakest of the major platform AI offerings. So in early 2026, Apple tapped Google to overhaul Siri, rename it Siri AI, and embed it across the operating system. The result is that the default assistant on every iPhone will be powered by Google’s models — routed, Apple insists, through its own hardware and cloud, with no data retained.
The privacy architecture is, in theory, a genuine improvement over the standard practice of shipping raw user queries to a remote server. But the architecture is the casing, not the engine. The engine determines who gets to participate in the market for on‑device AI. Google’s models will be integrated at the operating‑system level, accessible through Apple’s Shortcuts app, Safari, photo editing, and calendar. Other AI providers — the startups, the open‑source model developers, the companies whose models might run just as well on Apple’s neural engine — will have to go through the App Store, where commission rates can be as high as 30% and the review process has historically been hostile to apps that compete with Apple’s own services. The arrangement is a classic chokepoint: the platform owner picks a single supplier for the system‑level AI, locks in users who cannot switch operating systems without enormous cost, and then tells the world that the result is private because the processing stays on the device. The privacy claim is real, but it also serves as the technical justification for locking out third‑party AI vendors; the foreclosure flows directly from the architecture that promises data protection.
Federighi’s stagecraft included a swipe at unnamed AI developers who are “racing forward, seemingly pursuing AI for the sake of AI, without clear regard for the people, all of us, that it’s ultimately meant to serve.” The line drew applause. It also drew attention away from the fact that Apple’s own pursuit of AI has led it to the door of the company whose entire business model is the opposite of the regard Federighi described. The technique is a motte‑and‑bailey: the bailey is the claim that Apple’s AI is thoughtfully designed and respectful of users; the motte, which the company retreats to when pressed, is that the models came from Google but the data is handled well. The applause dies down; the agreements stand.
Across the Pacific, the same logic is playing out. Bernstein described Tencent as the benevolent gatekeeper of the Weixin AI ecosystem — a phrase that should set off every antitrust alarm you own. Tencent is integrating its mini‑programs so deeply into the Weixin AI agent layer that agentic assistants can order food, book a flight, and pay for a ride, but only through the Weixin API. Weixin is moving from being a messaging platform to a chokepoint over the entire local service economy. The benevolence is the benevolence of a toll‑booth operator who has successfully convinced the highway authority that he poured the concrete. Daiwa and Deutsche Bank analysts note that Chinese tech companies are showing clearer AI monetization and that token consumption and enterprise workflow execution will become the key commercial metrics. They are describing the mechanism of the tax. When Tencent signs agent‑to‑agent deals with five smartphone makers to embed Weixin as the default interface, it is not building a better product; it is foreclosing rival agents from scale. It is the Google‑default‑payment playbook, ported from search browsers to agentic assistants. The billions Google pays Apple to be the default search engine are the precedent; Tencent is trying to make Weixin the default reality for the AI‑native internet.
At a lower layer of the stack, the same extractive physics are at work. The consolidation signal coming from Singapore’s telecom sector — where the failed merger between Simba and M1 is merely pausing discussions for Keppel and StarHub — indicates that the incumbents know the AI data‑centre buildout, the gigawatt‑scale facilities Naver is planning with Nvidia, requires massive, predictable throughput. They are consolidating the pipes so they can raise the price of the water. The AI hype cycle, as market history shows, is always followed by the consolidation cycle. The capital expenditure is real; the market structure it produces is a monopsony.
This is the process Cory Doctorow named enshittification: the claw‑back of value from both the user and the business customer, achieved by locking the ecosystem so thoroughly that the platform can dictate the terms of every interaction. Doctorow identified four constraints that historically kept the enshittification lever from moving freely: competition, regulation, interoperability, and labor. On iOS, competition is absent because Apple has made it impossible for a rival app store or assistant to gain a foothold. Regulation has not yet caught up to the AI chokepoint, though the EU’s Digital Markets Act is beginning to probe whether Apple’s designation as a gatekeeper should extend to AI services. Interoperability — the ability for a user to replace Apple’s AI with a different one, seamlessly, without losing system integration — is structurally foreclosed. And labor, in the form of the engineers who built Siri’s predecessor, has been sidelined by the layoffs and reorganizations that followed the decision to outsource.
The regulatory response that could reverse this concentration is not obscure. If the AI agent becomes the primary interface for human‑computer interaction — the new browser, the new OS, the new search engine — then interoperability mandates must apply to that layer with the same force they should have applied to the original web. Adversarial interoperability, the right of a third‑party developer to build a tool that plugs into a platform’s agent ecosystem without the platform’s permission, is the only friction that keeps the gatekeepers honest. In the United States, the Computer Fraud and Abuse Act and the anti‑circumvention provisions of the DMCA make that kind of plugging in a felony. In China, the regulatory apparatus is even more tightly aligned with the platform companies’ interests. Without a legal right to interoperate, the benevolent gatekeeper eventually decides to charge for entry.
The regulatory mechanisms already exist — interoperability mandates, non‑discrimination obligations for dominant platforms, antitrust enforcement that treats chokepoints as what they are. They will only gain traction when the technical and civil‑society communities stop treating platform consolidation as an engineering inevitability and start demanding them. The deadline for drawing the line is always now.
But for now, the enclosure is quiet. The slide that said “Your data is not stored” should have carried an asterisk. The asterisk, rendered in the fine print that Apple’s lawyers know by heart, would read: “Ask Google.” We cannot, because Apple’s architecture does not permit it. That is, as the company’s marketing department would put it, a feature. The enclosure is quiet, the data is sandboxed, and the toll on everyone else’s AI will be collected at the door.