The Loop Apple's AI Moves Have Formed

In late June, Tim Cook explained why Apple was raising hardware prices across its lineup. The company had been absorbing the cost of a global memory shortage driven by AI data-center demand — what Cook called a "hundred-year flood" in component pricing.

we've been been trying to shield our customers from the increases, but the situation has become unsustainable. — Tim Cook

The pressure was real: Apple was simultaneously lobbying the U.S. government for permission to buy from blacklisted Chinese memory suppliers, something it does not do lightly ^[1]. Read together, the moves that followed convert that cost pressure into recurring revenue routed through Apple's hardware. None was announced as part of a strategy. Each arrived with its own rationale, on its own timeline. But taken together, they trace a closed circuit — and the newest piece, a device-leasing program called Apple Upgrade, closes the last gap. Consider the cutoff first. iOS 27, unveiled in June, restricts advanced AI features to the iPhone 15 Pro and newer; older devices can run the operating system but not the intelligence ^[2]. macOS 27 Golden Gate draws the same line, dropping support for Intel Macs entirely and requiring Apple silicon ^[3]. The effect is to tie AI feature access to device generation — if you want what Apple is calling intelligence, you need newer hardware. Then came the price hikes. In late June, Apple raised hardware prices globally, with Cook explicitly pointing to AI data-center demand for memory and components ^[1]. The increases landed just as the AI feature cutoff was making newer devices more desirable. Separately, on July 21, Apple announced Apple Upgrade, a subscription-based leasing program backed by Klarna that replaces the older iPhone Upgrade Program ^[4]. The structure lowers monthly payments compared to traditional financing, which keeps the upgrade cycle moving even as sticker prices climb. Apple is scheduled to launch it on July 28. The trade-in program, revised in late May, pulls in the same direction — but with a twist. Apple raised trade-in values for its own devices, offering up to $695 for an iPhone 16 Pro Max, while simultaneously cutting what it pays for Android phones ^[5]. The stated purpose was to recover and reuse RAM components during the memory shortage. But the mechanism has a second effect: trade-ins are paid in Apple Gift Card credits, which can only be spent on more Apple products. Old devices become a source of scarce components, and their owners become captive buyers. On top of the hardware layer, Apple raised subscription prices across Apple Music, Apple One, and AppleCare+ in July, citing rising licensing costs. And the company is weighing a paid tier for advanced Apple Intelligence features — image generation, complex conversation — while keeping basic Siri free. Mark Gurman captured the risk:

Apple is still at a place where it needs to prove to consumers that its AI technology is worth using, let alone worth paying for. — Mark Gurman

What makes this more than a pricing story is what is happening to the model layer at the same time. DeepSeek permanently cut its V4-Pro model prices by 75% in May, making it 12 to 19 times cheaper than OpenAI's GPT-5.5 for equivalent tasks. OpenAI's adjusted gross margin fell from 40% in 2024 to 33% in 2025 as inference costs quadrupled, and the company slashed its projected infrastructure spending from $1.4 trillion to $600 billion by 2030. OpenAI is now pivoting to enterprise, shuttering consumer products like Sora and Instant Checkout — vacating the consumer market Apple is filling with on-device intelligence. Anthropic, meanwhile, was hit with a record $1.5 billion copyright settlement for training on pirated works, adding a legal-cost burden that further pressures the model-only business ^[6]. Apple's posture toward this commoditization is visible in its sourcing. In China, Apple Intelligence runs on Alibaba's Qwen model. In the U.S., the iOS 27 public beta confirms Siri AI is powered by Google's Gemini, not OpenAI ^[3][7]. Apple is also evaluating PrismML's compression technology to shrink a 54-gigabyte model to under 4 gigabytes for on-device execution. The model is an interchangeable component. The value sits in the hardware that runs it. Even when model labs try to reach the device themselves, they travel through Apple's infrastructure. OpenAI updated Codex to enable desktop control on macOS — screen viewing, cursor control, app interaction — and Perplexity launched a $200-per-month "Personal Computer" AI assistant for Mac that orchestrates local files and native apps ^[8]. Both run on Apple hardware. Both are distributed through Apple's App Store, which takes a 15 to 30 percent commission. The model layer can compete on features, but it cannot compete on the substrate. The real threat to this arrangement is not a better model. It is regulators who are targeting the specific joints of the circuit. The Department of Justice is in settlement discussions with Apple over its 2024 lawsuit alleging the company monopolized the smartphone market across digital wallets, messaging, smartwatches, and super apps. Representative Alexandria Ocasio-Cortez called for breaking up Apple after the June price hikes, saying these companies

We need to break up a lot of these companies that are far, far too big and we need to be instituting consumer protections for people. — Alexandria Ocasio-Cortez

In the United Kingdom, the Competition and Markets Authority is moving to force Apple to allow steering — letting developers bypass the 30 percent App Store commission — and to open NFC to third-party wallets ^[9]. Both measures would cut into the toll-road economics that fund Apple's services layer. Apple is fighting them, arguing that opening its ecosystem exposes users to scams and security risks. That defense is worth sitting with. Apple's argument against regulatory intervention is that the closed loop protects users. Which is another way of saying the closed loop is the product — and that the product works because nothing gets in or out without Apple's cut. The question regulators are now asking, on both sides of the Atlantic, is whether that arrangement is a security feature or a tax.