Apple grew iPhone sales by 17.5% without joining the AI data center race

Major technology companies are spending tens of billions of dollars on data centers, accelerators, and power infrastructure. Apple has taken a different position: it aims to be the device, interface, and checkout through which people use AI. The model may belong to Google, OpenAI, or another provider, yet Apple can still make money from selling the computer, taking a subscription commission, or charging for expanded access to its own features.

So far, the formula looks convincing. In the September quarter, iPhone revenue rose 17.5% year over year to a record $57.6 billion for the period. Mac sales climbed 24% to $10.8 billion, their highest level in four years. The company does not disclose how much of this growth came specifically from AI, so the technology cannot be credited with the entire result. Management does, however, explicitly cite new features as a reason for device upgrades and users switching to Apple products.

AI without a proprietary cloud worth trillions

Apple combines three computing layers: on-device processing, its own servers, and models from external partners. The updated Siri uses Gemini technology, while sources say the company may offer a choice of third-party providers for writing and image-generation tools.

This approach differs markedly from its rivals' strategies. Apple's capital expenditure in the latest quarter was $2.2 billion, compared with $77.3 billion at Amazon and $51.4 billion at Alphabet. The comparison is imperfect: these are total company expenditures rather than dedicated AI budgets, and some model-related costs do not appear in them at all. Still, the gap illustrates the central point: Apple is not yet trying to own the entire computing stack.

Lower capital spending does not mean free AI. Apple's operating expenses rose 21% to $19.2 billion, partly because of research and development. Server-side Siri requests will also increase the cost of delivering services. The company still does not know what share of requests will remain on the device and what share will move to the cloud.

The Mac is becoming a local AI server

One practical example is Gap, which uses Mac Mini computers to process data locally for its loss-prevention system across hundreds of stores. Requests are not sent to an external cloud provider, so the business does not pay for every token and retains greater control over its data. Christie's uses local models on Mac Studio systems in workflows that include valuing objects.

The same logic applies to a small team. Suppose a support department needs to analyze internal documents containing sensitive data every day. Instead of continually sending text to an external API, it can buy several Mac Mini computers, deploy a suitable model locally, and obtain predictable costs. This is not a universal solution: powerful models, peak loads, and rapid scaling still often require the cloud. For stable workloads, however, local hardware can prove cheaper and more secure.

Controlling the gateway matters more than owning the model

The next source of revenue is paid usage limits. Apple plans to restrict access to features that require server-side models, including the new Siri and intelligent photo editing. Heavy users will be able to purchase additional access, probably through expanded iCloud plans.

Third-party subscriptions provide another channel. AppMagic estimates that App Store commissions from generative AI applications could bring Apple more than $1 billion in 2026. This revenue stream is not guaranteed, however: OpenAI and other providers want to move payments to their own websites so they do not have to share revenue.

The takeaway for developers is straightforward: separate local and server-side features from the outset, calculate the cost of one useful action rather than only the price of a token, and prepare clear usage limits. Product teams should examine who controls the customer relationship: the model provider, the device owner, or the app store. Apple demonstrates that the most profitable layer of the AI market may lie outside the model itself—in hardware, interfaces, distribution, and payments.