Apple Is Sitting Out the AI Building Boom. The Savings Aren’t Free

Big Tech is competing over construction budgets. Apple is cutting capital spending. In the September 2026 quarter, Amazon, Alphabet, Microsoft and Meta spent roughly $202 billion between them, with each company’s spending up 80–120% from a year earlier. Apple’s fell 31%, from $3.2 billion to $2.2 billion. Those figures cover company-wide cash purchases of property and equipment, including AI infrastructure, and exclude finance leases. They aren’t a tally of AI spending alone. The divergence is striking all the same: four companies are accelerating while Apple eases off.

For fiscal 2026, Apple’s capital spending fell 29% to $9 billion, while research and development spending rose 32% to $45.8 billion. Apple doesn’t disclose a separate AI budget. But that combination leaves plenty of room to develop products without matching its rivals’ construction bills. In a recent assessment, eMarketer analyst Jacob Bourne highlighted the strategy’s benefits for cash flow and financial risk. In an industry where spending another few tens of billions has become proof that you’re serious, financial restraint looks almost subversive.

Apple has good reasons to let someone else finance the race. It sells the devices and services through which AI reaches customers. Serving those customers can involve a mix of on-device processing, Apple’s own servers and outside technology. Its multiyear partnership with Google brings Gemini technology to its next-generation models and a more personalized Siri. Most iPhone buyers want an assistant that works; the lab that supplied the model weights is unlikely to determine their next phone purchase. Apple’s bet is that it can source some of the intelligence elsewhere while keeping the product, distribution and customer relationship.

The economics are compelling. We’ve previously examined the estimated cost of a $30 billion AI data center, of which $17.1 billion goes toward chips and server equipment. Much of that investment starts losing its competitive edge long before the building wears out. Each new generation of accelerators forces operators to revisit pricing, utilization and replacement schedules. Apple can buy what this race produces without owning all the equipment that makes it possible. Its suppliers get to explain to shareholders why last year’s technological marvel needs replacing.

If the industry builds too much capacity, this could prove an excellent strategy. Suppliers competing for business would make computing cheaper, better models would give Apple more options, and the company could acquire stronger capabilities without having to earn back the cost of the original buildout. A technology can succeed even as the investors who financed its first great expansion lose money. Apple is positioning itself to shop the resulting sale. For a company that already has an enormous audience, that is a serious commercial strategy, not an afternoon spent waiting for someone else to invent something.

The price is dependence. A key model supplier has a say in how quickly the technology improves, what it can do and the terms on which it is available. Several interchangeable suppliers would give Apple leverage. One indispensable supplier could claim an increasing share of the product’s profits. Switching models is also harder than finding cheaper electricity: quality, safety, assistant behavior and operating-system integration all need testing again. Spending less on the underlying technology can create obligations that barely register in a quarterly capex figure.

Running costs are already making themselves felt. Cloud inference has to be paid for as people use it; a popular feature generates more requests and bigger bills. Apple’s December-quarter gross margin forecast is 45–46%, below the 46.3% analysts had expected, with rising memory prices and the Siri AI rollout both weighing on the outlook. In its November assessment, Evercore ISI said the debate was increasingly shifting toward margins and earnings power. Keeping capex low avoids some infrastructure risk. It doesn’t make the computing behind each additional cloud-AI user free.

“the debate is increasingly shifting toward margins & EPS power as memory costs rise, Services growth moderates and Siri AI costs start to flow into the model.” — Evercore ISI

The bigger risk concerns who controls the experience. Apple currently owns the device, the operating system and the gateway to services. If an AI agent becomes the main intermediary between people and their digital lives, power could migrate toward whoever controls that agent. An outside model working inside Siri, under Apple’s rules, suits the iPhone maker perfectly well. But if people choose an independent assistant and take it with them across devices, relying on someone else’s intelligence becomes a strategic vulnerability. The billions saved on infrastructure won’t, by themselves, keep users loyal.

Apple’s bet deserves more respect than the usual verdict that it has fallen behind in AI. It could avoid the boom’s most expensive phase and still profit handsomely from the results. Two tests matter: how much profit AI usage leaves behind, and who keeps the customer relationship. If models become cheap, interchangeable components, Apple will look like the sensible company in the room. If the intelligence supplier gains control over the product, today’s savings will have bought tomorrow’s dependence. The most consequential number in this story has yet to appear in the accounts: what Apple will have to pay to stay in control.