Apple’s $14B Capex in a $700B AI World: Smart or Shortsighted?

The fiscal year 2026 marks a dramatic split in Big Tech’s AI ambitions. Amazon is plowing $200 billion into capex – mostly for AI infrastructure – while Alphabet (Google) targets $175–185 billion, almost double last year’s number, and Meta Platforms plans $115 to $135 billion. Together with Microsoft (on pace for around $145 billion), these hyperscalers are committing as much as $700 billion in a frenzied race to build out massive, gigawatt-scale training clusters and data centers.

Then there’s Apple (NASDAQ:AAPL) : projecting just a little over $14 billion in capex – essentially flat year-over-year, and a fraction of its peers. In fact, over Q1 FY’26, Apple’s capital spending actually fell year-over-year. When the world is going gung-ho on AI, why is Apple’s approach so radically different?

Apple AI Bet

Apple has taken a different route in AI. Instead of pouring tens of billions into building and training massive foundation models from scratch, it focuses on secure, tightly integrated inference that relies in part on on-device processing. For the most compute-intensive model training, Apple has been looking outward. It recently struck a deal to license Google’s Gemini, reportedly paying a mere $1 billion annually (just about 1% of annual free cash flows). That gives Apple access to a frontier-level model at a fraction of what rivals are spending to develop and maintain their own from the ground up.

Part of this comes down to Apple’s business model. Instead of chasing direct AI dollars through cloud subscriptions, consumption, or ad-boosted AI features, Apple embeds Apple Intelligence as free, privacy-first upgrades to drive premium hardware sales (new iPhones and Macs) while accelerating its digital services growth. To be sure, Apple has not abandoned internal model development. It is actively building its own in-house foundation and task-specific models, but at a smaller scale and with a product-first orientation rather than seeking to build the largest and most complex models.

The risk is straightforward. If the most powerful models stay scarce and tightly controlled, owning them could matter more than renting access. In that case, Apple’s restraint may look less like discipline and more like underinvestment in the core technology of the next era.

Some Big Advantages

Capital Efficiency and Margin Preservation: The primary benefit is the preservation of free cash flow. By avoiding the procurement of pricey GPUs such as Nvidia H100/Blackwell clusters, Apple avoids the massive depreciation schedules currently burdening the balance sheets of the likes of Amazon and Google. At this juncture, we’d agree that licensing a frontier model for $1 billion annually is financially superior to amortizing a $100 billion infrastructure build-out, particularly when the monetization of AI features remains relatively unproven. This allows Apple to maintain gross margins well above 40% while competitors contend with rising depreciation and energy costs.

Distributed Cost Structure: Apple’s vertical integration of silicon allows it to offload some of the computational costs to the end-user. By processing some inference tasks locally on the iPhone’s Neural Engine (NPU), Apple ensures that the energy and hardware wear associated with daily AI interactions are borne by the consumer’s device and not Apple’s cloud. This creates a distributed computing network of 2.2 billion active devices, offering a scale of inference capacity that centralized clouds cannot match in cost-efficiency.

Commoditization of the Model Layer: Strategic outsourcing positions the foundation model as a commodity utility rather than a differentiator. By treating the model provider (Google) as a backend vendor similar to a cloud storage host, Apple retains the ability to switch providers or employ a multi-model strategy. If a superior model emerges from OpenAI or Anthropic, Apple’s architecture allows for integration without the sunk cost of abandoned internal training runs.

Some Big Risks, Too

To be sure, this strategy carries real risks. Licensing Gemini creates a meaningful dependency on Alphabet for the core intelligence layer of the operating system. Unlike a component supplier, Google controls the reasoning engine behind the interaction. That gives Alphabet leverage over pricing, feature access, and timing. The deeper Gemini becomes embedded in iOS, the harder it may be for Apple to switch providers without friction.

There is also brand risk. Private Cloud Compute is designed to anonymize data, but complex queries still require a handoff to third-party models. Even if technically secure, the perception that sensitive requests leave Apple’s infrastructure could weaken one of its strongest differentiators, which is privacy.

Finally, Apple is implicitly betting that foundation models will commoditize. If frontier models remain scarce and strategically controlled, then owning large-scale training infrastructure could prove more defensible than licensing access. If models do not become utilities, Apple’s asset-light approach could look more like underinvestment rather than discipline.

Are you unsure if Apple is the right stock for your portfolio? Consider multi-asset portfolio approach to protect and grow your wealth.

How To Build A Resilient Wealth Strategy

In an environment of fear and greed, individual picks expose you to unnecessary risk. A comprehensive wealth approach positions you effectively to manage risk and capitalize on global trends.

What if you took advantage of the current commodity super cycle? Is a portfolio of 10% commodities, 10% gold, and 2% crypto, in addition to equities, likely to return more in the next 1-3 years? We’ve crunched the numbers. Our wealth management partner manages exactly these kinds of complex multi-asset strategies, blending real assets with high-performance equity sleeves like the Trefis High Quality Portfolio, which has returned > 105% since inception.