Apple Might Win the AI Race by Not Doing Anything… Yet

The Capex Gap is Not an Accident

The numbers tell the story plainly. Amazon is projected to spend $131.8 billion in capital expenditures in 2025. Alphabet, $91.4 billion. Meta, $72 billion. Apple? A comparatively modest $12.7 billion.

This is not a company that cannot afford to spend more. Apple generates some of the highest free cash flows in corporate history. The restraint is deliberate.

Figure 1: Cumulative capital expenditures for major Big Tech companies, 2020–2025 (Actual or Estimated). Apple's line (gold) barely registers against the steep acceleration of its peers—by end of 2025, Amazon will have spent more than seven times Apple's cumulative total over the same period. Source: Google Finance.⁽¹⁾

The Gemini Gambit: A Calculated Contradiction

Apple's conservative stance was recently complicated by one of its more revealing moves: partnering with Google to integrate the Gemini model into Apple Intelligence.

This decision is notable precisely because it appears to contradict Apple's deepest brand promise—user privacy. Routing queries to a third-party, cloud-based model is an admission that Apple's internal server-side models are not yet competitive. It is a rare public acknowledgment of a gap.

The genius, or the calculated risk, lies in how Apple manages this tension. The likely solution is to gate what gets sent to the cloud—trivial or non-sensitive queries go to Gemini, while anything personal stays on-device. It is privacy as a filter, not an absolute. Whether users accept this trade-off will be one of the defining product stories of 2026.

Where Apple Is Actually Winning: Edge AI

While the cloud AI arms race dominates headlines, Apple has been quietly building an insurmountable lead in a different category: edge AI—intelligence processed directly on the device.

On-device processing is the technological backbone of Apple's privacy commitment. When your iPhone recognizes a face, transcribes speech, or enhances a photo, none of that data leaves the device. That is not a marketing claim; it is an architectural reality enabled by Apple's custom silicon.

Apple's Playbook: Let Others Pay for the R&D

Apple has never been first. It has been best.

The history of innovation is littered with first movers who lost. What matters is who perfects a technology once the optimal use case is clear—and no company does that better than Apple.

The AI era is no different. Apple is letting OpenAI, Google, and Meta absorb the enormous costs and reputational risks of training frontier models. Once the commercially viable use cases crystallize, Apple will integrate the best-in-class solution into its ecosystem—instantly reaching hundreds of millions of users with a single software update. No competitor can match that distribution.

The Valuation Premium: What the Market Knows

Here is the remarkable part: despite avoiding the massive capex bets its rivals are making, Apple continues to trade at a premium Price-to-Earnings multiple compared to most of Big Tech.

That premium is not irrational. The market is pricing in four structural advantages:

In essence, the market is rewarding Apple for prioritizing profitability, privacy, and proven execution over the high-risk, high-reward gambles of the AI infrastructure race. The bet is not that Apple will train the largest model. The bet is that Apple will monetize whoever does.

References

1. Google Finance. "Capital Expenditures — Amazon, Alphabet, Microsoft, Meta, Apple." Accessed February 20, 2026. https://www.google.com/finance.
2. Apple Inc. "Apple Intelligence Overview." Apple Newsroom, 2025. https://www.apple.com/apple-intelligence/.
3. Gurman, Mark. "Apple to Add Google Gemini to iPhones in AI Push." Bloomberg, March 2024.
4. Apple Inc. "Apple Silicon — A-series and M-series Neural Engine." Apple Developer Documentation, 2025. https://developer.apple.com/machine-learning/.