Apple may have stumbled into an AI moat as the field commoditizes. While rivals pour billions into ever‑larger models, the gap between frontier, second‑best, and open‑source systems narrows—Gemma 4, Kimi K2.5 and GLM 5.1 now run comfortably on laptops. Each dollar spent on training shrinks the cost of earlier runs, turning intelligence into a utility for developers and consumers alike as they embed AI into daily workflows.

Apple’s strategy contrasts sharply with cash‑burning labs. OpenAI, valued at $300 B, recently killed Sora after it cost roughly $15 M daily against $2.1 M revenue, prompting a $1B equity stake to vanish. Micron redirected capacity toward AI only to see demand evaporate, sending its stock tumbling. Apple, meanwhile, sits on abundant cash and can fund modest AI spend without compromising balance sheets and scramble for market share.

The real moat lies in context, not raw model power. Apple already harvests personal signals from 2.5 billion devices—health metrics, photos, messages, location—allowing on‑device inference that never leaves the handset. By licensing Google’s Gemini for cloud‑scale queries at a price that dwarfs competitors’ compute bills, Apple leverages existing data pipelines to deliver AI that feels private and integrated, a distinct advantage over open‑AI services on iOS.