Apple's accidental moat: How the "AI Loser" may end up winning

The paradox of commoditized intelligence

A recent Substack post by ADLROCHA, amplified on Hacker News, argues that intelligence is rapidly becoming a commodity. Models improve, clones catch up, and the unit cost of deploying capability falls — including on modest, local hardware. It has been reported that increasingly capable open-source and mid-tier models (the author name-checks Gemma4, Kimi K2.5 and GLM 5.1) are already good enough for many real-world tasks. So what happens when the crown jewel — raw model supremacy — stops buying you the whole game?

Winners, losers, and runaway spending

While labs raced to burn compute and chase benchmarks, Apple stayed intentionally conservative. That pause looked like a loss. But it may have been an option. It has been reported that OpenAI’s video product Sora ran at eye-popping costs (allegedly $15M a day against roughly $2.1M in revenue) and that large infrastructure and wafer purchase plans with Samsung and SK Hynix were non-binding. It has also been reported that Micron reallocated consumer memory capacity toward AI demand and then reversed course when that demand softened. Those are the kinds of gambles that leave companies exposed — big upside, brutal downside.

The accidental moat and the question ahead

Here’s the emotional pivot: the company mocked as the “AI loser” might be the one best positioned for a world where intelligence is cheap but integration, privacy, hardware-software choreography, and distribution matter. Apple owns chips, OSes, an app ecosystem, and a consumer trust brand. When raw models commoditize, those assets aren’t worth less — they’re worth more. Will Apple move from patient to proactive? That’s the real cliffhanger.

Sources: adlrocha.substack.com, Hacker News