An article from Alfonso de la Rocha's Substack has exploded on Hacker News in recent days — 393 points and over 350 comments. The topic is simple, but the conclusion is counterintuitive: Apple, which almost everyone in the AI community dismisses as a laggard, might hold the strongest hand in the entire industry.
The argument goes something like this: While OpenAI, Anthropic, and Google compete to build the largest cloud models and the most expensive data centers, Apple has spent ten years weaving Neural Engine silicon into every iPhone, iPad, and Mac they sell. The M4 chip alone performs 38 trillion operations per second. That's no toy — it's serious inference capacity, and it's already in the hands of over a billion people.
What people are heavily debating in the thread is whether this actually matters in practice. Skeptics point out that Apple Intelligence is still embarrassingly weak compared to GPT-4o or Gemini Ultra. That's a fair point. But another faction responds: that's the wrong comparison. On-device AI doesn't compete on raw capacity — it competes on trust, speed, and availability.
Think about the use cases that actually matter to ordinary people: medical data you don't want to send to a cloud server, AI functions that work on a plane, things that happen instantly without network delay. Apple is the only company that can deliver this at scale because they control the entire stack — chip, OS, app framework, and user relationship.
From the research side, the numbers are actually quite impressive. The M4 performs ResNet50 image recognition in 15 ms versus the M1's 45 ms. The MLX framework is open source and specifically optimized for Unified Memory Architecture — something no one else has. And the entire chip uses 22W under load, which is ridiculously efficient compared to what a GPU cluster in the cloud consumes.
These are, of course, early signals from community sources, and HN is not peer review. But when 350 experienced engineers and product people spend a Monday debating this, it's worth paying attention.
The question hanging in the air: Will the next big AI shift be about who has the smartest model, or who has the most trusted, ubiquitous infrastructure? If it's the latter, Apple has already won — they just don't know it yet.