THE MACSPARKY DISPATCH

Hi Friend,

Right now, the frontier AI models are, at best, getting $200/month from AI consumers. And Sam Altman has said publicly that OpenAI loses money on every one of those subscriptions. The compute bills are just that large.

Apple is doing something entirely different.

At WWDC this year, they started making the case for local AI. Not with a frontier model of their own. They don't have one, and they're not trying to win that race. What they did instead was double down on the hardware advantage they already have and start wiring their platform for AI in ways that keep the processing on your device.

Apple has over a billion active devices. OpenAI and Anthropic are spending enormous amounts of money trying to get in front of those people. Apple already has them.

Put simply, Apple believes the customer experience is more important than the model. If they can get workable local AI on your Apple devices, Apple figures they'll win whatever this is.

And they are slowly getting there. The M5 chip is genuinely capable for local AI, and that is as bad as it gets as the Apple chip designs march forward. Apple isn't matching what frontier models do in a data center. But the tasks most people actually need AI for, like drafting, summarizing, searching, and organizing, are getting within reach of what local models can handle.

Likewise, Apple didn't adopt the MCP standard but instead seeks to again be a gateway to its customers with its own APIs. What Apple did was open up App Intents and a handful of related frameworks so AI can privately reach into your apps (and data!) and do things. It's the same idea as MCP, just Apple's version.

Apple is trying to position its devices as the place where AI meets your actual life. Your health data, your messages, your documents, your photos. None of that leaves. The processing happens on-device. Third parties who want to reach you through AI do so via Apple's hooks, on Apple's terms.

And if local AI on Apple silicon gets capable enough, they don't have to route your requests through a frontier model at all. They need fast enough hardware, a capable harness, and a big enough installed base.

They have the user base, and they're on their way with the hardware and the harness. If AI compute truly becomes a commodity, Apple could come out of this looking really good. At least that is one possible outcome, and it increasingly appears to be the one Apple is betting on.

Your pal,
David
​

​Read this post on macsparky.com​​

This Week in MacSparky Labs

It was a full week. The Lab Report covered Apple's price increases across nearly the entire hardware lineup. I got into what went up, what held, and why the Apple TV jump caught me the most off guard.

The Q2 Q&A ran long in the best way: robot assistant daily use, the move to Circle, local models on the new MacBook Pro, and where I think the next year is headed. And Thursday's Jam Session on network setup and security turned into exactly the kind of member-led conversation I was hoping for.

Every week in the Labs runs like this. If that sounds like your kind of place, join us here.

​This week's newsletter is sponsored by iPhone Life.

There's no shortage of iPhone content out there, but most of it covers the same handful of features.

The Tip of the Day Newsletter by iPhone Life goes after the stuff that doesn't get written about as much — the buried settings, the gestures you walked right past, the features that have been sitting there for years.

It's free, it lands daily, and over 700,000 people read it. If you want to be a little more fluent in your own device, it's worth thirty seconds to sign up.