Apple opens AI models to tech enthusiasts and plans AI server

Apple is now making its own AI language models available to technically minded users, and according to several media outlets it plans to build an advanced AI server for businesses, which isn't expected to arrive until around 2029.

The new operating system for Mac computers, macOS 27 (called Golden Gate), was released on September 14, 2026. It comes with a new tool called fm, which can be used in Terminal, a program where you type commands instead of clicking your way around. With fm, you can for example type fm chat to have a conversation with Apple's AI, fm respond to get quick answers that can be built into other programs, and fm schema to get answers in a specific, fixed format. The tool also works on iPhone with iOS 27, but it requires a newer Mac with Apple's own chip on which the Apple Intelligence feature is turned on.

Behind the tool are two versions of Apple's AI model: a smaller model designed to run quickly and power-efficiently directly on the device, and a larger, more advanced model called AFM Core Advanced, which requires more memory to use. Rumors have circulated that the models are based on Google's AI model Gemma, but no evidence has been found for that.

Developers who use the programming language Python can also access the models through a special toolkit from Apple, which however requires a newer version of Python and Apple's developer program Xcode.

Danish language support already in place

For Danish users, the good news is that Apple Intelligence has been able to understand and respond in Danish since fall 2025. This means the new fm tool should in principle also be usable in Danish, provided you have a compatible Mac and have the feature turned on. Some Apple Intelligence features are still limited in the EU for regulatory reasons, however, and it's unclear whether that also applies to the fm tool.

Apple returns to the server market

At the same time, several media outlets, including The Information, Ars Technica and The Verge, report that Apple is returning to a market the company left almost 20 years ago: selling servers to businesses. According to the reports, Apple is working on a powerful AI server built on the company's upcoming M8 Ultra chip, and there has reportedly also been contact with the chip company Nvidia about linking several chips together so they can compute even faster together.

According to the media, the background is the growing demand for computing power for AI, which among other things has made it hard for developers to get hold of Apple's existing Mac Studio and Mac Mini computers. The project reportedly received backing about a year ago from John Ternus, who at the time headed Apple's hardware development and is today the company's CEO.

It's important to remember, however, that Apple itself has not confirmed the plans. There's no official date, price or product name, and according to the media, Apple could still choose to drop the project or not use Nvidia's technology. In any case, a possible launch is several years in the future, around 2029.

What it means for you

If you're a technically curious Mac user, you can now test Apple's AI yourself directly from Terminal, including in Danish. If you have a newer Mac with an Apple chip, you can try it right away by turning on Apple Intelligence and typing fm chat. If you develop programs yourself, you can explore the Python toolkit if you have Xcode and an up-to-date Python version. The server plans you can safely sit back and follow from a distance: they're only rumors, and it will be several years before anything might land.