macOS is far more complex, and would have been even more so if they had modernized it with all its former built-in server parts and produced a professional suite of server tools which was the original objective after the us at NeXT merged with Apple.
The lack of EOF (standard and enterprise), PDO, D'OLE (which would have evolved to interface with Microsoft Services), a modern Mach microkernel instead of XNU, fully modern ObjC/up-to -date Cocoa with Swift as a training functional language later as co-equal, etc.
Lots of OS X features were halted to focus on iOS. When you do that personnel leave to work elsewhere on advanced kernels, operating systems, network protocols, etc.
At the time Apple paired down out of necessity. Now, they have no excuses. They need a far more robust open standards subsystem for services standards and extended developer language options because AI and HPC are massively in demand and 80%+ use fortran, and will never use Swift.
Fortran is king for scientific computation modeling. Nothing touches it. It's not an application, user space language. What it is exactly what they designed it to do: be the fastest, most scalable and precise mathematical processing language, bar none.
LLNL, CERN, all National Labs nationally and around the globe are steeped in Fortran. Nvidia, AMD, Intel, LLNL are the drivers behind LLVM Flang. All their AI initiatives and HPC are driven by it. Open AI: same, etc.
Apple should develop a full API to expose Metal ML and more to leverage it. As of now, Scientists on macOS need to use MacPorts and Home Brew to install it.
One can hope Apple develops tools to aide in its overall development within their extended AI vision.
AMD showed their MI350/355 and MI400 with their data center AIO solution, Helios in 2026, which includes UALINK (
https://ualinkconsortium.org/boardofdirectors/) which Apple has joined as a Director level contributor.
Ultra Ethernet will be next, both built into Helios. OpenAI made it clear they're all in on Epyc, MI latest and soon Helios as they also see open standards and not closed standards as their future--a shot at Nvidia.
UALINK does not include Nvidia.
I expect Apple will have Helios AIO in-house to train models to optimize along side their M Series processor designs.
Helios includes 72 MI400 series UDNA Instinct inference processors.
MI400: See attached slide
Apple will need these as they write their custom Fortran optimized stacks for the A series and M series processors.