// 9TO5MAC — MOBILE & WEB
Apple @ Work: The spiritual successor to the Xserve is here, thanks to Apple silicon and MLX
Apple @ Work is exclusively brought to you by Mosyle, the only Apple Unified Platform. Mosyle is the only solution that integrates in a single professional grade platform all the solutions necessary to seamlessly and automatically deploy, manage, and protect Apple devices at work. Over 45,000 organizations trust Mosyle to make millions of Apple devices work ready with no effort and at an affordable cost. Request your EXTENDED TRIAL today and understand why Mosyle is everything you need to work with Apple.
I’ve spent the past few weeks thinking about how Apple built up the best deal in enterprise AI. Two things happened that make me think Apple knew that all along (or maybe they didn’t and got lucky). Meta released its most capable open-weight model yet and optimized it to run on a Mac Studio. I also spent time with Osaurus, an app that runs a usable local AI model on something as small as a 16 GB Apple silicon laptop without needing a network connection. They even have two models that will run on 8 GB.
The Xserve has been dead since 2011, but its spiritual successor sits on a desk rather than in a rack.
About Apple @ Work: Bradley Chambers has been an Apple IT admin since 2009. Through his experience deploying and managing firewalls, switches, a mobile device management system, enterprise grade WiFi, 1000s of Macs, and 1000s of iPads, Bradley will highlight ways in which Apple IT managers deploy Apple devices, build networks to support them, train users, share stories from the trenches of IT management, and ways Apple could improve its products for IT departments.
For the last few years, the narrative has been that Apple fell behind in AI because it didn’t have a frontier model. Siri AI got delayed, and the Apple Intelligence rollout was not great. In hindsight, I think Apple was right not to run that race, as the frontier-model business has turned into somewhat of a capital bonfire. Every frontier lab is spending billions on GPUs to be the best for a few months at a time, and the moment a model is good enough, someone else releases an open-weight version that runs on hardware you already own.
Open-weight models prove it: Muse Spark Open has 816 billion parameters, 42 billion active parameters, and a million-token context window, and Meta is giving it away. Apple didn’t build that LLM. It built the hardware that runs it. With that being said, Siri is still far behind.
Look at what Apple has shipped while everyone was waiting on them to try to compete with Anthropic or OpenAI. With Apple silicon’s unified memory, the GPU can use up to 512 GB of memory on a single Mac Studio, instead of being limited to whatever fits on a graphics card.
MLX is Apple’s machine learning framework, and it’s built specifically for the unified memory architecture of Apple silicon. If you’ve used NumPy, the API will feel familiar right away. On top of that, MLX includes higher-level packages for neural networks and optimizers, along with function transformations for automatic differentiation and graph optimization, so you can build more complex models without giving up efficiency. It also has bindings in Swift, C++, and C, which means it runs on any Apple platform.
Apple’s own Machine Learning Research post shows four M5 Ultra Mac Studios running Muse Spark Open at over 36 tokens per second, which is faster than you can read, while peaking around 300 watts per machine.
Apple’s unified memory architecture lets the M5 Ultra’s integrated GPU use up to 512 GB per Mac Studio, so two machines can hold an entire model. And the chip’s 1.2 TB/s of memory bandwidth is fast enough to read the small slice each token needs. The result is two Mac Studios behave like one supercomputer with a terabyte of unified memory.