Your MacBook Could Be Running AI Workloads While You Sleep
Apple ML Research has built a system called Minions that pools the idle GPUs inside MacBooks and Mac desktops to run the kind of AI tasks that normally require expensive server farms.
Key points
- Apple ML Research's Minions system, described in a new research paper, turns idle Apple Silicon GPUs across a team's Mac devices into a shared virtual GPU cloud.
- The project targets machine learning inference, the process where a trained AI model reads your input and produces an answer, which is the step that costs companies the most money and time.
- Minions runs workloads dynamically, finding spare GPU capacity across whatever devices are switched on at any given moment rather than reserving dedicated hardware.
- The system is aimed at teams and research labs rather than individual users, though it runs on the same consumer MacBooks and Mac Studios most professionals already own.
If your MacBook sits idle most of the day, doing a bit of browsing and email while its GPU barely breaks a sweat, Apple ML Research thinks that wasted capacity could solve a real problem for AI teams.
The problem is queue time. Companies running AI inference, the live process where an AI model takes in a question or image and generates a response, pay a lot for dedicated GPU servers. Demand spikes and researchers wait. It drops and those servers sit idle too, just expensively so.
Minions, a distributed system described in a new paper from Apple ML Research, proposes a different answer: aggregate the GPUs, the specialised chips that do the heavy number-crunching AI needs, that are already sitting inside the Apple Silicon Macs a team owns. Stitch them together into a virtual pool. Route inference jobs to whichever machines are free.
What does this look like in practice?
Picture a university AI lab where most laptops are plugged in and sleeping at 2am. Minions can wake the GPU portions of those machines and run a large inference job across all of them at once, finishing work that'd otherwise wait until morning for a turn on a shared server.
It's dynamic, meaning it doesn't require pre-assigning any machine to any task. Capacity flows to wherever the work is, based on what's actually switched on and available at that moment.
This connects to something we've been tracking. On 1 September 2026, we reported on start-ups paying people to share spare computing power from home machines. Minions is a different lever on the same frustration: GPU scarcity is real, and teams are looking for creative fixes.
Should ordinary Mac users expect anything to change?
Not yet. Minions is a research paper, not a shipping product, and it's designed for organisations that can deploy it across a fleet of managed devices. Your personal MacBook won't start processing a stranger's AI queries.
Privacy catch worth knowing: if an organisation does deploy something like this, your device's GPU would be processing data you may not see. Any real-world rollout would need clear policies about what workloads run on whose hardware.
On price, there's no product to buy today. The appeal of the concept is cost: the GPUs already exist inside hardware teams have already paid for. Running inference on idle Macs costs electricity, not cloud billing.
Nvidia has dominated the GPU conversation for years, and our coverage has tracked that closely. What makes Minions interesting is that it doesn't compete with Nvidia directly. It asks whether the Apple Silicon chips already in millions of offices can cover some of the overflow before anyone reaches for a cloud GPU at all.
That's a genuinely fresh angle. Whether the latency, the delay between sending a request and getting an answer, is acceptable for real inference workloads at scale is the question this paper will need to answer convincingly before anyone ships it.
Common questions
What is Apple Silicon?
Apple Silicon is Apple's family of custom chips, such as the M3 and M4 series, that power recent MacBooks, Mac Minis and Mac Studios. They include a built-in GPU that handles graphics and, increasingly, AI tasks.
Does this mean Apple is launching a cloud computing service?
No. Minions is an academic research paper from Apple ML Research, not an announced product. It describes a proof-of-concept system, not something you can sign up for today.
What is the difference between training an AI model and inference?
Training is the long, expensive process of teaching an AI model using huge datasets. Inference is the quick, repeated step of actually using that trained model, such as asking a chatbot a question. It's what Minions is designed to handle.
