Why a network of Macs.
GPU marketplaces have tried for years to rent out spare GPUs, Macs included. Why would this work, and who uses the compute?
Cold start is the real problem.
A marketplace with nothing on either side has nothing to offer either side. Compute marketplaces mostly started with buyers in mind and asked owners to install server software, rent out machines and manage uptime.
MLX turns it around: a free app on a computer people already own, rewards from the first hour of work, and a checked job to do even before outside demand exists. Every Mac that joins makes the network more useful to the next one.
Data-centre uptime isn't the target.
Aggregators sell to buyers who expect a data centre's uptime from a single machine. A network of Macs works another way: jobs are small, each answer is checked by another Mac, and work goes to whichever Macs are free.
A Mac that closes its lid loses one job, not a customer. That suits work that can be split and checked; it doesn't try to be a server rack.
A closed loop.
The Macs that contribute compute are the same Macs whose owners use AI. MLX is built so they can serve each other: the people who run AI, those who lend their Macs to it, and the rewards between them on one public ledger, with no hyperscaler, data centre or model company in the middle.
Today Macs mine MLX for verified jobs and the Compute API (a pilot) pays USDC to the Macs that ran a job; paid AI jobs between Macs are planned.
For the people who use AI, MLX Node also finds the best AI for their Mac: the AI App Store lists AI apps, tools and open models, ranked only from public data. It is live on the web and in MLX Node, in a window of its own.
What nobody can know yet.
Nobody can size what a network of millions of Macs will do, the same way early networks couldn't be sized when they started. Chips change every year (M1 to M6 so far, and what comes next is unknown), and so does the software: MLX, open models, and ways to split work across machines that don't exist yet. The jobs the network runs today may look nothing like the jobs it runs in 2028.
Two companies make the chips that matter most for AI: one sells to data centres, the other puts its chips in the computers people already own. MLX builds on the second, and grows with every Mac that joins.
Join the Macs together first. What they're used for will follow.
None of this is a promise of demand, revenue or a price for MLX. See Risks.
