Cash in on the AI Boom by Renting Out Your Spare Compute

If your home server, gaming rig, or lonely laptop spends most of its life idle, a growing crop of companies would like to borrow it. Their pitch is simple: rent out spare compute for AI inference, the part where a pre-trained model answers a prompt, and get paid for the trouble.

The idea is showing up as a response to the giant data-center buildout behind the AI boom. Those facilities chew through electricity, pressure water supplies, create noise, and can be rough on the places that host them. Big model training is still likely to live there. But for smaller, often open-source models, companies are betting they can spread the work across homes and small businesses instead.

Far Labs, based in Abu Dhabi, is set to launch Far AI in the coming weeks. Evolving Edge, in Austin, Texas, is already in open beta. Bless Network, Salad, and Gradient have also entered the space over the last year. John Federico, who runs Evolving Edge, says he has spent years around computers and sees a simple fact: plenty of people already have powerful machines at home, broadband is widespread, and those machines could do useful work when they’re otherwise sitting still.

The selling point for hosts is that the setup is meant to be hands-off. You sign up, install an app, and let the platform run jobs only when you allow it. Federico says users can set a schedule, and the company says it only monitors resource usage. Far Labs says its system is built around least privilege, with isolated workloads, authenticated and encrypted communication, and explicit limits on GPU, CPU, memory, storage, and network use. Hosts can inspect resource use, pause the node, revoke access, and remove the software whenever they want.

The hard part is technical. Consumer devices are less powerful and less predictable than data-center hardware, so the companies lean on smaller models when they can, and split bigger inference tasks across multiple devices when they can’t. Evolving Edge uses Ray for that split. Far Labs says it has its own system that slices a model into pieces, distributes them, and recombines the results through an orchestrator and load balancer. Both companies argue that this can be cheaper than a traditional data center because there’s no big capital spend on the infrastructure.

They’re also selling resilience. A distributed network, they say, can shrug off failures better than one giant facility. Federico says that if the network is large enough, jobs can be routed to nearby devices, which can cut latency; Far Labs claims 100 milliseconds or less on its platform. Shazhaev also points to use cases that need fast, heavy responses, like in-game AI video generation. The bet here is not that data centers disappear. It’s that a lot of useful AI work never needed to live in one place to begin with.