Renting your Mac to AI could earn you $200 a month, but it's not for everyone.

Millions of Macs spend their days in standby, and this dormant power now attracts covetous attention. Eigen Labs, a Seattle startup, has been bringing together Apple Silicon machines within Darkbloom since April, a distributed inference network that has just become a paying provider on OpenRouter, the major marketplace for AI models. The company promises an average of $120 to $200 per month per machine.A French user wanted to verify it, backed by detailed logs, by plugging his Mac M5 Max in for a full day.

$4.11 in 24 hours: the promise holds, but only just

According to his logs, published with screenshots as evidence, the machine reportedly earned $4.11 in 24 hours, a rate of about $125 per month: the bottom of the advertised range, reached by the narrowest of margins. The stream, however, would never have stopped, even at 4 a.m., with 13,169 requests served and some 3.9 million tokens consumed, Gemma first, then GPT-OSS (the open models traveling through the network).

Yesterday I asked you for your predictions: how much does a Mac plugged in for 24 h earn on Darkbloom?

Everything is measured: $4.11 in 24 h, a rate of $125/month. Their promise of $120-200/month holds, just barely, at the bottom of the range. And only on MY machine.

The catch is in tweet 4. 🧵 pic.twitter.com/eImMlZF9mc

— Mickaël VILLERS (@MickaelV79228) August 25, 2026

As for the power outlet, the story borders on the anecdotal: an average of 33.7 watts and a peak of 109 watts, according to macOS software probes, meaning 0.79 kWh over the day and €0.20 at the French rate of €0.257 per kilowatt-hour. About 5% of the earnings therefore go to the meter, and the tester draws a blunt conclusion: on Apple Silicon, “electricity is a non-issue”.

The picture gets worse when you look at other people’s machines.While his M5 Max and 128 GB of memory were running without a break, another user’s 24 GB M3 would not have received a single job in eight hours. An M1 Pro left plugged in all night would, for its part, have remained at zero requests. Routing seems to prioritize large, proven configurations; the network’s advertised average therefore says nothing about what your machine will get (a nuance that weighs heavily in a potential buyer’s calculations).

The interested party nevertheless acknowledges having been wrong in his own estimates, which were based on generation speed. Real-world jobs mostly consist of digesting enormous prompts—up to 94,000 tokens, with payment calculated per token processed—an exercise in which the best-equipped chips soak up the work. Final caveats, and they matter: Qwen reportedly malfunctioned during the testing period, and the earnings were still sitting on Stripe at the time of publication, with the transfer requested but not yet received when we wrote these lines.

Under the hood is a distributed cousin of Apple’s Private Cloud Compute

The architecture is worth taking a closer look at, because it answers the uncomfortable question: would you entrust your prompts to a stranger’s Mac? At Darkbloom, inference runs in a single hardened Swift process, whose memory is locked by the macOS kernel: debuggers are refused, memory reads are blocked, and protections cannot be disabled without rebooting, which wipes the slate clean. Requests pass through a coordinator hosted in a confidential virtual machine, then are re-encrypted for the Apple chip responsible for the computation.Your Mac therefore works blindfolded: it opens the package in a sealed room for which even you do not have the key.

The principle bears a striking resemblance to Apple’s Private Cloud Compute, transposed to a swarm of computers owned by individuals. That swarm is indeed growing: more than 900 registered Macs (not all of which are active at the same time), more than 42 million requests served since April, and annualized revenue of $102,000 claimed by Eigen Labs. The terms of use nevertheless shut the door on the clever scheme of renting Mac servers in the cloud to sublease them to the network: personal machines only.

The public counters finally reveal a clear hierarchy: Qwen would generate around $26 per day for its best hosts, GPT-OSS $19, and Gemma $15. The annual range, meanwhile, would stretch from $192 to $4,983 depending on the configuration. In other words, the supplementary Mac mini and the 128 GB memory monster rarely compete on the same playing field.

👉🏻 Follow the latest tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.