Earn Cash with Old Macs Linking Them to AI Inference Services
Get Paid for Doing Nothing
Looking for some additional income from that Apple Silicon gathering dust? Eigen Labs, a tech company based in the United States, claims you can earn a decent amount—between $120 to $200 a month—by selling your unused computing power for AI inference. One of their engineers, Gajesh Naik, revealed that their Darkbloom project, a network of Apple devices, is now a monetized solution on OpenRouter. He encouraged Mac owners to contribute their machines. “We’ve processed 4.5 billion tokens and achieved $102K ARR,” Naik mentioned online. “We’re an open Mac network (currently 250 online). People are earning $120-200 a month per device. Have a Mac lying idle? Visit darkbloom.dev.”
Mac Mini Craze
Mac mini devices have become incredibly rare following the OpenClaw buying surge, with many waiting for the M5/M6 update. But if you possess an M1 MacBook Pro or any Apple Silicon devices lying around, you might as well put them to use, according to the hardware specs. Rented cloud machines aren’t permitted — those are the guidelines — but you can quickly establish your own setup with the Darkbloom software. It’s crafted to manage privacy concerns associated with running AI inference on external hardware.
Privacy Protections
“The inference engine operates within a single secured Swift process – no subprocesses, no server complications, no inter-process communication – using mlx-swift-lm on the Apple Silicon GPU,” Naik clearly states in a technical document. “macOS restricts all external access to this process’s memory: debuggers are blocked at the system call level (PT_DENY_ATTACH), memory readers face obstacles with Hardened Runtime, and these safeguards can’t be disabled without rebooting the entire system, which terminates the process and clears its data.” Essentially, an inference requester sends a command via HTTPS through an OpenAI-like API to a server running a Go-based coordinator in a Confidential VM (AMD SEV-SNP). It then assigns the task to an inference provider – someone with an Apple Silicon Mac and Darkbloom’s Swift CLI. The provider connects out via WebSocket, avoiding port or firewall issues, decrypts the task in a Secure Enclave, and proceeds to execute it using Apple’s machine learning framework. The plaintext only appears in the coordinator’s encrypted CVM memory, is never logged or stored, and is swiftly secured again for the provider, or so the documentation indicates.
The Journey of Darkbloom
Darkbloom started as a small project in April and now has over 900 providers (though not all are active consistently) who have managed over 42 million inference requests. They previously provided free inference power through OpenRouter but have since become a professional service. The highest earner on their roster at the moment, per Darkbloom’s statistics, is Qwen 3.6 35B A3B, generating $399 a month, followed by GPT-OSS 20B at $52 a month, and Gemma 4 26B at $27 a month. Usage isn’t always continuous, so top earners currently average $26/day, $19/day, and $15/day, respectively. They estimate that network participants can earn between $192 to $4,983 annually, factoring in an average electricity cost of nearly $2 a month (not to mention the ongoing fan noise). Regardless of the payment, it’s income for those running open weight models on their personal machines as opposed to large labs still contending with data center expenses and trying to market their high-end models.
Counting Earnings, Not Sheep
So there you have it. You don’t have to work late nights or make sacrifices to earn some extra cash. Just let your Mac do the work while you catch your favorite shows. It seems like an obvious choice to me, especially for a little extra pocket change. Visit GadgetLad for more useful tips like these!
