Idle Apple Silicon Macs can now generate passive income: the distributed network Darkbloom, run by Seattle-based Eigen Labs, has become a paid provider on OpenRouter. According to company estimates, Apple hardware owners can earn an average of $120 to $200 per month per connected machine by selling their unused compute capacity for AI inference.

Add AlexTech.ai asPreferred Source on Google

Macs as paid inference nodes on OpenRouter

Gajesh Naik, an engineer at Eigen Labs, confirmed that the network has already served approximately 4.5 billion tokens and generated an ARR (annual recurring revenue) of $102,000. Currently, around 250 Macs are online, while the total number of registered providers exceeds 900. The most requested model is Qwen 3.6 35B A3B, followed by GPT-OSS 20B and Gemma 4 26B. Average daily earnings range from $15 to $26 depending on the model and usage, with an estimated annual net income between $192 and $4,983 per participant, after deducting an estimated electricity cost of about $2 monthly.

Hardened architecture and privacy guarantees

The Darkbloom software is designed to mitigate security risks associated with running prompts and responses on third-party hardware. The inference engine operates within a single hardened Swift process, without subprocesses or local servers, leveraging mlx-swift-lm on the Apple Silicon GPU. The macOS kernel blocks external access to the process memory: debuggers are denied at the system call level (PT_DENY_ATTACH), and memory-reading APIs are blocked by Hardened Runtime. These protections cannot be disabled without a reboot, which terminates the process and erases its data. Plaintext is exposed only within the hardware-encrypted memory of a coordinator in a Confidential VM (AMD SEV-SNP) and is immediately re-encrypted for the provider, following a security model similar to Apple's Private Cloud Compute.

From free tier to decentralized data market

Darkbloom began as an experiment in April and has already processed over 42 million inference requests. The shift to a paid provider marks a turning point: revenue goes directly to owners of open-weight models on personal PCs, rather than to frontier labs with infrastructure debt. Hardware requirements are detailed in the official documentation, and Mac minis remain the optimal choice due to scarce inventory ahead of the M5/M6 refresh, but even an M1 MacBook Pro can contribute. The initiative fits into a broader trend of AI decentralization, as previously explored with the MLX framework that turns Macs into AI workstations and Apple's extension of Private Cloud Compute to external infrastructure.