Conjure cash with old Macs by linking them to AI inference Borg
Darkbloom distributed inference network promises to put underutilized Apple kit to work
Eigen Labs, a Seattle-based tech company, estimates that Apple Silicon device owners can earn between $120 and $200 per month by renting out their idle compute power for AI inference. Gajesh Naik, an engineer at the company, announced the Darkbloom project, a distributed network of Apple Silicon devices, on Friday. The network has already served 4.5 billion tokens and generated $102K in annual recurring revenue (ARR).
Naik stated that the network now consists of approximately 250 online Mac machines and that users can potentially earn $120-200 per month per machine. He encourages Mac owners to add their devices to Darkbloom via darkbloom.dev. Suitable devices include Mac mini, M1 MacBook Pro, and other Apple Silicon-equipped macOS devices. The network does not support cloud-provider rental arbitrage.
Darkbloom aims to address privacy concerns by running the inference engine directly within a single, hardened Swift process, with no subprocesses, local servers, or inter-process communication. The inference engine uses MLX on the Apple Silicon GPU and is protected by various security measures, such as denying external access to the process's memory, blocking memory-reading APIs, and preventing modifications without rebooting.
When an inference consumer sends a request over HTTPS via an OpenAI-compatible API, it is routed to a coordinator running in a Confidential VM. The coordinator decrypts the inference job within a Secure Enclave, runs it on the Apple Silicon GPU using MLX, and re-encrypts the plaintext for the selected provider. This ensures that the plaintext is only exposed within the coordinator's hardware-encrypted memory and is never logged or retained.
Darkbloom began as an experiment in April and has since grown to include over 900 providers, with more than 42 million inference requests served. The network shifted from free to paid inference services. Currently, the top-earning model is Qwen 3.6 35B A3B, priced at $399 per month of usage, followed by GPT-OSS 20B at $52 per month and Gemma 4 26B at $27 per month. Average earnings per year range from $192 to $4,983, after accounting for estimated electricity costs of about $2 per month.
Written by urgent.news from The Register Science's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
Also reported by 1 other outlet
- Conjure cash with old Macs by linking them to AI inference Borg theregister.com
