Eigen Labs unveiled Darkbloom, a peer‑to‑peer inference platform that taps the roughly 100 million idle Apple Silicon Macs on the planet. By routing requests directly to verified Mac Studio, MacBook Pro or Mac Mini nodes, the service mimics OpenAI’s API while eliminating the hyperscaler middle‑man. Internal tests show up to 70 % lower costs versus traditional cloud providers today.

Operators install a tiny launchd service, then earn USD for every token processed for each device. Because Apple Silicon’s power draw sits between $0.01 and $0.03 per hour, electricity is the only variable expense; the rest flows straight to the hardware owner. Darkbloom claims 95 % of revenue stays with the operator, a stark contrast to the three‑layer markup chain that inflates end‑user prices.

Privacy rests on four independent guarantees: end‑to‑end encrypted prompts, hardware‑bound keys stored in Apple’s secure enclave, a runtime hardened against debugging, and signed responses that publish a full attestation chain. No operator can inspect user data, yet developers can drop‑in the Darkbloom endpoint with existing OpenAI SDKs. The result is a low‑cost, private inference layer that runs on hardware many already own widely.