Darkbloom starts paid OpenRouter inference on idle Macs

Darkbloom says it has moved from free usage to paid production traffic on OpenRouter, turning its distributed network of Apple Silicon Macs into a real inference provider rather than a demo pool.

The company says 250 Macs are online in apartments and offices, with 4.5B tokens served so far. It also says operators are earning 120-200 USD per month per Mac for compute that would otherwise sit idle. OpenRouter now lists Darkbloom as a provider, which gives developers a standard route to try its capacity alongside other model suppliers.

The bigger point is not the token count by itself. Darkbloom is trying to prove that consumer-owned Apple Silicon can become a paid inference layer for open models. Its site describes an OpenAI-compatible API, encrypted requests, hardware verification and a hardened runtime intended to keep prompts hidden from machine operators.

That makes the story useful to watch for two reasons: it is another attempt to push AI inference outside the usual GPU cloud stack, and it tests whether small distributed hardware owners can earn meaningful revenue from model traffic. The risk is also obvious: reliability, trust and sustained demand have to hold up once developers are paying, not just experimenting.