Darkbloom Pitches Idle Macs for Private Inference, and HN Tests the Trust Model

HN's Darkbloom discussion had the right kind of skepticism for a decentralized AI compute pitch. Darkbloom, an Eigen Labs research preview, describes a network where idle Apple Silicon machines serve private inference through an OpenAI-compatible API. The site says requests are encrypted before leaving the user's device, routed by a coordinator, decrypted inside a hardened process, and signed by the machine that produced the response.

The appeal is easy to see. Darkbloom argues that more than 100 million Apple Silicon machines sit idle for much of the day, while AI compute passes through GPU vendors, hyperscalers, API providers, and then end users. If already-owned Macs can safely handle useful inference, users could pay less and hardware owners could earn from machines they already have. The page claims up to 70% lower costs compared with centralized alternatives and highlights electricity costs around $0.01–0.03 per hour depending on workload.

HN did not stop at the pitch. Commenters questioned the revenue math: if a Mac mini can supposedly pay for itself quickly, why would the network not simply buy its own Macs? One user who tried the software reported health checks and attestations but no real inference requests during a short test, which pushed the conversation toward demand rather than supply. Others focused on the operator side, especially the trust implications of installing management software on a machine that might also be used for normal personal activity.

The privacy model drew the hardest technical questions. Community discussion noted that Apple's Secure Enclave is not the same as a general-purpose SGX, TDX, or SEV-style enclave for arbitrary code, so claims about verifiable private inference on consumer Macs need careful reading. That is the value of the thread: Darkbloom is an interesting attempt to turn idle local hardware into AI infrastructure, but HN treated privacy, attestation, and marketplace bootstrapping as the real product.