Darkbloom – Private inference on idle Macs
Darkbloom proposes an ambitious decentralized AI inference network, aiming to harness the collective power of idle Apple Silicon Macs for cheaper compute and passive income. This innovative model promises to cut out hyperscalers and offer 'verifiable private inference.' However, the Hacker News community intensely scrutinizes both the economic feasibility of its lofty earning estimates and the technical validity of its privacy assurances.
The Lowdown
Darkbloom introduces a novel platform designed to democratize AI inference by utilizing the vast, untapped compute power of idle Apple Silicon machines. By bypassing traditional cloud providers, Darkbloom seeks to dramatically reduce costs for AI users while providing a passive income stream for Mac owners. The project highlights a 'marketplace problem' in AI, where middlemen inflate costs, and positions itself as a distributed alternative akin to Airbnb or Uber.
- Decentralized Compute: Darkbloom aims to leverage over 100 million Apple Silicon Macs that sit idle daily, connecting them to AI inference demand.
- Cost Reduction & Operator Earnings: It promises users up to 70% lower AI inference costs, with operators retaining 95-100% of the revenue from their hardware.
- Verifiable Private Inference: The core security claim rests on a four-layer approach including end-to-end encryption, hardware-verified keys (Apple's secure enclave), a hardened runtime that blocks inspection, and a public attestation chain.
- OpenAI-Compatible API: The service offers an OpenAI-compatible API supporting streaming, image generation, speech-to-text, and various large language models.
- Earnings Estimates: Darkbloom provides significant earning estimates for Mac owners, suggesting thousands of dollars annually for continuous operation.
In essence, Darkbloom envisions a future where AI processing is distributed, affordable, and private, directly challenging the centralized hyperscaler model.
The Gossip
Profits and Payouts Perplexed
The most prominent discussion revolves around the profitability claims. Many users are skeptical of Darkbloom's earning estimates, with some performing back-of-the-envelope calculations suggesting significantly lower actual revenue than advertised. Critics question why Darkbloom wouldn't simply buy Macs itself if the profits were so high, while others point out that increased supply from a successful platform would inevitably drive down prices and, therefore, individual earnings. Some early adopters reported receiving zero inference requests, leading to doubts about demand, though supporters noted the platform's newness. The conversation also touched on the potential for this model to disproportionately benefit those in lower-income regions, who might find smaller earnings more significant.
Privacy Pledge Problematics
A deep dive into Darkbloom's privacy and security claims revealed considerable skepticism. Commenters challenged the concept of 'verifiable privacy' on current macOS hardware, noting that Apple Silicon lacks public SGX/TDX-style enclaves for arbitrary code. Concerns were raised about the MDM (Mobile Device Management) installation, with some fearing it could give the platform undue control over their machines, despite clarifications on minimal MDM permissions. Technical experts dissected Darkbloom's paper, pointing out potential flaws in its OS hardening approach (like the ability to patch macOS kernels or disable SIP) and a lack of verifiable binary attestation. One user performed an independent analysis of the installer, flagging issues like a non-notarized binary, MDM enrollment, and collection of serial numbers, concluding it was too risky to install.
Early Experiences and Expectations
Several users shared their initial experiences with Darkbloom, which were largely underwhelming. Reports included difficulties downloading and loading models, as well as a complete lack of inference requests after running the software for a significant period. Some attributed this to the platform's nascent stage, suggesting it needed time to build demand. There was also discussion about the chicken-and-egg problem of bootstrapping a marketplace, with some suggesting Darkbloom should generate its own requests to incentivize early participation. Despite these early hiccups, the underlying concept was praised for its potential to disrupt the centralized AI compute market, though many expressed a desire for it to be open-source rather than a private business.