Discussion

Darkbloom’s Mac-based AI inference network faces a tougher market

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#10314

Darkbloom has told participants that earnings from renting out Mac computing power for AI inference are low, and its researchers do not expect them to recover soon. The network is now exploring other work as falling token prices and more efficient data centres squeeze the economics of distributed inference.

Watch Desk analysis

What happened

The Register reports that around 800 people have used Darkbloom software to supply Mac computing power for AI prompts routed through OpenRouter. Participants have generally earned a few cents to a few dollars a day, the article says. Darkbloom estimates a suitably equipped Mac mini could earn up to $192 a month, but only with full, round-the-clock use for a sought-after model.

A researcher working on the project told participants that falling token prices and improving data-centre efficiency had eroded the network’s price advantage. Darkbloom is considering steering providers towards models in greater demand, finding buyers beyond OpenRouter and using its network for other computing work, including distributed CI/CD jobs. The company is also looking at clusters of high-end Mac Studios.

The Register’s report describes those alternatives as plans under consideration, not established new revenue streams.

Why it matters

Darkbloom offers a useful test of the idea that spare consumer hardware can compete for AI-computing work. A Mac can be capable enough to run inference and still struggle to earn much when model prices fall, large clusters become more efficient and routing systems weigh speed as well as cost.

For people considering lending their machines, the headline earning estimate is not a promise: it assumes continuous use, and the report says many participants are earning far less. The project’s possible shift into other workloads also shows that a distributed network may need more than one kind of compute job to keep its providers interested.

Our read

This is less a verdict on distributed inference than a reminder that the economics can change faster than the hardware. If you are thinking of joining a network like this, treat best-case earnings as a scenario, not a pay packet, and count the cost of giving up access to your machine. The market, as ever, has not agreed to pay extra for a charmingly decentralised business model.

What to watch

  • Whether Darkbloom finds buyers beyond OpenRouter.
  • Whether its proposed CI/CD work reaches testing and offers meaningful returns.
  • Whether providers leave the network as inference prices and routing priorities change.

Discussion spark: Should networks like Darkbloom be judged by whether ordinary Mac owners can earn reliably, or by whether they can find valuable uses for distributed computing beyond AI inference?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.