Firm that lets you rent out your idle Mac for AI inference workloads warns of plummeting token costs — Darkbloom network sees earnings drop to $1.23 a day as Anthropic tokens fall 77%

Darkbloom, a decentralized AI inference network that aggregates and sells spare compute capacity from Macs, told its members to expect revenues to fall as token prices continue to drop. According to The Register, the price per million tokens for Anthropic dropped by 77% in under five months, falling to $0.62 in October 2026 from $2.65 in May.

“Data centers have become cheaper. Output per NVIDIA cluster is up ~20x this year, and their margins are wide enough to cut prices 90 percent and still profit at 50 percent,” said Kydo, a researcher from Eigen Labs, the developer of the compute platform. “The market caught up to our price. We could go lower, but then most of you would earn even less [than] nothing.”

It’s estimated that someone running an M5 Pro Mac Mini with 64GB of Unified Memory could earn as much as $192 per month. This is the best-case scenario, though; the median earnings of the top 50 operators on Darkbloom’s leaderboard are about $159.86 per month. This translates to about $5.26 daily or $1,918.50 annually. However, if we account for Anthropic’s 77% token price drop, earnings will fall to a measly $36.77 per month, or about $1.23 per day.

AI token prices are falling so fast that they've even outpaced Moore’s Law, with some reports saying AI is becoming 50% cheaper every quarter and up to 13 times more affordable every year. News of companies blowing through their annual AI budgets in months has less to do with per-token price and more with the number of tokens, as agentic AI use outpaces token price drops. Aside from that, OpenRouter, Darkbloom’s primary client, has started prioritizing decode speed, quantization, and prefill speed, putting users without a top-of-the-line Max or Ultra Apple chip at a disadvantage.

Despite that, Darkbloom says it will not sit idly by and let some users lose out because of these changes. The network is analyzing model demand to advise members on which models are in higher demand right now. It’s also looking at other kinds of workload beyond inference, including distributed CI/CD jobs, improving network performance to better compete against other computer providers/aggregators, and finding other clients outside of OpenRouter.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.