Mac owners earning money by renting out their unused computing power for artificial intelligence workloads are facing declining returns as the cost of running AI models continues to fall.
Darkbloom, a decentralized AI inference network developed by Eigen Labs, has warned participants that earnings may remain low as large data centers become more efficient and competitive.
The company allows people with compatible Apple computers to contribute spare GPU processing capacity to a shared network. Those machines handle AI inference requests, generating payments for their owners.
However, the economics of this model are changing. Anthropic's average token prices reportedly declined by 77% between May and October 2026, while major AI infrastructure providers continue reducing their operating costs.
An estimate based on those price changes suggests that earnings previously equivalent to approximately $160 per month could fall to about $37 per month, or $1.23 per day.
AI Token Prices Have Fallen Sharply Since May
The cost of processing AI tokens has declined considerably over the past several months.
Tokens are the units used to measure text processed or generated by language models. Companies typically charge customers based on the number of input and output tokens consumed.
For Anthropic, the reported average price per million tokens declined from $2.65 on May 1 to $0.62 on October 22, 2026.
That represents a reduction of approximately 77% in less than six months.
One reason is the improving efficiency of large AI data centers, which can process more requests using increasingly optimized hardware and software.
A researcher involved with Darkbloom estimated that output per NVIDIA GPU cluster has increased approximately twentyfold during 2026.
The researcher also argued that major infrastructure providers have sufficient margins to reduce prices substantially while remaining profitable. These figures are the researcher's assessment rather than independently established industry averages.
For smaller computing networks, falling prices make it harder to compete with centralized infrastructure.
Darkbloom Earnings Could Fall to $1.23 Per Day
Darkbloom operates a distributed network of approximately 800 participants who contribute computing resources from compatible Mac systems.
The platform has previously advertised potential earnings of up to $192 per month for an M5 Pro Mac mini equipped with 64GB of unified memory.
That estimate assumes continuous operation and strong demand for the AI models running on the device.
Actual earnings vary considerably depending on hardware performance, network demand, and the number of requests assigned to a computer.
The median earnings among the platform's top 50 operators were approximately $159.86 per month.
Applying a 77% reduction to that figure produces an estimated monthly return of $36.77.
| Metric | Reported amount |
|---|---|
| Anthropic average token price in May 2026 | $2.65 per million |
| Anthropic average token price in October 2026 | $0.62 per million |
| Token price decline | 77% |
| Previous median monthly earnings among top 50 operators | $159.86 |
| Estimated monthly earnings after equivalent reduction | $36.77 |
| Estimated daily earnings | $1.23 |
The $1.23 daily figure is a projection based on falling token prices, not a confirmed average payout across the entire Darkbloom network.
Actual returns also depend on demand, workload allocation, electricity expenses, and hardware capabilities.
OpenRouter Is Prioritizing Faster AI Processing
Darkbloom faces another challenge as OpenRouter, one of its main customers, changes how it selects AI processing providers.
Instead of focusing primarily on low pricing, OpenRouter is placing greater importance on processing performance.
This includes decode speed, which measures how quickly a model generates output tokens, and prefill speed, which affects how quickly it processes incoming prompts.
Quantization efficiency is also becoming more important because it allows AI models to operate with reduced memory requirements.
These changes could disadvantage older Mac systems or models without powerful Apple Silicon Max and Ultra processors.
As a result, contributing an ordinary Mac to the network may become less financially attractive, even when the device is otherwise unused.
Darkbloom Is Exploring Other Ways to Generate Revenue
Despite declining inference earnings, Darkbloom is investigating ways to improve its business model.
The company is studying demand for individual AI models to help participants choose workloads with better earning potential.
It is also exploring distributed software development tasks, including continuous integration and continuous delivery workloads.
Additional efforts include improving network performance and attracting customers beyond OpenRouter.
These changes could help the platform diversify its revenue sources, although their financial impact remains uncertain.
The situation illustrates a growing challenge for individuals hoping to generate passive income from consumer computing hardware. As AI infrastructure becomes more efficient, the value of selling spare processing capacity may continue to decline, particularly for devices that cannot match the performance of dedicated data center systems.
Discussion (0)
Be the first to comment.