Darkbloom Warns of Plummeting Earnings as AI Inference Token Prices Collapse 77% in Five Months

Darkbloom Faces Revenue Crisis as AI Inference Market Shifts

Darkbloom, a decentralized AI inference network developed by Seattle-based Eigen Labs, has issued a stark warning to its 800 participants who rent out idle Mac compute capacity: expect plummeting earnings as the AI inference market undergoes dramatic transformation1

1

2

2

. The platform, which allows Mac users to rent out their GPUs for AI workloads through OpenRouter, an AI model gateway, has seen revenues collapse as declining token prices reshape the competitive landscape.

Drop in Anthropic's Token Prices Devastates User Income

The numbers paint a grim picture for participants. Anthropic token prices have crashed 77% in under five months, falling from $2.65 per million tokens on May 1, 2026 to just $0.62 on October 22, 20261

1

2

2

. This dramatic decline directly impacts what Mac users earn through the distributed inference network. Someone running an M5 Pro Mac Mini with 64GB of Unified Memory could previously earn up to $192 per month under optimal conditions. The median earnings among the top 50 operators on Darkbloom's leaderboard stood at approximately $159.86 per month, translating to $5.26 daily or $1,918.50 annually1

1

. Accounting for the 77% price collapse, those figures now plummet to a mere $36.77 per month, or approximately $1.23 per day1

1

.

Data Centers Flood Market with Cheaper Compute Capacity

Kydo, a researcher from Eigen Labs working on the platform, explained the structural forces driving this collapse in a message to Darkbloom participants on the company's Slack channel2

2

. "Data centers have become cheaper. Output per NVIDIA cluster is up approximately 20x this year, and their margins are wide enough to cut prices 90 percent and still profit at 50 percent," Kydo stated1

1

. The dramatic improvements in NVIDIA clusters have flooded the market with tokens, eliminating Darkbloom's previous pricing advantage. "The market caught up to our price," he acknowledged. "We could go lower, but then most of you would earn even less nothing"1

1

.

AI Inference Prices Fall Faster Than Moore's Law

The pace of price decline has been extraordinary. AI token prices are falling so rapidly they've outpaced Moore's Law, with some reports indicating AI is becoming 50% cheaper every quarter and up to 13 times more affordable annually1

1

. Ramp's AI Token Price Index confirms this trend, showing similar slippage across major providers. The decline has been driven partly by computational advances like DeepSeek's DeepInfer framework, which has transformed the economics of AI inference2

2

.

OpenRouter Prioritizes Speed Over Price

Adding to Darkbloom's challenges, OpenRouter, the platform's primary client, has shifted its routing strategy away from price-based selection1

1

2

2

. The AI model gateway now emphasizes decode speed, quantization, and prefill speed when routing inference jobs. This change disadvantages users without top-of-the-line Max or Ultra Apple Silicon chips, as older or less powerful Mac hardware struggles to compete on performance metrics. For many participants, "inference won't be worth running," Kydo warned2

2

.

Darkbloom Pivots Strategy Beyond AI Inference

Despite celebrating the processing of its trillionth token, Darkbloom acknowledges the need for strategic adaptation2

2

. The network is analyzing model demand to guide members toward higher-demand models and exploring workload diversification beyond traditional inference. The company is investigating distributed CI/CD jobs, prompted by observations that AI agents now generate pull requests faster than CI/CD pipelines can process them2

2

. This new direction will require sandboxing capabilities, which Darkbloom expects to begin testing soon. The platform is also focusing on clusters of high-end Mac Studio systems, now available with up to 512GB of RAM, utilizing Apple's Remote Direct Memory Access over Thunderbolt2

2

. Additionally, Darkbloom aims to improve network performance, particularly time to first token, and actively seek token buyers beyond OpenRouter to diversify revenue streams1

1

2

2

. Kydo concluded his message with measured optimism: "The next stretch will be a fight, and I expect many machines will leave and Slack will get quieter. But those who stay will be first in line when new earning opportunities come along. We are a driven team and well capitalized. We will not give up"2

2

.