If you are using AI tools in your business - ChatGPT, Claude, custom automations, AI-powered software - you are paying more than you probably need to.
Here is why, and here is where it might be headed.
The Hidden Markup in Every AI Call
When you pay for an AI API, the price goes through three layers of markup before it reaches you.
A chip manufacturer (NVIDIA) sells processing power to large cloud providers (Amazon, Google, Microsoft). Those providers rent it to AI companies. Those companies charge you. Each layer takes a cut.
That is not a complaint about those companies - that is just how supply chains work. But the result is that end users pay multiples of what the underlying compute actually costs to run.
Meanwhile, something significant has happened quietly: Apple has shipped more than 100 million machines running Apple Silicon chips - M1, M2, M3, and M4 series. Those chips have serious AI hardware built in: high-bandwidth unified memory, neural engines, and the ability to run large language models locally. And most of those machines sit idle for 18 or more hours a day.
What Darkbloom Is Building
A company called Darkbloom, out of Eigen Labs Research, launched this week with a simple idea: connect that idle compute to demand.
The model works like this. Mac owners install software that lets their machine accept AI inference requests when it is idle. They earn money. Businesses that need AI calls get them at lower prices - because idle hardware has near-zero marginal cost.
The numbers Darkbloom is reporting: up to 70% lower costs compared to centralized providers. Hardware operators keep 95% of revenue. Requests are end-to-end encrypted so operators cannot observe what is being processed.
The API is OpenAI-compatible, which means any software or workflow already using the OpenAI API can be redirected to Darkbloom with a single line of code change.
Why This Matters for Small Business
If your business spends $100 a month on AI API costs, a 70% reduction brings that to $30. At $500 a month, you go to $150. At $2,000 a month - which is not unusual for businesses running automations, customer service tools, or content pipelines - you are looking at $600 instead.
That is real money over a year.
The comparison Darkbloom uses is intentional: Airbnb turned idle rooms into an asset. Uber turned idle cars into a service. Rooftop solar turned idle rooftop space into an energy source. Darkbloom is pitching the same logic for idle Mac compute.
The Things to Understand Before You Switch
This is brand new. Darkbloom launched this week and is still in early access. The tradeoffs worth thinking through:
Latency may vary. Centralized providers optimize heavily for speed. Distributed networks can have more variability, depending on which machines are available.
Privacy and compliance. Requests are encrypted and operators cannot see them - but if your business processes regulated data (health information, financial records, legal documents), you should verify whether a decentralized network meets your compliance requirements before routing those calls through it.
Uptime guarantees. Large providers have SLAs (service level agreements) that commit to uptime. A peer-to-peer network does not carry the same guarantees.
It is OpenAI-compatible - but check your models. If you rely on specific models only available through OpenAI or Anthropic, not all of those may be available through Darkbloom yet.
The Bottom Line
AI compute costs are one of the fastest-growing expenses for small businesses using automation. They tend to creep up quietly - a few automations here, a customer service bot there - and then suddenly you have a $300 line item you didn't plan for.
Darkbloom is not plug-and-play for every use case yet. But it is worth knowing this infrastructure is being built. For businesses that are API-heavy and cost-sensitive, checking back in 60 to 90 days when the platform has more history makes sense.
For Mac owners: the earning side of this is an interesting experiment. If your M2 MacBook sits closed and plugged in eight hours a night, it could be doing something with that time. The math is: $0.01 to $0.03 per hour in electricity. Revenue potential depends on demand on the network.
The decentralized AI cost curve is bending. This is one of the first real signs of it.
- Platform: darkbloom.dev
- API compatibility: OpenAI-compatible (drop-in replacement)
- Reported cost savings: Up to 70% vs. centralized providers
- Status: Early access, launched April 2026
Source: Darkbloom - Eigen Labs Research - via Hacker News, April 16, 2026