Dark Bloom: Rent Out Your Mac for AI Inference and Get Paid
Dark Bloom pays Mac owners to share idle compute for distributed AI inference. Here's how the network, privacy claims, and payouts actually work.
What is Dark Bloom?
Dark Bloom is a distributed inference network that lets Mac owners share unused compute with people who want to run AI models, in exchange for money. Instead of routing requests through a centralized data center, Dark Bloom splits inference across a network of individual Macs, each running open-weight models like Qwen, Gemma, and GPT-OSS. Anyone with a compatible Mac can install the software, enroll their device, and start earning based on how much inference work their machine handles.
TL;DR
- Dark Bloom turns idle Mac hardware into nodes on a peer-to-peer inference network, serving open-weight models such as Qwen, Gemma, and GPT-OSS to paying users.
- Enrollment happens through a CLI tool available at dark bloom.dev, with a Mac app reportedly in development, and payouts are handled through a connected Stripe account.
- The project claims a privacy-focused architecture where inference runs inside a single hardened Swift process using Apple’s MLX framework, aiming to prevent the machine’s owner from viewing prompts or responses passing through their device.
- Dark Bloom already serves models through OpenRouter, reportedly at lower prices than other providers, and had processed billions of tokens within its first week or so of public availability.
- Current hardware requirements reportedly call for at least 48GB of RAM, which rules out entry-level Macs but covers most Mac Studios, Mac minis, and higher-end MacBook Pros.
- The project is very early stage, self-described as days old at the time of its public attention spike, with an openly available codebase that outside reviewers can inspect.
- Earnings estimates vary by hardware. A high-memory Mac Studio configuration was cited as capable of earning somewhere in the tens of dollars per month, alongside a modest increase in electricity costs.
Other agents ship a demo. Remy ships an app.
Real backend. Real database. Real auth. Real plumbing. Remy has it all.
How does Dark Bloom actually work?
The basic mechanic is simple: you install software on your Mac, it enrolls your machine as a provider node, and the network routes inference requests to your hardware when someone needs to run a model. Your Mac’s GPU processes the request using Apple’s MLX inference engine, tuned for Apple silicon, and the results get sent back to whoever requested them. You get paid for the compute you contributed.
This is different from traditional cloud inference, where a company runs its own server farms and charges customers to access them. Dark Bloom instead aggregates thousands of individually owned machines into something that functions like a distributed data center, except there’s no single building, no centralized ownership, and the hardware belongs to regular people rather than a corporation.
The models currently served through the network are open-weight models: Qwen, Gemma, and GPT-OSS were named as examples. These are small enough to run efficiently on consumer Apple silicon while still being useful for a wide range of tasks. Dark Bloom’s inference is also available through OpenRouter, a marketplace that aggregates access to many different AI model providers, reportedly at a price point cheaper than competing providers on that platform.
How is user privacy protected on someone else’s Mac?
This is the part of the pitch that raises the most questions, and Dark Bloom has published a white paper addressing it directly. The stated problem: if you run inference on a stranger’s computer, that stranger has root access and physical possession of the machine. In theory, they could observe the prompts you send and the responses that come back.
Dark Bloom’s answer is architectural. According to the project’s own description, inference runs entirely inside a single hardened Swift process, with no subprocesses, no local server, and no interprocess communication that could expose data along the way. It uses MLX Swift LM, Apple’s inference library for the Apple silicon GPU, to keep everything contained within that one process. The goal is to eliminate the software pathways through which a machine’s owner could inspect what’s flowing through their own hardware during someone else’s inference request.
This doesn’t mean the system is provably unbreakable. It means the design intentionally narrows the ways data could leak, and the code is published for outside review. Independent AI-assisted code review, cited during early coverage of the project, reportedly found no hidden malware, no cryptocurrency mining, and no credential theft in the codebase, though that kind of automated review has real limits and shouldn’t be mistaken for a formal security audit.
Is Dark Bloom worth setting up on your Mac?
That depends on what you’re optimizing for.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
If you’re chasing meaningful income, temper expectations. Estimates for a well-equipped Mac Studio with a large amount of memory point to earnings in the tens of dollars per month, not a replacement for a day job. Apple silicon is efficient, so the incremental electricity cost of running inference in the background is small, but it’s not zero, and your actual profit depends on your local electricity rates and how much your machine gets utilized by the network.
If you’re interested in the underlying idea, distributed compute as an alternative to concentrated data centers, the calculus changes. Traditional AI infrastructure depends on massive, centralized facilities that draw enormous amounts of power and have become a point of local controversy in many communities. A network like Dark Bloom proposes something structurally different: computers already sitting idle in homes and offices doing useful work instead.
Two caveats are worth flagging. First, the project currently pays out 100% of revenue earned to node operators, but that revenue split isn’t guaranteed to stay that way as the business matures. Second, this is genuinely early software. Expect rough edges, hardware requirements that shift over time, and rapid changes as the team responds to demand.
What do you need to run a Dark Bloom node?
Enrollment currently happens through a command-line tool, downloadable from the project’s website, with a dedicated Mac app reportedly planned but not yet released. Setup involves creating a Dark Bloom account, installing the CLI, and enrolling your device through macOS’s device management settings. The process includes a hardware verification step that tests your machine’s inference speed in tokens per second before fully activating it as a provider node.
Hardware requirements have already tightened once. The project reportedly raised its minimum memory requirement to 48GB of RAM after facing more demand than expected, using the higher bar as a quality filter. That threshold excludes lower-memory Macs but still covers most Mac Studios, Mac minis, and higher-end MacBook Pro configurations.
Payment flows through Stripe rather than a direct connection to your bank account, meaning Dark Bloom itself doesn’t have direct access to your banking details, only the ability to send payouts through Stripe as an intermediary.
Frequently Asked Questions
What is Dark Bloom used for?
It’s a network that lets Mac owners share idle compute so others can run open-weight AI models like Qwen, Gemma, and GPT-OSS, functioning as a distributed alternative to centralized cloud inference.
How much money can you make running Dark Bloom?
Earnings depend on your hardware and how much the network uses your machine. A high-memory Mac Studio configuration was estimated at roughly tens of dollars per month, minus a modest increase in electricity costs.
Is Dark Bloom safe to install?
The project publishes its code openly for inspection, and early independent review reportedly found no malware, mining, or credential theft. That said, it’s an early-stage project without a formal third-party security audit, so anyone installing it should weigh that risk themselves.
Does Dark Bloom expose your data to strangers running your Mac’s compute?
The project’s stated architecture runs inference inside a single hardened process using Apple’s MLX framework, specifically designed to prevent the machine’s owner from observing prompts or responses. This is a design claim backed by a published white paper, not an independently verified guarantee.
What Mac do you need to join the network?
Current requirements reportedly call for at least 48GB of RAM, which covers most Mac Studios, Mac minis, and higher-end MacBook Pro models, though the threshold may change as the project develops.
