Eigen Labs, the team behind the Ethereum restaking protocol EigenLayer, published Project Darkbloom on April 15, 2026, opening a research preview at darkbloom.dev that allows owners of Apple Silicon Macs to contribute idle compute capacity to a distributed AI inference network. The project was authored by Gajesh Naik, a Senior Research Engineer at Eigen Labs, a two-time exit founder, and a former staffer at Solana Labs. The project targets a structural inefficiency: more than 100 million Apple Silicon machines have shipped since 2020, and most sit unused for 18 or more hours each day. Darkbloom is designed to put that stranded capacity to work running large language model inference for paying users, while keeping the underlying request data invisible to the machine running it.

Darkbloom is the latest initiative under EigenCloud, the developer-facing product layer that Eigen Labs launched on September 30, 2025, introducing EigenAI and EigenCompute as the primary interfaces for building on top of EigenLayer. Readers familiar with EigenLayer the restaking protocol should understand EigenCloud as the product layer that sits above it, translating protocol-level infrastructure into developer-facing services.

How the Privacy Architecture Works

The system relies on four stacked security layers. User requests are encrypted on the client device before transmission. The Mac running inference is verified through Apple's Secure Enclave, with an attestation chain anchored to Apple's root certificate authority, confirming that the hardware is authentic.

The runtime environment uses OS-level process locking to block debugger attachment and external memory inspection.

Finally, every response is signed by the specific machine that produced it using published attestation chains, creating a publicly verifiable record of which hardware performed each inference.

The project's documentation puts it plainly: "The operator runs your inference. They cannot see your data."

At launch, the network supports several open-weight models, including Gemma 4 26B (a multimodal mixture-of-experts model), Qwen3.5 27B, Qwen3.5 122B, MiniMax M2.5 (a 239-billion parameter mixture-of-experts model with 11 billion active parameters, noted for state-of-the-art coding benchmarks), and Cohere Transcribe for speech recognition.

The API is compatible with OpenAI tooling, supporting streaming responses via Server-Sent Events (SSE), function calling, image generation via FLUX.2 on Apple's Metal GPU framework, and speech-to-text through Cohere Transcribe.

The Economics for Operators and Users

Darkbloom's pricing runs at approximately 50% below major inference aggregators such as OpenRouter.

Project documentation credits structural efficiency rather than subsidies: no data center facilities, no cooling overhead, no networking overhead, and no layered reseller margins.

For Mac owners contributing compute, the project claims profit margins of roughly 90% after electricity costs.

The revenue share structure is listed as 95% on the EigenCloud blog but as 100% on the Darkbloom project site. Eigen Labs had not responded to a request for comment on this discrepancy at the time of publication, and potential node operators should treat the figure as an open question until the team clarifies.

Users pay per token with no subscription requirement and no minimum spend.

Where EigenLayer Stands Right Now

EigenLayer is a restaking protocol that allows users to "restake" their staked ETH, meaning that same capital secures multiple services simultaneously. Those services are referred to within EigenLayer as Actively Validated Services, or AVS.

The protocol held $18 billion in restaked ETH as of February 2026, across roughly 1,900 active operators, representing a 93.9% share of the restaking market. Competitors Symbiotic and Karak hold $897 million and $102 million in total value locked, respectively.

The EIGEN token is trading at approximately $0.1645 as of April 15, 2026, with a market cap near $113 million. That is a decline of roughly 97% from its all-time high of $5.65.

A token unlock of 36.82 million EIGEN (about 7.54% of circulating supply) occurred April 1, 2026.

Based on available project documentation, Darkbloom does not appear to introduce a new token; it is a research and product initiative built within the existing EigenCloud infrastructure.

What This Means for Developers Outside the US

Cloud compute costs consume between 50% and 70% of AI startup budgets. That burden falls hardest on developers in markets where USD-denominated pricing intersects with volatile local currencies, including India, Nigeria, Kenya, and Pakistan.

Darkbloom's pay-per-token model eliminates minimum spend requirements and provides access to frontier models at a lower baseline price. The model selection is particularly relevant outside the United States. Qwen (from Alibaba) and MiniMax are Chinese-origin open-weight models with strong multilingual performance across South and Southeast Asian languages, making them popular choices for non-English AI applications. Their open-weight availability also enables local deployment, which matters in contexts where reducing dependence on external cloud providers is a priority.

Because Darkbloom's API is OpenAI-compatible, teams already building on standard tooling can switch with minimal integration work.

One important distinction applies here: the earning opportunity (contributing idle compute for revenue) currently favors Mac owners in higher-income markets where Apple Silicon penetration is highest. Developers in Africa and South Asia are more likely to access Darkbloom as inference consumers rather than compute suppliers, at least in the near term.

Darkbloom carries the label "research preview" for substantive reasons. Its distributed architecture introduces known engineering challenges: network latency between nodes can equal or exceed computation time, and the system is specifically designed to address cold starts, disconnects, and variable node availability. Developers in regions with inconsistent broadband should factor these constraints into any evaluation.

On the regulatory side, data sovereignty laws are tightening across multiple emerging markets, including India's DPDP Act, Nigeria's NDPR, and Kenya's Data Protection Act. Darkbloom's hardware-backed design, where not even the node operator can observe inference data, offers a technically defensible privacy posture that centralized cloud APIs cannot match by design.

What Comes Next

Darkbloom enters a growing field of decentralized compute networks. The broader DePIN sector (Decentralized Physical Infrastructure Networks, meaning crypto-coordinated real-world hardware networks) grew from $5.2 billion to more than $19 billion in market cap within 12 months, a gain of roughly 265%, with AI-focused projects comprising 48% of that total. That expansion is set against a global AI compute market projected to grow from $9 billion in 2024 to $100 billion by 2032. An estimated 280-plus crypto-AI projects currently require trust-minimized model evaluation infrastructure, a demand pool that makes Darkbloom's positioning within EigenLayer strategically significant.

Existing players including Render Network, Akash Network, Aethir, and io.net compete on price and availability, but none specifically leverage Apple's Secure Enclave for privacy guarantees.

The codebase is open-source and a companion research paper has been published alongside the preview. As of publication, Eigen Labs has not announced a general availability timeline.