SkyCloud

SkyCloud is a decentralized cloud network that allows devices to share idle NPU power for collaborative AI training and inference.

Core Concepts

Decentralized Compute

SkyCloud turns every device into a potential compute node:

• Idle NPUs: Share when not in use
• Privacy Preserving: Federated learning, data never leaves device
• Token Rewards: Earn SkyTokens for compute contributions

Network Architecture

┌─────────────────────────────────────────┐
│           SkyCloud Network             │
├─────────────────────────────────────────┤
│  ┌─────────┐  ┌─────────┐  ┌─────────┐ │
│  │ Device  │  │ Device  │  │ Device  │ │
│  │  (NPU)  │  │  (NPU)  │  │  (NPU)  │ │
│  └────┬────┘  └────┬────┘  └────┬────┘ │
│       │            │            │      │
│       └────────────┴────────────┘      │
│                   │                     │
│         ┌─────────▼─────────┐         │
│         │   Orchestrator    │         │
│         │  (DeepComet-1B)   │         │
│         └───────────────────┘         │
└─────────────────────────────────────────┘

Use Cases

Collaborative Training

Train large models across thousands of devices:

# SkyCloud training job
job = SkyCloud.Job({
    model: "deepcomet-7b",
    dataset: "web_corpus_v3",
    nodes: 1000,
    privacy: "differential_privacy",
    reward: "sky_tokens"
})

job.start()

Inference at Scale

Distributed inference for low latency:

# Deploy model to edge
endpoint = SkyCloud.deploy({
    model: "deepcomet-1b",
    regions: ["us-east", "eu-west", "ap-south"],
    min_latency: "10ms"
})

Token Economics

Roadmap

1. Testnet: Q4 2026 — Beta network with 100 nodes
2. Mainnet: Q2 2027 — Production network launch
3. Scale: Q4 2027 — 10,000+ active nodes

The world's compute, shared.