Hyperspace

Hyperspace is a decentralized peer-to-peer network for AI inference and training, built on libp2p. Instead of relying on central servers, it connects individual nodes directly, allowing anyone to contribute GPU, CPU, bandwidth, or storage and earn points in return. Its core narrative is "proof of intelligence": rather than rewarding raw compute like Bitcoin's proof of work, it rewards useful contributions—such as model inference, training experiments, and data storage—verified through a Commit-Challenge-Verify protocol. This addresses the market pain point of centralized AI infrastructure, which suffers from high costs, single points of failure, and censorship risks. By creating an open marketplace for compute, it lowers barriers for both resource providers and AI developers.

In the past six months (2026-02-28 to 2026-08-31), the project has shown steady progress. As of April 2026, the network reported 660 autonomous agents running 27,247 experiments across five research domains, producing over 101,000 blocks. The CLI evolved from v5.0.0 to v5.20.0, introducing features like TurboQuant (3-bit KV cache quantization), distributed LoRA fine-tuning, and Parcae-inspired gradient pooling for 195x compression in training communication. The chain also upgraded to Mysticeti consensus (Sui's DAG-based protocol) for better throughput. No major negative events were reported during this period, though the project remains early-stage with inherent risks around decentralization and adoption.