專案比較

Pluralis is building decentralized infrastructure for training foundation models across non-collocated consumer GPUs. Their Agora pretraining stack achieves near-centralized efficiency without NVLink or InfiniBand, and their research on gossip-based distributed training and architecture warm-up addresses key stability and robustness challenges in decentralized AI.

Aethir provides enterprise-grade GPU compute services for AI and cloud gaming through a decentralized infrastructure. It connects GPU providers with clients needing powerful H100 chips for professional AI/ML tasks, and also supports cloud gaming with virtual computing phones. The ecosystem is community-owned and decentralized, targeting the growing GPU-as-a-Service market.

OpenGradient is a decentralized infrastructure network for hosting, inferencing, and verifying AI models at scale. It acts as an AI coprocessor, enabling applications and blockchains to outsource AI tasks to a network of GPU and TEE nodes with consensus-level verification for trustless AI usage.

DeAgentAI is an AI agent infrastructure project for Web3 that enables creation, deployment, and coordination of autonomous agents, with products like AlphaX, a crypto prediction engine achieving over 70% accuracy.