The @near_ai team ran major open-weight AI models by enabling TEE on H100s to test throughput for decentralized confidential machine learning. Overhead was as low as 1.5% for some models compared to not using TEE. This is definitely the most practical approach to decentralized and confidential ML right now
NEAR Protocol
NEAR Protocol24.6.2025
The race to own AI is on and Big Tech is winning. They want your data. Your value. Your control. FORK THAT. NEAR was built to flip the script: 🔓 Open-source agents 🤖 AI you control ⚡️ Infra that just works No middlemen. No friction. Just AI for everyone. Hold it NEAR. This is what it means to build the User-Owned Future:
Full DCML paper here, full of more important research findings for the future of user-owned AI:
19,77K