According to @DeepSeek_AI, DeepSeek has disclosed the training methods behind its advanced mixture-of-experts AI models, DeepSeek-R1 and DeepSeek-V3, achieving significant cost savings by utilizing 2,048 Nvidia H800 GPUs and implementing memory-efficient techniques such as FP8 precision (source: DeepSeek_AI Twitter, June 2024). These breakthroughs in AI training efficiency could lower entry barriers for blockchain projects leveraging AI, potentially reducing development costs for decentralized AI protocols and increasing competition in AI-powered crypto tokens. The adoption of advanced GPU hardware and novel optimization methods may also influence demand for related crypto infrastructure and layer-1 blockchain projects.