DeepSeek, a Chinese AI developer, revealed in a peer-reviewed article that it spent only $294,000 to train its R1 model using 512 Nvidia H800 chips, a significantly lower cost than US rivals, sparking renewed debate over China's role in the AI industry and raising questions about the technology and costs involved in AI development.
China plans to deploy over 115,000 Nvidia AI chips across multiple data centers in Xinjiang, despite US export restrictions, highlighting ongoing efforts to advance AI capabilities and potential illicit procurement channels, amid geopolitical tensions and sanctions.
The U.S. Energy Department is developing a new supercomputer named Doudna at Lawrence Berkeley National Laboratory, set to arrive in 2026, which will integrate commercial AI technologies, notably Nvidia chips, to significantly boost computational speed and versatility for scientific research and AI model training.