Chinese AI developer DeepSeek claims it spent only Ksh.37.9 million ($294,000) to train its R1 AI model, a figure significantly lower than estimates for U.S. rivals. This revelation, published in Nature, reignites debate about China's position in the global AI race and raises questions from U.S. officials regarding their technology access.

• DeepSeek's R1 model training cost $294,000 using 512 Nvidia H800 chips over 80 hours.
• DeepSeek acknowledged possessing A100 chips, used in preparatory stages, despite U.S. export controls on advanced AI chips to China.
• The company addressed accusations of "distilling" OpenAI's models, stating that their V3 model's training data incidentally contained "a significant number of OpenAI-model-generated answers."