Chinese firm claims it trained AI model for $3 million using 2,000 GPUs

A Chinese company, 01.ai, claims it trained its AI model, Yi-Lightning, for just $3 million using 2,000 GPUs. This contrasts sharply with OpenAI's reported spending of $80-$100 million to train GPT-4. Chinese tech firms face challenges due to U.S. export bans on advanced hardware, limiting access to high-performance GPUs. Despite this, 01.ai's CEO, Kai-Fu Lee, highlighted innovative engineering solutions that improved efficiency and reduced costs. Yi-Lightning is now ranked sixth globally in AI performance, demonstrating that 01.ai has managed to compete effectively despite limited resources. The company's inference cost is significantly lower than that of similar models.