**DeepSeek Trains AI Model for $294,000, Significantly Lower Than US Costs**
*By Akash Pandey | Sep 18, 2025, 06:57 PM*
Chinese artificial intelligence (AI) company DeepSeek has revealed that its R1 model was trained at a dramatically lower cost compared to US competitors. The company disclosed this information in a peer-reviewed article recently published in the academic journal *Nature*. This revelation is expected to fuel new discussions about China’s position in the global AI race and its transparency amid ongoing export restrictions.
### Cost Comparison Highlights Stark Contrast
DeepSeek’s reasoning-focused R1 model was trained for $294,000 using 512 NVIDIA H800 chips. This figure contrasts sharply with the estimated “much more” than $100 million that OpenAI CEO Sam Altman cited as the cost of “foundational model training” in 2023. While Altman’s company has not shared detailed expenses, it is well known that training large language models for AI chatbots generally requires running powerful chip clusters continuously for weeks or even months, resulting in very high costs.
### Controversy Over Chip Usage
DeepSeek’s claims regarding its relatively low development costs and chip technology have faced skepticism from US companies and officials. The H800 chips used by DeepSeek were specifically designed by NVIDIA for the Chinese market after the US government banned exports of its more advanced H100 and A100 AI chips to China in October 2022. DeepSeek insists it used only lawfully acquired H800 chips—not the restricted H100 models.
### Admission of Using Banned Chips in Development
However, in supplementary information accompanying their *Nature* article, DeepSeek acknowledged for the first time that it owns A100 chips and used them during early stages of development. The researchers explained: “Regarding our research on DeepSeek-R1, we utilized the A100 GPUs to prepare for the experiments with a smaller model.” After this initial phase, the full R1 model training was conducted over 80 hours exclusively on the 512-chip cluster of H800 chips.
### Innovative Training Methodology
The DeepSeek team shared insights into their unique approach to training the R1 model. They employed a system of rewards, mimicking how humans learn from experiences and mistakes. This strategy allowed them to overcome some of the costly computational and scaling challenges typically associated with teaching AI models human-like reasoning.
### Advancing AI Efficiency and Accessibility
DeepSeek’s work represents a major advancement toward making sophisticated AI systems more efficient and accessible. Their cost-effective training approach could reshape how next-generation AI models are developed globally, particularly amid ongoing geopolitical and technological tensions surrounding AI hardware access and export controls.
https://www.newsbytesapp.com/news/science/deepseek-trained-its-ai-for-294-000-vs-openai-s-100m-cost/story
About the Author
