**DeepSeek Trains AI Model for $294,000, Significantly Below US Costs**
*By Akash Pandey | Sep 18, 2025, 06:57 PM*
Chinese artificial intelligence (AI) company DeepSeek has disclosed that its R1 model was trained at a fraction of the cost reported by US competitors. This information was revealed in a recent peer-reviewed article published in the prestigious academic journal *Nature*.
### Cost Comparison Highlights Stark Contrast
DeepSeek’s reasoning-focused R1 model cost just $294,000 to train, utilizing 512 NVIDIA H800 chips. This is drastically lower than the “much more” than $100 million reportedly spent by OpenAI on foundational model training in 2023, as stated by CEO Sam Altman. However, OpenAI has not released detailed cost breakdowns for its models.
Training large language models typically demands running extensive clusters of high-performance chips for weeks or even months, which contributes to the steep costs associated with AI development.
### Controversy Over Chip Usage
DeepSeek’s claims about its cost efficiency and chip technology have drawn scrutiny from US companies and officials. The H800 chips used were designed by NVIDIA specifically for the Chinese market following the US government’s ban on exporting its more powerful H100 and A100 AI chips to China in October 2022.
Despite the ban, DeepSeek insists it only employs lawfully acquired H800 chips and not the prohibited H100 models.
### Admission of Using Banned Chips in Early Development
In supplementary materials accompanying their *Nature* publication, DeepSeek acknowledged for the first time that it owns A100 chips and used them during the preparatory stages of development. The researchers stated, “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 R1 model was trained for 80 hours using a 512 chip-cluster composed solely of H800 chips.
### Unique Training Methodology
DeepSeek’s team also shared insights into their innovative training approach for R1. They implemented a reward-based system resembling human learning from experience and mistakes. This strategy allowed them to overcome many computational and scaling challenges typically involved in teaching AI models sophisticated reasoning skills.
### Significance
DeepSeek’s work represents a significant breakthrough in making advanced AI systems more efficient and accessible, potentially reshaping the competitive landscape of the global AI race. At the same time, it raises important questions about technology transparency and supply chain controls amid ongoing export restrictions.
—
*Stay tuned for further updates on AI development and industry insights.*
https://www.newsbytesapp.com/news/science/deepseek-trained-its-ai-for-294-000-vs-openai-s-100m-cost/story
