DeepSeek has forged its path in the AI sector by deploying cutting-edge models that require significantly less hardware than its competitors. For instance, reports suggest that
DeepSeek v3 was trained using 8-11 times less budget than typical models, specifically 16k H100s, leading to the reduction of GPU usage by a staggering 95% compared to traditional methods (
source). This efficient use of resources can fundamentally transform the AI landscape, presenting an existential threat to Nvidiaās AI supremacy.
The ability of DeepSeek to achieve such significant savings in AI training costs is creating new pathways for businesses to leverage AI without the monumental overhead associated with traditional GPU-based training. As highlighted by
Fortune, these types of breakthroughs have made investors and AI practitioners reconsider their approach to hardware investment. The overall aim here appears focused on efficiency over sheer power.