The developers of DeepSeek assert that their model “leads the pack among open-source models and competes with the most sophisticated closed-source models worldwide.” App analytics firm Sensor Tower notes that its rapid rise has taken Silicon Valley by surprise, challenging the prevailing notion of US supremacy in artificial intelligence and raising questions about the effectiveness of Washington’s export controls on China’s advancements in AI and semiconductor technology.
Both DeepSeek-V3 and ChatGPT depend on high-performance chips for their training processes. Despite US restrictions on the export of advanced chips to China, DeepSeek researchers have recently revealed that their model was trained using Nvidia’s H800 chips, which reportedly cost less than $6 million. While the validity of these claims is contested, the company’s streamlined training methodology has sparked debate regarding the efficacy of export limitations aimed at curbing China’s AI development.
Founded in 2023 and headquartered in Hangzhou, DeepSeek is among several Chinese startups that have emerged in the AI space since Baidu launched the country’s first large-language model. However, DeepSeek stands out as the first Chinese AI model to achieve significant recognition within the US tech community, with some analysts suggesting it may rival or even exceed the capabilities of leading American models.
The app’s success highlights China’s increasing influence in the AI domain and indicates a potential transformation in the global AI landscape.