In January 2025, Chinese AI lab DeepSeek unveiled its large language model DeepSeek-V3, marking a significant moment in the global AI landscape. The model was built using a fraction of the resources typically required by leading US companies like OpenAI and Meta, demonstrating a masterclass in the optimal use of limited resources. This development has challenged the long-held dominance of the United States in artificial intelligence and sparked a sense of urgency, particularly within the Trump administration. DeepSeek's technological feat has surprised Silicon Valley and the world, proving that innovation requires foresight as much as investment. The emergence of DeepSeek highlights China's growing influence in the AI sector, even amidst US export controls on advanced AI chips. These restrictions, intended to limit the scope of Chinese tech firms, may not have been as effective as intended, as DeepSeek's breakthrough demonstrates.

DeepSeek reportedly spent just $5.6 million to develop the DeepSeek-V3 model, a surprisingly low figure compared to the investments made by OpenAI, Google, and Microsoft. For example, OpenAI reportedly spent $100 million to train its GPT-4 model. Despite using GPUs considered last generation in the US, DeepSeek's results rival those of much more expensive models like GPT-4 and Meta's Llama. DeepSeek's success is a testament to its resource-efficient approach and challenges traditional assumptions about the costs and infrastructure required to build competitive AI systems. Nvidia, a leading AI chip provider, has praised DeepSeek’s innovative use of test-time scaling, a method that enhances performance during model deployment. This recognition underscores how DeepSeek’s strategy is redefining industry standards, with implications that extend far beyond China.

DeepSeek, based in Hangzhou, was founded in 2023 by Lian Wenfeng, who is also the co-founder of the quantitative hedge fund High-Flyer. Wenfeng's decision to work on Artificial General Intelligence was motivated by scientific curiosity rather than profits. He aimed to work with PhD students from China’s premier universities who were aspirational, and many of the team members had been published in top journals with numerous awards. DeepSeek's open-source nature, reflecting Wenfeng's ethos and belief system, has earned admiration from the global AI community. The company's ability to attract top-tier talent by offering competitive salaries and opportunities to work on cutting-edge projects has also contributed to its rapid rise. Government policies, generous funding, and a strong pipeline of AI graduates have further supported the emergence of companies like DeepSeek in China.

Google's CEO Sundar Pichai acknowledged the impressive work of DeepSeek's team. During Alphabet's Q4 2024 earnings call, Pichai noted that DeepSeek has a tremendous team and has done very good work. However, he also emphasized Google's focus on driving efficiency in serving AI models, highlighting Gemini's strengths in balancing cost, performance, and latency. Pichai pointed out that Gemini 2.0 Flash models are among the most efficient models available, including DeepSeek's V3 and R1. Google DeepMind CEO Demis Hassabis also praised DeepSeek's AI model as "probably the best work" from China, acknowledging its impressive engineering and geopolitical impact. Despite this praise, Hassabis suggested that DeepSeek's models do not represent any new scientific advances and are merely utilizing existing AI techniques, adding that the hype around DeepSeek has been "exaggerated a little bit".