Hancheng Cao, an assistant professor at Emory University, notes that DeepSeek represents a significant breakthrough for researchers and developers with limited resources, especially from the Global South. Despite facing US export restrictions on advanced chips, DeepSeek has propelled innovation within China’s AI sector, focusing on efficiency and collaboration. To develop its model R1, DeepSeek adapted its training techniques to lessen the demand on Nvidia GPUs, which are limited in performance for the Chinese market. Researchers commend R1’s capability to handle complex tasks in mathematics and coding, employing a step-by-step “chain of thought” method similar to ChatGPT’s. Dimitris Papailiopoulos from Microsoft's AI Frontiers lab highlights DeepSeek's engineering simplicity, emphasizing the model's focus on accuracy over detailing every logical step, thus optimizing computing time. DeepSeek also introduced six smaller versions of R1, with some reportedly outperforming OpenAI’s o1-mini, which they have openly sourced. Founded in July 2023 by Liang Wenfeng in Hangzhou, DeepSeek emerged from High-Flyer, a hedge fund Liang established in 2015. Liang aspires to develop artificial general intelligence (AGI), despite the challenges posed by the resource-intensive nature of training large language models (LLMs) within the constraints of US semiconductor sanctions. While large corporations dominate China's AI landscape, DeepSeek’s independence is notable as it does not seek additional funding.

Zihan Wang, a former employee, cites that staff have access to abundant computing resources, a rarity for new graduates. Liang has amassed a significant stockpile of Nvidia A100 chips before sanctions, allowing DeepSeek to leverage these resources effectively alongside lower-powered alternatives. Chinese firms generally face challenges due to less efficient engineering practices, requiring more computing power for similar outcomes. However, DeepSeek has found ways to optimize memory usage and accelerate calculations without compromising accuracy, showcasing adaptability in the face of hardware limitations. The trend toward open-source practices is growing among Chinese firms, with Alibaba Cloud releasing numerous AI models while startups like Minimax adopt similar strategies. A white paper from the China Academy of Information and Communications Technology reveals that 36% of the world's AI LLMs have emerged from China, establishing it as a key player in the sector. Matt Sheehan argues that US constraints have forced Chinese companies to optimize limited computing resources, potentially leading to industry consolidation. Recently, Alibaba Cloud partnered with the startup 01. AI to enhance collaboration, suggesting an evolving division of labor in the rapidly changing AI landscape.