The MIT Technology Review has reported a groundbreaking achievement in artificial intelligence by Chinese startup DeepSeek. Despite stringent U.S. sanctions restricting advanced semiconductor exports, the company has unveiled DeepSeek R1, an open-source reasoning model that rivals—and in some cases outperforms—OpenAI’s ChatGPT o1.
This milestone underscores the resilience and ingenuity of Chinese AI researchers, who have transformed resource constraints into opportunities for innovation.
DeepSeek R1 excels in handling complex reasoning tasks across domains such as mathematics and coding. The model employs a “chain of thought” approach, breaking down queries step by step to deliver precise answers. Dimitris Papailiopoulos, a principal researcher at Microsoft’s AI Frontiers lab, lauded its simplicity:
“DeepSeek aimed for accurate answers rather than detailing every logical step, significantly reducing computing time while maintaining a high level of effectiveness.”
Notably, R1 operates at a fraction of the cost of comparable models, making it an invaluable tool for researchers and developers in resource-constrained settings, particularly in the Global South.
Innovation Under Pressure
DeepSeek’s rise is remarkable, given the increasing restrictions imposed by U.S. export controls on high-performance chips. These measures, designed to curb China’s AI progress, have instead driven Chinese startups to prioritise efficiency and innovation.
To circumvent hardware limitations, DeepSeek optimised its training process for Nvidia GPUs tailored to the Chinese market, which have capped performance. By innovating to reduce memory usage and computation times, the company created a model that matches or surpasses competitors without requiring high-end hardware.
“Rather than weakening China’s AI capabilities, the sanctions appear to be driving startups like DeepSeek to innovate,” said Matt Sheehan, an AI researcher at the Carnegie Endowment for International Peace, as cited by MIT Technology Review.
Open Source and Collaboration
DeepSeek has adopted an open-source approach, releasing six smaller R1 versions. These lightweight models, capable of running locally on laptops, include one that outperforms OpenAI’s o1-mini in specific benchmarks.
This ethos aligns with a broader trend in China’s AI ecosystem. According to a white paper by the China Academy of Information and Communications Technology, 36% of the world’s 1,328 large language models originate in China. Both tech giants like Alibaba Cloud and startups such as Minimax are fostering collaboration by open-sourcing their models.
The Vision Behind DeepSeek
Founded in 2023 by Liang Wenfeng, an alumnus of Zhejiang University, DeepSeek owes much of its success to Liang’s foresight. Anticipating U.S. export controls, he stockpiled Nvidia A100 chips—now restricted—and combined them with lower-power GPUs to develop R1.
Liang, who remains actively involved in research, also founded High-Flyer, a hedge fund backing DeepSeek’s efforts. His vision has propelled the company into the global AI spotlight.
A New Era for Chinese AI
DeepSeek’s achievement exemplifies a broader shift in China’s AI landscape, where constraints have spurred energy-efficient, cost-effective solutions.
As U.S. export controls intensify, Chinese firms are consolidating resources and forming strategic partnerships. For instance, Alibaba Cloud recently collaborated with Beijing-based startup 01.AI to launch an “industrial large model laboratory.” Such initiatives reflect a growing emphasis on cooperation.
“The U.S. export control has essentially backed Chinese companies into a corner where they have to be far more efficient with their limited computing resources,” said Matt Sheehan to MIT Technology Review.
DeepSeek R1 demonstrates the transformative potential of innovation born from necessity. This breakthrough not only solidifies China’s position in the AI arena but also paves the way for greater global collaboration in artificial intelligence.
