Chinese AI startup DeepSeek is accelerating the launch of its next-generation artificial intelligence model, R2, after its previous release triggered a massive sell-off in global stock markets.
The Hangzhou-based company had initially planned to debut R2 in early May but is now pushing for an earlier release, according to sources familiar with the matter. The new model aims to enhance coding capabilities and expand AI reasoning beyond English, marking another step in DeepSeek’s rapid rise in the global AI landscape.
DeepSeek gained international attention in January when its R1 model, built using less-powerful Nvidia chips, rivaled AI systems developed by U.S. tech giants that had spent hundreds of billions of dollars on their models. The development has fueled concerns in Washington, where AI leadership has been identified as a national security priority.
China’s government and businesses have swiftly embraced DeepSeek’s technology, with dozens of companies integrating its models into their systems. However, concerns over privacy and security have led some governments, including South Korea and Italy, to remove DeepSeek from their national app stores.
DeepSeek’s rise is largely attributed to its founder, Liang Wenfeng, a former quantitative hedge fund manager who built his fortune through High-Flyer, one of China’s most successful quant funds. Liang, known for his reserved nature, has steered DeepSeek more like a research lab than a traditional tech firm.
Unlike China’s major tech companies, which are often criticized for rigid hierarchies and demanding work cultures, DeepSeek operates with a flat management structure. Employees describe a collaborative environment with standard eight-hour workdays, a stark contrast to the “996” culture—working 9 a.m. to 9 p.m., six days a week—common in China’s tech industry.
DeepSeek’s strategy also differs from rivals like Baidu and Alibaba. While they focused on commercializing AI applications, Liang prioritized model refinement, an approach that has proven effective.
DeepSeek’s ability to develop competitive AI models at a fraction of the cost is rooted in High-Flyer’s long-term investment in AI research and computing power. Before DeepSeek was established, High-Flyer had already invested billions of yuan in AI infrastructure, including two supercomputing clusters powered by Nvidia A100 chips.
The accumulation of computing resources initially raised concerns among Chinese regulators, who questioned why a hedge fund required such vast AI capabilities. However, authorities allowed the company to proceed—a decision that later proved critical when the U.S. imposed an export ban on advanced AI chips to China in 2022.
DeepSeek’s reliance on efficient AI architectures, such as Mixture-of-Experts (MoE) and Multihead Latent Attention, has further reduced computing costs. Analysts at Bernstein estimate that DeepSeek’s pricing is 20 to 40 times lower than that of OpenAI’s equivalent models.
DeepSeek’s rapid rise has drawn support from China’s leadership. In January, Liang was invited to a high-level meeting with Premier Li Qiang as the representative of China’s AI sector—an endorsement that has accelerated state-backed adoption of its technology.
Chinese state-owned enterprises, including major energy companies, have integrated DeepSeek into their systems, while tech giants like Lenovo, Baidu, and Tencent have incorporated its AI models into their products.
However, as DeepSeek expands, it faces increasing scrutiny from Western governments. AI expert Stephen Wu warns that if DeepSeek becomes the dominant AI model for Chinese state institutions, regulators in the U.S. and Europe may tighten restrictions on AI chip exports and software collaborations.
For Liang, securing access to advanced chips remains the biggest challenge. “Our problem has never been funding,” he told Chinese media last year. “It’s the embargo on high-end chips.”
