DeepSeek, a startup based in Hangzhou and founded in 2023, has sparked major upheaval in global stock markets and within the artificial intelligence sector by proving its innovative capabilities. Its new R1 AI model, which runs on locally manufactured Ascend chips by Huawei among others, marks a crucial milestone by demonstrating that high-performance AI can be achieved without reliance on Nvidia’s advanced hardware, challenging longstanding assumptions in the industry.
The technology has been rapidly embraced by a host of prominent Chinese companies. Telecom leader Huawei, for example, has adopted DeepSeek to operate on its domestically produced Ascend processors, while other key players such as Alibaba, Tencent, Baidu, and Bytedance have integrated the technology into their cloud services. This widespread adoption underlines a broader strategic trend where both chipmakers and cloud service providers incorporate advanced, open-source AI models to enhance their offerings and attract users.
DeepSeek’s impact is further underscored by support from leading Chinese AI chipmakers like Moore Threads, Tencent-backed Enflame, Baidu’s Kunlunxin, and Hygon Information Technology, all of which are now running the model on their respective hardware. This acceptance by major players and tech giants across different sectors reflects a remarkable level of confidence in the model’s performance and cost-effectiveness, especially when compared to established rivals like ChatGPT.
The global reverberations of DeepSeek’s success extend beyond China’s borders. While governments in Taiwan, Australia, South Korea, and Italy have raised data security and privacy concerns leading to bans or restrictions on the service, the model's widespread adoption by both American and Chinese tech giants suggests a robust endorsement of its technical prowess. Despite ongoing challenges such as a shortage of advanced chipmaking equipment due to US export restrictions, industry analysts note the impressive technical agility demonstrated by DeepSeek as it quickly adapts its models to operate on diverse chip architectures.