DeepSeek's rise heralds seismic change in AI landscape

Overcoming US technology restrictions, Chinese companies lead wave of innovation

Competition is heating up in the fast-developing artificial intelligence sector as the emergence of startup DeepSeek underscores China's growing innovation capacities in cutting-edge technology, signaling a profound shift in the global AI landscape, experts said.

Despite tightened United States export controls on advanced AI chips, the meteoric rise of DeepSeek has challenged the belief that massive computing resources are essential for AI breakthroughs, they said. The emergence of players like Deep-Seek will also help bolster the evolution of the global AI ecosystem in an open, collaborative and inclusive manner, they added.

US containment of technology will not stymie China's innovation progress, but instead will accelerate Chinese enterprises' efforts to achieve technological breakthroughs, the experts said.

DeepSeek-R1, the latest opensource AI model developed by Deep-Seek, was recently launched with a performance on a par with leading models from US-based OpenAI in tasks such as mathematics, coding and natural language reasoning. However, it achieves these outstanding results at a fraction of the cost and computing power of its foreign peers.

The model has taken the world by surprise and sent shock waves through the tech industry. It quickly soared to the top of the Apple app store's free downloads, surpassing ChatGPT and securing the No 1 spot on the free app rankings in China.

Leading Chinese cloud computing companies, including Alibaba Cloud, Baidu AI Cloud, Tencent Cloud and Huawei Cloud, as well as China's three largest telecom operators, have all integrated DeepSeek's AI models into their platforms. The China-developed model has also attracted attention from US companies including Amazon, Microsoft and Nvidia.

Domestic breakthroughs

DeepSeek's achievements offer a glimpse into how Chinese tech companies are striving to make technological breakthroughs in large language models, or LLMs, and propelling the use of generative AI technologies in a wide range of sectors.

Alibaba Cloud, the cloud computing arm of Alibaba Group, recently unveiled its latest AI model, Qwen 2.5-Max. The company says it boasts enhanced capabilities in math and coding, and has outperformed other leading models like OpenAI's GPT-4o and DeepSeek's V3 model.

The Qwen model also offers a low-cost alternative to DeepSeek. US computer scientists have developed a new reasoning model that was trained for less than $50 with the help of Alibaba's open-source technology.

Researchers from Stanford University and the University of California, Berkeley, trained their AI models on the back of Alibaba's Qwen 2.5-32B-Instruct model, lowering the costs of AI training, according to a research paper published recently.

"We remain committed to investing in advanced AI infrastructure to foster the widespread adoption of generative AI technologies across different industries," said Zhou Jingren, chief technology officer of Alibaba Cloud Intelligence. The company is ready to empower developers and corporations of all sizes, enhance their abilities to leverage AI technologies, and further stimulate the growth of the open-source community, Zhou added.

Wu Yongming, CEO of Alibaba Group, stressed the importance of an open-source approach to LLMs, which will lower the threshold for the development and application of AI, greatly reduce the costs of computing power and boost the popularization of the state-of-the-art technology across a wide array of industries.

Open source allows researchers, developers and users to access the model's underlying code, and parameters, enabling them to use, modify or enhance the model to suit their needs.

At present, more than 300,000 enterprises have access to Alibaba's LLMs, which have a broad range of application in fields such as code programming, drug research and development, space exploration and manufacturing.

Baidu has unveiled the latest version of its LLM, Ernie 4.0 Turbo, which offers faster responses and improved performance in handling complex queries. Developers are also able to integrate this advanced technology via Baidu's Qianfan AI platform.

The Beijing-based tech company launched its LLM Ernie Bot in March 2023 after OpenAI unveiled ChatGPT in November 2022. So far, its AI-powered chatbot has garnered over 430 million users.

Robin Li, co-founder and CEO of Baidu, said the inference cost, or computational expense, of foundation models can be reduced by more than 90 percent over 12 months, and added that there's a correlation between cost decrease and productivity gains.

"If you can reduce the cost by a certain percentage, then that means your productivity increases by that kind of percentage. I think that's pretty much the nature of innovation," Li said, adding that he remained "optimistic about the future of AI", as even at the current level, LLM can create significant value in a range of scenarios.

The biggest difference between China and some Western countries in terms of AI, Li said, lies in its application, which is driving the rapid development of the AI industry in China.

Baidu has also recently launched the third-generation Wanka cluster based on its self-developed Kunlun chips, marking a milestone in the AI computing power field and significantly improving the training efficiency of AI models.

Wanka cluster is a high-performance computing system composed of 10,000 or more GPU computing accelerator chips that are mainly used to train and fine-tune AI models.

ByteDance, the owner of the popular short video app TikTok, released Doubao 1.5 pro, an upgrade to its flagship AI model, in late January, which it claims surpasses GPT-4o in categories such as coding, reasoning and Chinese language processing, while adopting a "resource-efficient" training approach that does not sacrifice performance.

Overcoming barriers

Bai Ming, a researcher at the Chinese Academy of International Trade and Economic Cooperation, said DeepSeek's rapid ascent has demonstrated that AI breakthroughs can be achieved through innovation and overcoming political barriers, and have overthrown the existing AI development paradigm that relies heavily on massive investments in infrastructure and hardware.

Washington's attempt to contain China's technological innovation and advancement "has proved to be completely futile and is bound to fail", Bai said, adding that "it will spur Chinese enterprises to double down on efforts on independent innovation, in order to make progress in crucial technologies."

Market observers said DeepSeek's success provides a new opportunity for international AI cooperation and showcases China's innovative strength and open attitude in the AI domain by providing a low-cost and highly efficient AI platform for global developers. This has also fostered the sharing of AI technologies around the world, as well as collaboration and innovation.

"Homegrown AI models have made significant breakthroughs based on low costs, high performance and open-source features despite facing limited access to advanced chips," said Pan Helin, a member of the Expert Committee for Information and Communication Economy, which is part of the Ministry of Industry and Information Technology.

"Chinese AI companies have the ability to take the lead in global AI innovation, as they have sought an alternative AI development approach that emphasizes efficiency and open-source collaboration, while reshaping the global AI landscape," Pan said.

These companies' technological breakthroughs prove that AI innovation is no longer exclusively dependent on significant computational resources, and can thrive even under restricted conditions. Pan stressed the need to foster international technological collaboration and innovation through open-source principles.

Former Google CEO Eric Schmidt described DeepSeek's rise as "a turning point" in the AI race. China's ability to compete with big US tech companies while using fewer resources demonstrates the need for the US to intensify its open-source AI efforts, Schmidt added.