The Chinese AI ecosystem has been boosted over the past year by the emergence of local company DeepSeek’s chatbot, which surprised everyone by matching the capabilities of its US competitors, such as ChatGPT, but at a lower cost and with less powerful chips.
A sign of the sector’s euphoria: two of China’s leading AI start-ups—Zhipu AI and MiniMax—made a big splash on the Hong Kong Stock Exchange in early January. However, Zhipu AI co-founder Tang Jie warned that despite Chinese advances in large open-source AI models, the gap with the United States “may actually widen.”
“Large-scale models in the United States are mostly closed. We must recognize the challenges and gaps we face,” he said at a recent conference in Beijing.
DeepSeek and its Chinese rivals have been betting on free, open-source technology, a strategy that quickly attracts users but generates less revenue than private, closed systems. In just two weeks, shares in Zhipu AI, which provides chatbots to Chinese companies, rose 80%, while MiniMax, which targets the consumer market with multimedia AI tools, soared 150%.
Read also: New AI Nexus Lab by NYP built on AWS Cloud makes artificial intelligence adoption accessible to SMEs
The US giants in the sector, OpenAI (ChatGPT) and Anthropic (Claude), are not yet listed on the stock exchange.
Despite a valuation that has exceeded USD 500 billion (EUR 426 billion), OpenAI does not expect to be profitable before 2029, due to the colossal costs of building its IT infrastructure.
Similarly, Zhipu AI and MiniMax are recording growing losses as costs increase, particularly for training new models: both “burn money faster than they can generate sustainable revenue streams,” said Poe Zhao, an expert and founder of Hello China Tech, quoted by Agence France-Presse.
For Zhao, 2026 will be a critical year for AI in China, which will have to prove whether companies “can go beyond code and generate real commercial value,” which is crucial for their survival.
“The challenge is not only technological. It is also the high cost of processing, in a context of sanctions, and the difficult balance between innovating within a strict regulatory framework,” explained Nick Patience, from the consulting firm Futurum, quoted by AFP.
Geopolitical tensions may slow down Chinese AI. Washington restricts the export to China of advanced microprocessors, particularly the most sophisticated ones from the American company Nvidia, and precision chip manufacturing equipment. When using chips produced in China, Chinese AI programmers need two to four times more processing power to train their models, according to Lian Jye Su, an analyst at Omdia.
Although Beijing distributes generous subsidies to stimulate innovation and rival the US, it also demands concrete applications of AI. In January, China announced its goal of implementing three to five large generalist AI models in the manufacturing industry by 2027, while also planning to increase available processing capacity.
China, which sees AI as an engine for the real economy, “is trying to build the world’s AI-powered factory,” Patience summarized. China still has the size of its domestic market, a vast pool of engineers, and moderate energy costs working in its favor, according to Tang Heiwai of the University of Hong Kong.
In the long term, “these factors will give China greater resilience than the US to assert itself as an AI superpower,” he argued.
