China’s Low-Cost AI Race Heats Up
One year after Chinese startup DeepSeek disrupted the global technology landscape with a budget-friendly artificial intelligence model, China’s AI sector is preparing for another wave of competitive launches. This time, DeepSeek will not stand alone. A growing number of domestic rivals are ready to unveil new models, many designed to attract broader consumer adoption.
DeepSeek’s rapid rise in early 2025 — during the Lunar New Year holiday — reshaped China’s AI ecosystem. By delivering a powerful, low-cost, open-source model despite U.S. export restrictions on advanced chips, the Hangzhou-based firm challenged assumptions about the level of computing power required to build cutting-edge AI systems.
As China approaches its longest annual holiday period, beginning February 15, several major players are expected to introduce upgraded products. Industry observers say expectations are now significantly higher.
“The real surprise would be if the new releases disappoint,” said Alfredo Montufar-Helu, managing director at Ankura Consulting in Beijing. “The market now expects strong performance.”
New Model Launches Gain Momentum
Zhipu AI this week introduced an updated model featuring stronger coding capabilities and improved performance on extended, autonomous tasks. Meanwhile, ByteDance unveiled Seedance 2.0, a video-generation AI system described by state media as capable of producing cinematic-quality content within seconds.
ByteDance is also preparing upgrades to Doubao, currently China’s most widely used AI chatbot, which has more than 155 million weekly active users, according to QuestMobile data.
DeepSeek itself is reportedly preparing its next-generation V4 model. Alibaba is expected to release its Qwen 3.5 series, which promises enhancements in mathematical reasoning and coding. Developer activity related to the upcoming Qwen models has already appeared on open-source platform Hugging Face, typically signaling an imminent launch.
While official release dates have not been confirmed, competition is intensifying across the sector.
Low-Cost, Open Source Becomes the Standard
DeepSeek’s original model launch in January 2025 had global repercussions. The announcement triggered a technology market selloff and erased hundreds of billions of dollars in market value from major semiconductor firms in a single trading session.
The startup’s claim that it could produce AI models comparable to leading U.S. systems — at a fraction of the cost — unsettled investors who had assumed that only companies investing tens of billions of dollars in computing infrastructure could compete at the highest level.
A recent RAND report on U.S.–China AI competition estimated that Chinese AI systems operate at roughly one-quarter to one-sixth of the cost of comparable American models.
Analysts say DeepSeek demonstrated that high-quality AI can be developed even under resource constraints. Its combination of open-source accessibility, advanced reasoning performance, and significantly lower deployment costs has influenced how Chinese firms now approach foundation model development.
Before DeepSeek’s breakthrough, several Chinese executives had argued that closed-source systems would dominate the industry. However, after DeepSeek’s assistant briefly surpassed ChatGPT in U.S. App Store downloads, major firms including Baidu began opening parts of their own AI models.
Today, Chinese companies — from established giants like Baidu, Tencent, and ByteDance to startups such as Moonshot — are among the most active contributors on open-source AI platforms.
Commercialisation vs. Pure Research
Although DeepSeek remains focused primarily on improving core model performance, competitors are increasingly turning their attention to consumer integration and monetisation strategies.
Alibaba, for example, has begun testing features that allow users to complete purchases directly through its Qwen chatbot. The shift reflects growing pressure on publicly listed companies to generate returns from heavy AI infrastructure investments.
DeepSeek’s structure gives it a different set of incentives. The company is backed by a quantitative hedge fund controlled by founder Liang Wenfeng, allowing it to prioritize research and long-term model development without the same level of shareholder pressure.
A More Competitive Phase
As China’s AI ecosystem matures, the market is entering a new phase marked by both imitation and differentiation. Open-source development and aggressive pricing are becoming standard practice, while firms compete to embed AI deeper into consumer services.
With multiple high-profile releases expected around the Lunar New Year, the coming weeks could signal whether China’s low-cost AI model strategy continues to challenge global industry assumptions — or whether expectations have now risen too high.
Source: (Reuters)