China’s AI Models Creating Buzz and New Winners
China’s affordable artificial intelligence models are rapidly gaining traction worldwide, leading to emerging success stories in the stock market.
As AI development expands to cover a range of tasks, a new economy is forming around “tokens,” the essential units of data in large language models. The global uptick in token usage is providing China with a competitive edge, thanks to its low-cost electricity and a wealth of AI developers vying for innovation.
Smaller startups are attracting investor interest over traditional technology giants, as the AI sector holds more growth promise. Newly listed companies like MiniMax Group Inc. and Knowledge Atlas Technology Joint Stock Co., known as Zhipu, have already exceeded a market valuation of $40 billion, outpacing bigger players like Baidu and Kuaishou.
Even major firms like Alibaba have seen significant jumps in stock prices, despite broader losses. Victoria Mio, a portfolio manager at Janus Henderson Group, describes this shift as a structural revaluation rather than a fleeting trend. “Investors are responding to visible token usage and the emerging viability of AI as a monetizable activity,” she noted.
Mirroring last year’s trends, investors are banking on the idea that China’s cost efficiency will position it as a long-term leader. The government is leveraging this momentum to enhance the international competitiveness of its tech sector, with “token exports” becoming a recurring theme in state media.
Understanding Tokens
A token represents a small piece of data—about four characters. While AI firms typically provide low-cost or free monthly plans, the token limits can be restrictive. Developers exceeding those limits pay based on their token consumption, with output tokens sometimes costing several times more than input ones.
Tokens are now essential in the AI economy, serving not just as a transaction unit, but also as a benchmark for measuring developer productivity within technology firms.
Data from LLM distribution platform OpenRouter indicates that the top three models in token usage recently are all from Chinese brands: Xiaomi’s MiMo, Alibaba’s Qwen, and DeepSeek. Notably, MiniMax has two models in the top ten, and the smaller company StepFun has entered the mix.
Investment manager Bush Chew from Aberdeen Standard Investment believes that “China has the edge in providing tokens and models to global users,” pointing out that constructing data centers in China is significantly cheaper than in the U.S.
MiniMax, remarkably, ranks fourth in market share as of mid-April, trailing only major U.S. companies like Alphabet and OpenAI. Its models cost about $1 per million output tokens, far lower than Google’s and Anthropic’s offerings.
Shares of Zhipu climbed as much as 16% recently, supported by the OpenRouter report, while MiniMax rose by 7.1%. The news that DeepSeek aims to raise at least $300 million has also contributed to the positive sentiment.
Goldman Sachs recognizes MiniMax as a leading contender among Chinese AI modeling firms poised for growth, citing its strong commercialization and comprehensive product offerings.
Both MiniMax and Zhipu are part of a group dubbed the “Six Tigers,” which includes startups providing specialized AI services often supported by larger corporations. Alibaba maintains a stake in MiniMax.
Having gone public in Hong Kong in January, Zhipu’s stock has surged nearly 750%, while MiniMax has increased by 440%. In contrast, Alibaba’s stock has fallen about 3.6%, and Tencent is down 12% in the same period.
The Rise of New Startups
With investors eagerly chasing new startups, more companies like StepFun, Moonshot AI, and Baichuan are expected to launch IPOs soon. However, the swift rise in the stock prices of unprofitable newcomers is raising eyebrows among some investors.
Franklin Templeton’s Nicholas Chui expressed concerns about the rapid valuation increases. “The valuation expansion has been quick, likely outpacing actual earnings growth,” he explained, emphasizing the importance of model introductions going forward.
The market is experiencing stiff competition, with larger companies not remaining idle. Tencent, known for its WeChat app, has potential advantages in the AI landscape, though its recent efforts have not been particularly impressive.
Cost effectiveness remains a crucial consideration for users. The popularity of tools like OpenClaw, which automates numerous daily tasks, is driving a fierce demand for low-cost tokens from niche models. The industry trend of “token maximization,” where engineers maximize token usage, has also gained traction.
While the U.S. retains a technology lead, China’s cost-effective AI applications are capturing a substantial global market share. Reports indicate that China’s daily token consumption hit over 140 trillion in March, a significant increase from 100 billion at the start of 2024.
Mio from Janus Henderson suggests that investors should assess China’s AI landscape by focusing on utilization rates and monetization efficiency rather than solely on the quality of top models. “China is rapidly integrating AI services into real-world applications at low costs,” she remarked.
