“It was an expression of growing self-confidence. The Chinese AI founder wrote laconically that it "wouldn't take that long"—thereby challenging the world's richest person. Elon Musk had written on his platform, X, that it would take until the first quarter of next year for Chinese teams to develop an artificial intelligence as powerful as Anthropic’s celebrated and feared model. Tang Jie, however, predicted that such a model would emerge from China this year. For the West, this would be devastating news. After all, it would mean that all the security risks associated with Anthropic’s models would now originate from China.
Tang has every reason to be confident. He is an AI professor holding a chair at the elite Tsinghua University and is one of the field's most respected researchers. Recently, his company, Zhipu AI, unveiled a new model called GLM-5.2; it ranks just behind Anthropic and OpenAI in standard benchmarks and is proving highly popular among developers. The company’s stock has skyrocketed; the share price has increased sixteenfold since the IPO earlier this year, making it arguably the most successful AI stock in the world. Forbes currently estimates Tang’s net worth at around six billion dollars. The professor is now a billionaire.
While clear frontrunners like OpenAI and Anthropic have emerged in Silicon Valley, the AI race in China remains comparatively open. Every few months, a new team makes headlines—a testament to the market's intense competition and reminiscent of other Chinese sectors, such as electric vehicles. First came DeepSeek, which put Chinese AI on the map. Then Moonshot AI rose to prominence, followed by Alibaba with Qwen; now, Zhipu AI is considered leading the way. Many other companies are in the race, such as the likewise publicly traded Minimax or ByteDance, whose AI app Doubao ranks among the most popular in China. "Ask Doubao" has already become something of a household phrase in the Middle Kingdom.
However, experts from the consulting firm BCG state in an analysis published this week that the US remains in the lead. Yet China is catching up rapidly and has built an increasingly independent AI ecosystem.
The authors examined six dimensions in this regard: capital, talent, intellectual property, data, energy, and computing power.
Accordingly, China is focusing increasingly on basic research, publications, and patents, as well as on domestic chip development and the application of artificial intelligence in the real economy.
In 2025, one-third of the most cited AI research papers originated in China—twelve percentage points more than from the United States.
China has also long held the lead in the number of AI patents.
The United States, meanwhile, continues to benefit from its pool of AI talent and, above all, from the massive capital it deploys to finance the best AI models and the necessary computing power.
Since 2023, American AI start-ups have raised a good $380 billion from venture capitalists—nine times more than their Chinese counterparts.
The 20 largest American tech companies—including Google, Amazon, and Meta—invested $439 billion last year, with a significant portion going toward artificial intelligence and data centers. Investments are expected to rise to more than $800 billion this year. By comparison, the $63 billion invested by the 20 largest Chinese technology companies seems almost modest. They invest 9 percent of their revenue, whereas their US counterparts invest 16 percent.
However, the Chinese state plans to invest around $295 billion in nationwide AI data centers over the next five years.
Yet China holds a decisive advantage in the data center race: while the US is already grappling with power grid bottlenecks in many areas, China is expanding its grid at record speed.
Some experts even consider overcapacity a possibility by 2030.
China already generates more than twice as much electricity as the US.
Part of the strategy involves building data centers in China’s less developed western regions, where there is already often a surplus of solar, wind, and hydroelectric energy. According to the Chinese government, 70 percent of all new data centers are currently being built there.
China remains competitive in the field of foundation models, according to BCG experts. The openness of the ecosystem has played a role here: unlike Anthropic, Google, or OpenAI, Chinese providers publish their architectures allowing other developers to build upon them.
However, the primary focus for Chinese players has been to offer AI at significantly lower costs while maintaining comparable quality. The latest version of the "Kimi" model from the start-up Moonshot AI costs approximately $1.71 per million tokens—the text units that AI models process and generate. By comparison, OpenAI’s latest model costs users $11.25 per million tokens. Consequently, many companies worldwide are opting to use Chinese AI models due to cost considerations.
At the same time, Beijing is driving the domestic adoption of its own models, thereby boosting demand. China is projected to account for more than half of global AI usage in the second quarter of this year—despite representing 17 percent of the world's population.
A more positive attitude toward AI—compared to the US—may play a role here: according to BCG data, 86 percent of Chinese people believe AI offers more benefits than drawbacks, whereas in the United States, that figure stands at just 41 percent.
In the People's Republic, this is reflected in the aggressive push by tech companies to bring AI agents to market. These agents can independently order food or book flights for users. Tencent is currently driving integration into WeChat—an app ubiquitous in China used for chatting, payments, travel bookings, and—via mini-apps—countless other daily functions. Alibaba has similar plans for its rival app, Alipay.
For its AI initiatives, China is increasingly relying on a supply chain independent of the US, particularly for model training. Although the US government is once again permitting more exports of advanced Nvidia chips to China, China’s largest Nvidia customer—TikTok parent company ByteDance—is barred from using the US-developed semiconductors. State-backed data centers are required to use domestic chips, with hopes pinned on advancements in Huawei’s chip technology.
The latest model from the Chinese AI company DeepSeek, for instance, was optimized to run on Huawei chips, even though it was trained—at least in part—on Nvidia chips.
"Lower-cost Chinese models running on cheaper Chinese chips could offer an attractive package for many countries," notes the BCG analysis—particularly for nations in the Global South that are favorably disposed toward China. However, interest could also emerge in Europe, given the skyrocketing costs of deploying American AI models and the uncertainty surrounding potential export controls on the most advanced models. By the end of 2025, open-source Chinese AI models had already surpassed American models in terms of download numbers.
At the same time, the growing incompatibility between Chinese and American AI means fewer options for Europe to mix technologies from both regions. Apple, for example, already partners with Alibaba for its iPhone AI in China, while working with OpenAI in the rest of the world. That said, BCG’s metrics rank Europe as the most advanced among "middle powers" in the field of artificial intelligence—though it still lags far behind the US and China.
French frontrunner Mistral, for instance, remains three to ten months behind top AI labs technically—an eternity in the AI era.” [1]
1. China jagt die USA im KI-Rennen: Der Herausforderer holt auf und entwickelt zunehmend Technik, die für amerikanische Anbieter inkompatibel ist. Das hat auch für Unternehmen aus Europa weitreichende Folgen. Frankfurter Allgemeine Zeitung; Frankfurt. 03 July 2026: 25. Von Maximilian Sachse, Frankfurt, und Gustav Theile, Shanghai
Komentarų nėra:
Rašyti komentarą