Alexis Bienvenu

Has China lost the AI battle?

Anthropic, an American AI company that produces the Claude model, is preparing for its initial public offering. Estimated market capitalisation: between 1,500 and 2,000 billion dollars according to sources, although the final range has not yet been set.

Moonshot AI, a Chinese AI company behind the Kimi model – one of the country’s most high-performing models – is also, according to the Wall Street Journal, preparing for a possible initial public offering. Its latest known private valuation stands at around $50 billion, according to Bloomberg – thirty to forty times lower.

The gap in valuation is staggering. It is all the more surprising given that, in January 2025, the performance of the Chinese DeepSeek model had sent shivers down investors’ spines, as they feared the US might lose its dominance in the field. Just under two years later, the markets no longer seem to fear an imminent shift away from American dominance. So is the AI race lost for China?

In some respects, the US lead is indeed very clear. Chinese models can currently only compete with US models on certain tasks. In programming, for example, the gap between the two sides of the Pacific is virtually non-existent: according to the LiveBench 2.0 test dated 25 October 2026, DeepSeek V4.1 Flash scored 96 points, almost on a par with the 97.7 points achieved by Anthropic’s Claude Fable 5.1. However, in long-form agentic tasks, the best American systems score around 78 points, compared with 41 for the best Chinese model, namely DeepSeek. This is a considerable gap.

US players have a second major advantage: their models are already integrated into a vast ecosystem of professional software. An agent connected to Microsoft 365, Salesforce or Amazon Web Services, for example, can, if authorised, access the company’s internal data to take action. Chinese providers do not benefit from the same level of integration into Western companies’ systems, as these companies are often reluctant to entrust their sensitive data to them.

Finally, the United States is by far the leader in terms of computing power. According to Epoch AI, in May 2025 the United States accounted for around 75 per cent of the global computing power of publicly identified major AI centres, compared with 15 per cent for China. They also enjoy privileged access to cutting-edge technologies: Nvidia accelerators, TSMC chips, the latest-generation machines from ASML, and advanced memory produced by SK Hynix, Samsung and Micron. China, by contrast, faces restrictions that limit its ability to acquire this high-end hardware.

Nevertheless, China is not irrevocably condemned to a secondary role in AI. Firstly, its models are significantly cheaper. According to LiveBench 2.0, a successful task costs around $0.18 with DeepSeek, compared with several dollars using Anthropic’s most advanced models.

Furthermore, China has a considerable advantage: the scale and speed of development of its electricity system. The country has been investing heavily for years in renewable energy, coal and nuclear power, whilst the United States lags significantly behind in this type of infrastructure. This momentum could support the expansion of AI, provided that China manages to overcome the challenge of transporting large volumes of electricity between its main centres of production and consumption. According to the IEA’s central scenario[1] , the electricity consumption of Chinese data centres could thus grow by around 170 per cent by 2030, compared with around 130 per cent in the United States.

In short, US dominance in AI – currently unchallenged – faces vulnerabilities precisely where China is concentrating its main strengths. In many respects, the two countries appear to be mirror images of one another in terms of their strengths and challenges.

The contest between the two powers is therefore far from over. Judging by what China has achieved over the past twenty years in the industrial sector, it is entirely possible that a more economical and less flashy strategy – yet one supported as a vital priority by a state-led planning system – may prove competitive in the long term.

Draft completed on 30 October 2026

Alexis Bienvenu, Fund Manager, La Financière de l’Échiquier (LFDE)

Disclaimers: These data and opinions, as well as the sectors and companies mentioned, are provided for information purposes only and, as such, do not constitute an offer to buy or sell any security, nor do they constitute investment advice or financial analysis. The opinions are those of the author; LFDE shall not be held liable for them under any circumstances. Past performance is not indicative of future results.

[1] International Energy Agency (IEA) ‘Energy and AI’ Report, April 2025