Appeared in Substack on 10 December 2025:
The other day at dinner with a recent Rhodes Scholar who got his degree from Brown University, we had a very interesting conversation about the state of artificial intelligence, especially its recent impact and influence. When the topic turned to AI developments in China, this young man simply threw up his arms and said something like, “well Chinese industry does what the Chinese Communist Party (CCP) wants” quite disparaging and unimpressed by what was going on in China. It is uncanny how monolithic the reaction of most Americans is vis-à-vis China, even the best and the brightest. Being unenlightened about the world’s largest economy is going to come back and bite America’s domestic growth. In fact, it is my contention that the U.S. can learn a lot about how to compete successfully in AI by emulating what China is doing.
First, unlike the widespread Pavlovian reaction to China, the CCP is neither telling entrepreneurs exactly what AI to build and what not to, nor are they hiding behind bushes monitoring every move made in the industry. In fact, what the Chinese government is doing is to support, rather than direct, the AI industry in multiple ways. The government lets the Chinese entrepreneurs figure out what to do with AI, what to build and what industries to target. This is how the Chinese industrial giants Alibaba, Tencent, BYD and JD, among others, became very successful. All private companies, and all have a light touch from the Chinese government. Of course, the largest companies in China are all state-owned; these include State Grid Corp of China, China National Petroleum, Sinopec, China Construction Bank, Sinochem etc.
The Chinese government certainly has an Industrial Policy aimed at enhancing the development of new technologies and processes including the development of semiconductors, semiconductor manufacturing, artificial intelligence and computing. In each case, the government provides support in the form of infrastructure development and expects the industry and companies to meet certain targets and performance metrics. Companies that perform are given added bonuses and subsidies. And there is little attempt to curb the financial rewards of successful entrepreneurs who make billions of dollars in personal wealth. In the case of AI, the industrial policy has set the target that there should be 90% penetration by 2030, very aggressive indeed! Note that a recent McKinsey study found that there was only 1% penetration of AI in U.S. companies at the end of 2024.
Using adoption rate as a measure of success rather than the stock market valuation of the companies building AI models and algorithms should lead to a better outcome and faster expansion of AI. Companies producing AI models will focus on adoption rather than profits and this in turn should lead to improvements in the models and algorithms as well. It should be noted that the Chinese government strongly encourages AI companies to develop open-source AI models, and this makes adoption much easier and cheaper. China wants AI to benefit industrial development and the standard-of-living of its citizens rather than just be a technology to profit companies and an elite few.
The Chinese government supports the AI industry by investing in the rapid expansion of infrastructure needed for AI, both directly and indirectly. China is aggressively building data centers to support AI development. The main resource that fuels AI is vast quantities of data that are used to train and improve models. Storing, managing and manipulating data requires data centers with ultra-fast servers/computers. In 2024, China had a capex (capital expenditure) of $90 billion about three times that of the U.S. In addition, it is 60% cheaper to build data centers in China as compared to the U.S. not including the purchase of GPUs (Graphics Processing Units) that are the main semiconductors propelling AI applications.
AI data centers are really hungry for electricity. If recent figures are correct, China is expanding its power grid (i.e. electricity capacity) about 3-5 times as fast as the U.S. each year. Given its commitment to renewable energy, it is much easier to expand the electricity capacity in China than in the U.S. and much cheaper. According to IEA (International Energy Agency) when we add up all the phases, constructing a 1 MW solar power plant typically takes about 1-2 years from the start of planning to the final commissioning. A thermal power plant takes 5-6 years. Renewable energy has achieved cost parity as well in the U.S. and would be cheaper in China. Utility-scale solar and onshore wind cost $23 to 139/MWh compared to coal costing $68-166/MWh and natural gas $77-130/MWh. In fact, new electricity to power data centers for AI costs about 40% less in China.
The infrastructure investment by the Chinese government makes it easier for Chinese entrepreneurs to rapidly expand AI. However, China has an important disadvantage: It is banned from being able to get the state-of-art GPU semiconductors from the U.S. This barrier did not prevent a very versatile LLM model, DeepSeek, being released by the Chinese company High-Flyer in November 2023. The company launched an eponymous chatbot alongside its DeepSeek-R1 model in January 2025. Its training cost was reported to be significantly lower than other LLMs. The company claims that it trained its version 3 of the DeepSeek model for US$6 million—far less than the US$100 million cost for OpenAI’s GPT-4 in 2023 and using approximately one-tenth the computing power consumed by Meta’s comparable model, Llama 3.1.
The company trained its models during ongoing trade restrictions on AI chip exports to China, using weaker AI chips intended for export and employing fewer units overall. DeepSeek significantly reduced training expenses by incorporating techniques such as mixture of experts (MoE) layers. The company recruits AI researchers from top Chinese universities and also hires from outside traditional computer science fields to broaden its models’ knowledge and capabilities. What the Chinese AI companies lack in the best semiconductors is more than made up by having brilliant engineers who come up with ways of optimizing the systems under these constraints leading to more efficient solutions. China graduates more than 1.38 million engineers each year, about seven times more than does the U.S.
While DeepSeek captured the public imagination like ChatGPT because of the ease of use on the underlying LLM models, several Chinese companies have also aggressively entered the AI race, including Huawei, the technology giant and Alibaba, the eCommerce giant. Qwen (Tangyi Qianwen = universal questions), Alibaba’s AI model recently won an international competition in which AI models were asked to trade crypto currencies and stocks; DeepSeek came second. Huawei has several AI models in its Pangu series and each new version is much more powerful than the previous one. Pangu is a primordial creation figure in Chinese mythology and in Taoism. According to legend, Pangu separated heaven and earth, and his body later became geographic features such as mountains and flowing water. In other words, Huawei’s vision is for Pangu to define the global AI landscape on which everyone can live and thrive! Quite an impertinent goal!
DeepSeek, Qwen and Pangu are all open source, cost-effective, and high-performing AI models. It is very easy for companies that want to use the AI models from Chinese companies to simply upload them and use them without having to connect to them via an API (application programming interface). This enables much faster and deeper penetration of AI into the Chinese economy. Companies that use either DeepSeek or Qwen or any other AI model that could be launched in the future could take advantage of the technology without having to worry about matters of data privacy or losing control of cybersecurity.
China’s industrial policy has resulted in tens of billions being mobilized to support developments in the domestic semiconductor industry. The sanctions and tariffs imposed by the U.S. has also spurred these developments to the point where the conservative Economist magazine wrote recently that “China’s chip industry will surprise the world”. Recently there was a big splash when Chinese GPU startup Moore Threads debuted on the STAR Market on December 5, 2025, with its opening price soaring to CNY650 (US$91.94) per share—a staggering 468.78% increase, creating one of the most remarkable public market debuts in recent memory. Called “China’s Nvidia” its Founder-CEO James Zhang worked in the global semiconductor business (including Dell and HP) for almost 20 years and led Nvidia’s China unit for 14 years until 2020 when he founded Moore Threads.
There are many players in chip design in China including Huawei and upstarts such as Enflame Technology and Biren technology in addition to Moore Threads, not to mention Huawei itself. The fastest chip, as of this writing is the Nvidia H100 processor but many companies use H20 a less sophisticated version. According to Dylan Patel, the CEO and Chief Analyst of SemiAnalysis, “the performance gap between Huawei [Ascend chip] and H20 is less than a full generation [i.e less than a year]”. Nvidia is a chip design company, and almost all its manufacturing is done by TSMC (Taiwan Semiconductor Manufacturing Company Inc.), the world’s largest chip foundry which is barred from taking any chip order from a Chinese company. So the Chinese government is aggressively investing in chip foundries to the tune of billions of dollars; the largest one being SMIC (Semiconductor Manufacturing International Corporation). SMIC is still far behind TSMC, but this could also play a role in the geopolitical maneuvering of China – just saying!!
China is only a year or two away from competing with the best U.S. chip designers. Semiconductor manufacturing in China is several years behind the U.S. suppliers like TMSC, and there is a lack of suppliers of chipmaking equipment like the Dutch company ASML (Advanced Semiconductor Materials Lithography). There are several Chinese companies trying to break the stronghold of ASML including SiCarrier Technologies, but lithography could take years to master, if not a decade or so. Of course, China has a very potent resource for overcoming barriers: smart engineers. Almost one-half of all AI engineers in the world have degrees from Chinese universities. The Chinese engineers are hot in pursuit of coming up with innovative alternatives for producing chips, rather than imitations, to the current state-of-art in chip manufacturing such as lithography. In an indirect way, the sanctions aimed at the development of AI in China might actually result in providing a strong incentive for the country to create better ways to develop and manufacture GPUs and to dominate global AI. Only time will tell.
Anand Anandalingam - Personal Web Site 