I discussed that embodied AI, vertical AI applications across industries, and mass adoption of low-cost AI are the main trends coming out of China in the coming 2 or 3 years. The underlying assumption of my forecast is China will have the capability to lead the AI development despite US attempt at holding back its progress.
In this last part, I’ll discuss why the US will fail in the AI war, like in the trade war and chip war it initiated. Make no mistake – the US declared war on China when it put a chokehold on China’s advanced chip imports in 2022. China imported $413 billion worth of computer chips then, accounting for 15% of its total import. This was China’s single biggest import category, surpassing its $300 billion oil import (and China is the world’s largest oil importer).
Xie Feng, Chinese ambassador to the US, drew the line last week – “trade war, tech war, and whatever other kind of war the US wants to impose on China, we will fight to the end.”
As the US attempts to strangle China’s AI development at the foundational level, China is responding by accelerating indigenous AI tech development and quickly rolling out AI applications across a broad range of industries and economic sectors.
China’s edge in the AI war
As discussed, there are three levels of AI tech stack to form a complete ecosystem – chips, foundational LLMs, and applications.
The US is trying to deny China access to the most advanced chips and LLMs. However, China is making rapid progress to become self-sufficient:
– Chips: Huawei has already rolled out its Ascend series of locally made AI chips which are closing in on Nvidia in performance; Alibaba is developing cutting edge RISC-V chips based on open source technology; Huawei has also made steady progress in developing domestic EUV lithography machine.
A research team at Peking University has developed a bismuth-based 2D transistor that outperforms the most advanced commercial chips from Intel, TSMC, Samsung, and Belgium’s Interuniversity Microelectronics Centre. The new chip is 40% faster than the latest 3-nanometre silicon chips from Intel and TSMC while consuming 10% less energy. This innovation could allow China to bypass the challenges of silicon-based chipmaking entirely. https://interestingengineering.com/innovation/chinas-chip-runs-40-faster-without-silicon
– LLMs: DeepSeek and Qwen have closed the gap with ChatGPT, Llama and other LLMs from the US. The low-cost open-source nature of these Chinese LLMs will help them to reach a broader developer/user base and iterate at a faster rate.
I have already discussed China’s advantage in AI applications with the world’s largest industrial base and consumer market in part 2 of the series. I have also touched on the business model advantages of Chinese AI companies as well as long-term government support and planning.
In this part, I will emphasize two other critical advantages China has: talent and energy.
Talent pool
Needless to say, the most critical factor in long term development in AI is the size and quality of the human capital a country has. In this front, China has an overwhelming advantage.
China graduates annually 3.5 million STEM students with bachelor’s and master’s degrees (virtually 100% of them Chinese) while US graduates 790,000 (including 25% international students). Chinese universities are ranked higher and higher each year and now account for half of the top 50 research universities (not limited to AI) globally according to Nature magazine. See my article “whose universities are better?” https://huabinoliver.substack.com/p/whose-universities-are-better-china
According to ITIF, a major research foundation, Chinese AI researchers published three times as many AI papers as US researchers in the last 5 years. Among the 1% most-cited AI research, Chinese papers outnumber US by a 2:1 ratio. China produces half the world’s top AI researchers and while some still go to work outside of China, 90% who have gone to a graduate school in China are staying in China.
On the other hand, in the US, a recent Paulson Institute report shows there are more graduates of Chinese universities working in top AI labs in the US (38%) than graduates from US schools (36%). If Chinese researchers educated from US schools are included, over half researchers in top American AI labs are ethnic Chinese.
Some Washington lawmakers and industry leaders are calling for outright bans on Chinese nationals from work in the AI field for trumped up national security concerns. Such a move will not only severely cripple AI progress in the US, but also inevitably result in an exodus of top talent from the US back to China, handing China an easy victory.
Energy
A less obvious Chinese advantage in the AI race is energy.
It is now well understood that the rapid technological progress of AI has profound energy sector implications. AI technology is effectively the result of three inputs: chips, data, and electricity.
Computing power is not just about chip performance but the energy to power them. The nearly inexhaustible demand for model training and inference (or AI applications) requires vast amount of electricity to power data centers. This is why companies such as Microsoft and Amazon are exploring building nuclear power plants of their own to meet their ever-increasing electricity demand.
The U.S. power sector has essentially zero “spare capacity” to meet new data center demands following two decades of near-zero demand growth nationwide as the country deindustrializes. Every new gigawatt of data center demand must be matched by a new gigawatt of capacity generation.
For roughly two decades, top-line US national electricity consumption has stagnated, growing at a compound annual growth rate of 0.7 percent since 2007. The electric power industry as a whole has been decelerating since the 1970s.
Commercial strategy, regulatory norms, and policy debates in the US power industry have been conditioned by this seemingly inexorable flatline trajectory and are now out of date.
As a result, existing plants are decades old and use updated technologies. Few can be upgraded, and new power plants will take years and billions of dollars to build. This is similar to the general degradation of infrastructure in the US.
In addition to AI, electricity demand is also growing from other electricity-intensive industries like semiconductor fabrication and battery manufacturing. As the US attempts to reindustrialize, demand for energy-intensive industries like mining, minerals processing, chemical production, metallurgy, etc. will explode. AI needs to compete for resources with all these other demands.
Analysis from Rand, McKinsey and Goldman Sachs project US electricity demand for AI data centers will grow between 400 to 600% by 2030. Currently planned data center sites alone is projected to drive 2 percent annual growth in total US electricity demand. By 2030, total data center electricity consumption will exceed the state of California today (around 240 Terawatt Hours).
In comparison, Chinese electricity demand and supply have been growing in line with economic and industrial growth rates (5-10% annually) for the past three decades. China overtook the US in electricity production in 2011. China today produces more than twice as much electricity (8,392 TWh) as the US (4,065 TWh). China’s clean energy and renewable energy production is four times that of the US and the gap is widening.
China has deployed the world’s most sophisticated nuclear power plants, smart grids with ultra-high voltage transmission systems, and electricity storage facilities. It has been on the leading edge of electrification for a decade, especially with the explosive growth of its EV industry in the last few years.
Last year China State Grid, the world’s largest utility, announced a plan to rollout a national smart grid over the next 10 years at an investment of $800 billion.
As power generation and transmission is fully state-owned, China can make long-term strategic investments in its power sector, including electricity supply for AI data centers.
China’s utility rates are set as a public service while utilities in the US are privatized and for-profit. China’s electricity tariff is $0.075/KWh vs. $0.165/KWh in the US.
In summary, as the AI war intensifies, China is strengthening each part of the AI tech stack, overcoming bottlenecks imposed by the US, and pushing for faster innovation and commercialization. In the long run, China’s advantages in human capital and energy infrastructure will give it an extra competitive edge.
