China-US AI Technology Competition: Who’s Winning in Key Inputs?

After China revealed its own homegrown large-language model, DeepSeek, in January 2025, the artificial intelligence (AI) competition intensified. Much of the conversation on this new technology has focused on semiconductors or model speeds, but the race is very much dependent on several upstream factors: energy, rare earth elements, and talent. These critical inputs into the AI industry face vastly different structures in the two countries and may determine the pace and scale of AI innovation.

Energy: Powering the AI Revolution

AI models use massive amounts of energy to power their computations. Ensuring a consistent and growing energy supply to power data centers and cool servers is now an essential part of national AI strategies. The United States and China have different energy ecosystems, with alternative methods of pricing, regulating, and sustaining energy for AI endeavors.

United States Energy Landscape

The U.S. is home to some of the most advanced AI data hubs in the world. For example, Data Center Alley in Northern Virginia sees 70 percent of the world’s digital traffic pass through its server farms. These facilities require so much power that coal plants in West Virginia, originally slated for retirement under clean energy guidelines, remain operational to serve this continuous demand.

Other data centers use cleaner energy. Data centers in Oregon and Washington run on hydroelectricity produced by the Columbia River. In addition, recognizing the need for sustainable power supplies, companies like Amazon are investing in small modular nuclear reactors to fuel their data centers. Amazon’s aim is to bring 5GW of new energy projects online by 2039.

Louisiana is receiving a large investment from Meta, which is planning a $10 billion hyperscale AI data campus in Richland Parish. This massive project will require a $6 billion investment in new electric infrastructure, including three gas turbines and substantial solar arrays to ensure operational stability.

Due to increasing investment in electricity generation, the energy requirements of AI data centers have placed a great amount of pressure on local power grids. A notable incident occurred in July 2024, when 60 data centers in Northern Virginia were disconnected from the grid due to a surge protector failure. This incident forced operators to rapidly reduce power generation to prevent widespread outages, and demonstrated the challenges that utilities face in providing data centers with growing amounts of power.

In addition to energy supply, electricity prices and supplier choice are factors that AI producers consider. Electricity produced from U.S. energy sources faces prices that are market driven, with high regional variation due to local energy resources, infrastructure, and regulation. Deregulated markets in some states allow consumers to choose suppliers. In order to maintain a stable supply of data output, data centers often negotiate long-term contracts for electricity.

AI data centers are increasingly located in deregulated states such as Texas, where wholesale electricity pricing allows for competitive contracts that are directly negotiated with generators. Data centers located in high-cost areas such as California or Hawai‘i face significantly higher operational expenses. Different electricity pricing costs result in variable operational expenses for AI workloads.

AI computing power demands are also reshaping grid operations and investment signals in deregulated wholesale markets managed by Regional Transmission Organizations. These markets determine which power plants are dispatched based on lowest cost, often favoring renewables with near-zero marginal costs.

However, AI data centers add new baseload and peak demand, placing additional strain on transmission capacity and grid flexibility. In regions without capacity markets like Texas, prices spike during demand surges, which can create high costs or outages.

China’s Energy Landscape

In contrast to the United States, China’s energy industry is centralized and planned. The government guides the location of data hubs to optimize national grid loads. The Eastern Data, Western Computing plan, for example, pairs data centers with renewable energy in underdeveloped western regions such as Guizhou, which has abundant hydropower. In Sichuan, the Tianfu Intelligent Computing Center is powered by surplus hydro and wind power in a tightly integrated system that supports various sectors.

China’s state planning mechanism allows the government to align power grid investments with AI industry development. Clear targets are set in Five-Year Plans. Grid development, AI infrastructure, and renewable energy are planned together, allowing for high scalability and energy resilience that market-based economies like the United States often lack.

Electricity prices in China average $0.08/KwH compared to $0.18/kWh in the U.S. The government provides state-guided tiered pricing for strategic sectors such as AI. Pilot market reforms in regions like Guangdong and Inner Mongolia have introduced time-of-use pricing and two-part tariffs for large users. This encourages users to shift demand to off-peak periods to maintain grid efficiency.

It is important to note that China’s National Development and Reform Commission has laid out energy-efficient rules for advanced chips. This would disqualify Nvidia’s custom-made China chip, H20, from use. Huawei has reportedly doubled the yield rate on its own AI chips as it seeks to catch up to cutting-edge productivity rates. Relatedly, China’s DeepSeek AI uses far less energy than existing Western models, which will help reduce China’s energy demand on the other side of the ledger.

Rare Earth Elements

Rare earth elements are used in semiconductors, electric motors, and data center cooling systems – all key elements in the AI sector. Rare earth elements are difficult to extract and refine, and have become a geopolitically sensitive industry, since the United States and China have vastly different available resources and capacity in the sector.

Rare Earths in the U.S.

The United States has some rare earth deposits, in particular at Mountain Pass, California, but insufficient processing facilities. Consequently, most rare earth element refining is carried out in China. The U.S. CHIPS and Science Act of 2022 provided funding for domestic rare earth processing, but such industries will take years to build up.

The U.S. remains vulnerable to supply chain shocks as China restricts exports of certain rare earth elements and technologies as a result of the China-U.S. trade conflict. China placed export controls on gallium and germanium – critical materials for chips – following U.S. sanctions on Chinese AI chipmakers, and later restricted exports of tungsten, tellurium, bismuth, indium, and molybdenum. Additional restrictions were placed in April 2025 on exports to the U.S. of medium and heavy rare earth metals like samarium, gadolinium, terbium, dysprosium, lutetium, scandium, and yttrium in response to tariffs imposed by the U.S. on China. Soon after, China began to require its rare earth firms to secure licenses to export materials internationally.

The drive to secure rare earth elements has led the Trump administration to seek resources in Ukraine and Greenland. Trump has stated that his proposed minerals deal with Ukraine could provide the United States with $500 billion in rare earths, although experts warn that such deposits may not actually exist. Trump’s drive to acquire Greenland for its access to trade routes and rare minerals has garnered criticism from Greenland and Denmark. Greenland’s prime minister recently called out a visit by U.S. officials as “highly aggressive.”

Even if these deposits are secured for U.S. exploitation, it wouldn’t solve the problem of China’s near-monopoly on the processing of rare earth minerals.

China’s Rare Earth Dominance

China dominates the rare earth industry, controlling over 70 percent of production and over 85 percent of refining capacity. The nation also holds 40 percent of global reserves, with a supply chain protected by quotas and export controls. And China is clearly willing to make the most of its dominance: as noted above, Beijing banned the export of gallium, germanium, antimony, and superhard materials to the U.S. in December 2024.

China’s rare earth dominance spans multiple aspects of the industry, including mining, processing, and R&D. Since the 1990s, China has treated rare earths as a strategic asset, banning foreign ownership of rare earth mines. China’s June 2024 “Rare Earth Management Regulation” declared that rare earth resources are state-owned. This means that rare earths would be more strictly managed under greater standardization and traceability. The industry is controlled by four major firms, which benefit from state subsidies.

In terms of processing, Chinese engineers have improved the solvent extraction process to refine rare earths. Extraction requires the separation and refinement of rare earth oxides, metals, alloys, and magnets, which is chemical-intensive and requires expertise. Notably, China banned the export of rare earth extraction and separation technologies in December 2023 to maintain its dominance in this area.

However, China’s legacy of severe environmental damage, particularly in southern China, has led to toxic runoff and water and soil pollution. Rare earth mining has produced wastewater that can acidify soil and groundwater. Mining waste can also create radioactive materials as well as heavy metal contamination. As a result, the government has increased environmental policy enforcement for clean production and ecological protection, worrying some experts that the supply of rare earths would be constrained by the new regulation.

Talent: The Human Infrastructure of AI

AI competition relies on people, including researchers, engineers, and domain experts. This is not just a numbers game; it is about creating an ecosystem that nurtures interdisciplinary collaboration, academic freedom, and access to data infrastructure. The United States and China have two different talent environments developing along divergent trajectories.

Talent in the U.S.: Elite but Scarce

A June 2024 analysis by CBRE found that AI’s share of total U.S. tech job postings increased to 14.3 percent from 8.8 percent in 2019. This strong demand has resulted in rising wages for top AI talent. On the supply side, the United States leads the way in top-tier AI research as well as academic excellence. Top AI talent comes from institutions like MIT and Stanford.

However, the U.S. faces labor shortages, since the AI industry is growing faster than the U.S. education system can supply workers. Major industries wish to hire workers in AI but most top-notch engineers are hired immediately by large technology firms such as Google and OpenAI. In addition, American universities do not produce a sufficient number of AI graduates, as the discipline is still growing across higher education. Uneven salary distributions complicate the AI labor market, as top tech companies pay AI engineers an average salary of $120,000 to $250,000, pricing out smaller companies.

Immigration restrictions also limit the talent pipeline. The tightening of H-1B visa regulations has resulted in a dropoff of high-skilled talent into the United States. This is problematic in a country that has fallen behind in STEM fields, so much so that the National Science Board has labeled U.S. educational progress as a “STEM talent crisis.” Lower-tier jobs such as data labeling are often outsourced to firms in the Philippines or Kenya, reducing domestic capacity for large-scale data operations.

The CHIPS Act earmarked $13 billion for workforce development, and the National AI Initiative funds AI fellowships and institutes. Private companies, such as Google AI Residency and OpenAI Scholars, also fund training programs. However, training programs are nascent, and training gaps remain.

China’s AI Talent Pipeline

China produces over 3.5 million STEM graduates per year, over four times that of the United States. The nation educates students in the area of AI across all levels of education. At the primary school level, AI education has focused on Python programming and access to robotics and drone labs. High school students are required to take AI coursework in an information technology course. At the college level, China has over 200 universities offering an AI major. China has also implemented government-led training and subsidized degree programs.

China also has linkages among education, industry, and government through programs like the New Generation Artificial Intelligence Development Plan, which created specialized AI degree tracks and funded “AI Little Giant” incubators for government-certified high-tech SMEs. Large metropolitan cities, such as Beijing and Shenzhen, provide rent-free zones or salary subsidies for AI talent.

Although China previously suffered from brain drain, with top students leaving the country to work abroad, close to 90 percent of Chinese students with AI degrees now remain in China. The number of elite Chinese AI researchers has expanded to 26 percent of NeurIPS presenters in 2022, close to the U.S. share. While the United States still leads the way in attracting top AI talent, China is quickly catching up.

However, it has been pointed out that AI educational curricula often lag behind technology advancements. The study is also hindered due to rigid departmental boundaries, which prevent interdisciplinary collaboration. Faculty members may also lack hands-on experience, which results in insufficient knowledge transfer to students. This indicates that the average quality of Chinese AI education may not be as sophisticated as that in the United States.

Conclusion

The China-U.S. AI race is shaped by far more than just access to chips. Energy, talent, and rare earth materials are all critical inputs. The AI ecosystems and their supply chains in each country differ greatly. The U.S. excels in private sector innovation and frontier research, but struggles with energy fragmentation, higher costs, and talent scarcity. China, on the other hand, has a state-led AI ecosystem with cost-effective infrastructure and rapid scaling. However, its system faces limits in academic flexibility and innovation culture.

Winning the AI race will depend not just on producing faster chips, but on securing and expanding key supply chains that will enable long-term, large-scale AI development. The success of the two countries will hinge on how well they can manage to supply critical and complex inputs that form the backbone of the AI revolution.