AI Enters the Danger Zone as the US-China Race Moves Toward Confrontation

Artificial intelligence is no longer simply a promising technology sector. It has become a central pillar of economic, military and sovereign power for the world’s major powers.

As US President Donald Trump and Chinese President Xi Jinping prepare to hold talks in Washington later this month on AI governance, a darker picture is emerging of a technological race in which semiconductors, electricity and high-performance computing are increasingly intertwined with intelligence operations, strategic deterrence and military calculations.

From technological supremacy to a logic of containment

A report by the Center for a New American Security (CNAS), cited by China’s South China Morning Post, reveals that some US national security circles have begun discussing extraordinary scenarios aimed at preventing China from reaching artificial general intelligence, or AGI, first.

Artificial general intelligence refers to a hypothetical system capable of achieving human-level performance across a broad range of tasks.

Jacob Stokes, deputy director of the Indo-Pacific Security Program at CNAS, has called on US institutions, including the Department of War and the National Security Agency, to assess what intelligence would be required to justify a range of options extending from diplomatic measures to espionage and cyber operations, and ultimately to military action.

According to Stokes’ argument, policymakers should begin by examining the unique characteristics of the technology, in a process reminiscent of the way strategic thinking evolved during the nuclear age, when governments moved from understanding the underlying science to assessing its political and security consequences, according to the newspaper.

What makes the report particularly striking is that it does not focus solely on responses after a technological breakthrough has occurred. Instead, it raises the possibility of considering preemptive strikes against data centers that could potentially train or operate a Chinese AGI system.

The report also discusses the possibility of reverse-engineering Chinese technology or obtaining it through intelligence operations. From Stokes’ perspective, such actions could potentially be justified on legal and normative grounds against the backdrop of previous US accusations that Beijing has engaged in intellectual-property theft.

The idea of striking Chinese data centers based on their potential future use, however, carries an extraordinary risk. Such an attack could trigger a direct conflict between two nuclear-armed powers, with potentially devastating consequences for all sides.

William Hartung, an arms analyst with the Washington-based Government Oversight Project, has warned of precisely this danger. According to the South China Morning Post, he cautioned that attacking Chinese data centers because of how they might be used in the future could ignite a direct war between two nuclear powers.

Data centers become strategic assets

This line of thinking reflects a fundamental shift in the way data centers are viewed. They are no longer necessarily regarded as purely civilian infrastructure or facilities designed primarily to support commercial applications. Increasingly, they are being treated as strategic assets capable of influencing the balance of power.

The ability to train and operate advanced AI models depends on an enormous ecosystem of semiconductors, electricity, cooling systems and high-speed networks. As a result, this infrastructure could, under an escalation scenario, become a potential military target.

The danger lies in the changing definition of critical infrastructure. The more artificial intelligence is considered a decisive national-security asset, the more vulnerable the digital infrastructure and energy systems supporting it become to strategic calculations involving deterrence, disruption and preemptive action.

Estimates from Epoch AI, cited in the report, suggest that the United States currently maintains a lead of several months in advanced AI models. Yet Stokes warns that China could potentially achieve a sudden breakthrough.

His concern is partly based on China’s progress in robotics and embodied AI, a field involving systems that connect perception and algorithms with actions in the physical world.

China’s growing presence in humanoid robotics has become one visible expression of this broader technological push. Beijing has invested heavily in artificial intelligence and robotics, seeking to strengthen its position across a sector that increasingly connects software, industrial automation, autonomous systems and physical machines.

Electricity becomes the hidden battlefield

The Chinese newspaper Global Times, meanwhile, shifts the discussion away from models and algorithms toward a less glamorous but potentially more decisive factor: electricity.

As AI models grow larger, access to cheap, stable and transferable electricity at massive scale becomes increasingly important. Energy availability can influence training speed, inference costs and the ability to deploy AI services across entire markets.

The newspaper points to the Sogou 8000 facility in Zhengzhou, Henan Province, describing it as China’s first fully domestic AI supercomputing complex. The facility reportedly contains 100,000 computing cards and represents an important node in China’s national high-performance computing network.

The facility reportedly occupies only around 2,000 square meters and uses phase-change liquid immersion cooling. According to Global Times, the system improves energy efficiency by directing most of the electricity consumed toward computing rather than allowing a large share of it to be lost through conventional heat-management processes.

The newspaper says the facility’s computing capacity could cover between 5% and 10% of domestic Chinese token demand if it were dedicated entirely to inference.

Global Times links this efficiency to China’s unified electricity grid, its ultra-high-voltage transmission networks and the expansion of renewable-energy infrastructure.

The newspaper cites Zhang Xiaorong, director of the Institute of Advanced Technology Research in Beijing, who argues that this infrastructure offers a comprehensive response to the rapidly growing electricity demands of the AI era. The system allows electricity generated in areas where production costs are lower to be transmitted over long distances to major load centers and computing hubs.

The grid gap and the cost of AI

According to Global Times, China’s computing centers consumed approximately 170 billion kilowatt-hours of electricity in 2025, equivalent to around 1.6% of total national consumption. That figure could approach 800 billion kilowatt-hours by 2030.

In the United States, data centers consumed approximately 192 billion kilowatt-hours in 2024, representing about 4.7% of national electricity consumption. Under a reference scenario, US data-center consumption could reach 649 billion kilowatt-hours by 2030, equivalent to 11.8% of the country’s electricity use.

The newspaper argues that the US challenge is not simply the amount of electricity available annually, but the difficulty of delivering sufficient power, with the required reliability and quality, directly to data centers.

The US electricity grid is divided into three major, relatively independent interconnections and includes hundreds of operators and utility companies. Wood Mackenzie estimates that US grids could satisfy only around 28% of registered data-center interconnection requests, which collectively exceed 1,000 gigawatts. In addition, approximately 70% of US transmission lines have been operating for more than 25 years, according to a report by the American Society of Civil Engineers.

In the Chinese narrative, these weaknesses in the US power grid become a potential competitive advantage for Beijing.

Electricity accounts for more than 60% of data-center operating costs, according to industry experts cited by the newspaper. Xing Honglie, senior director of renewable-energy consulting at UL Solutions, says lower electricity costs are one reason Chinese AI models can offer prices several times lower per million tokens than leading US models with comparable measured quality.

The emerging confrontation in the AI race, therefore, is not simply about which country reaches a more powerful model first. It is increasingly about which country has the infrastructure required to operate, scale and protect that technology.

The United States continues to hold advantages in cutting-edge AI training chips, software ecosystems, the performance ceiling of advanced proprietary models, capital intensity and highly skilled talent. China, meanwhile, is betting on its unified national electricity grid, clean-energy expansion, domestic deployment and lower inference costs.

The contrast reveals a crucial reality of the AI competition: technological leadership depends not only on algorithms, but also on the physical infrastructure required to run them.

A dangerous new phase in the AI race

Trump and Xi’s upcoming talks in Washington are expected to open a channel for discussions on more responsible AI governance. Yet the growing presence of scenarios involving espionage, cyber operations and even preventive military strikes illustrates how quickly the strategic debate is evolving.

The central danger is that AI infrastructure could move from being a bridge for technological cooperation and economic competition to becoming a direct point of confrontation between two nuclear powers.

Data centers, semiconductor supply chains, electricity networks and computing capacity are increasingly becoming part of the geopolitical equation. As governments view AI as an essential component of national power, the infrastructure behind it could acquire the same strategic significance.

That creates a difficult paradox. The more important artificial intelligence becomes to national security, the greater the temptation to treat the systems that support it as strategic targets. Without clear rules, crisis-management mechanisms and mutual safeguards, a technological race designed to establish economic and military advantage could create entirely new pathways toward escalation.

The US-China AI competition is therefore entering a potentially dangerous phase. The decisive contest may no longer be limited to who develops the most advanced model. It may ultimately depend on who controls the chips, electricity, data centers, networks and industrial capacity needed to turn artificial intelligence into real-world power.