The U.S. AI data center boom reveals a critical dependence on Chinese components

The multibillion-dollar boom in artificial intelligence data centers in the United States is exposing a strategic vulnerability that until recently went unnoticed: much of the hardware that keeps those facilities running —not the chips, but the equipment that supplies their electricity— depends heavily on Chinese manufacturers. According to analysts consulted by CNBC, Chinese companies supply significant shares of the transformers, switchgear, batteries and optical technology used in US data center infrastructure.

Until now, the debate over the technological rivalry between Washington and Beijing had focused almost exclusively on the 'compute stack': the advanced semiconductors and AI hardware in which the United States holds a clear advantage. However, according to Laveena Iyer, senior analyst at The Economist Group, scrutiny of the Chinese presence in data centers' 'power stack' has increased markedly over the past eight months, marking a shift in focus from the earlier concern centered solely on compute.

A data center's 'power stack' is the set of equipment that ensures the facility operates with minimal interruptions. It includes transformers, which convert high-voltage electricity from the grid into the lower levels needed by servers and cooling systems; switchgear, which bundles switches, fuses and circuit breakers; and batteries, needed for backup power. According to Ben Boucher, senior supply chain analyst at Wood Mackenzie, one of the most notable areas of dependence is substation transformers, which hyperscalers typically install on site to step down transmission voltage.

The figures provided by Yury Dvorkin, associate professor at Johns Hopkins University, give a concrete sense of the scale of the problem: China's share in certain categories of transformers and switchgear is around 30%, and in batteries it accounts for more than 40% of US imports. Dvorkin adds that the exposure is not limited to finished components but extends 'upstream' to materials such as copper, electrical steel and battery cathode materials, which makes that supply chain harder to replace quickly.

In optical technology —which uses light through fiber cables rather than electrical signals over copper to transmit large volumes of data— Chinese dominance is even more pronounced. Companies such as Zhongji Innolight and Eoptolink lead global revenues for data center optical transceivers, and Chinese firms together account for roughly two-thirds of the world's unit supply, according to research firm Counterpoint.

The political response in Washington has intensified. Last week, President Donald Trump signed an executive order declaring a national emergency over what he called an 'extraordinary threat' to the United States from foreign-made bulk-power system equipment. The order authorizes the Department of Energy to ban or impose conditions on certain transactions involving components used in the power grid and in data centers. In a statement accompanying the order, Trump said that 'the threat to the United States from the foreign supply of bulk-power system electric equipment has become even more acute' since his first term, and that 'the rapid growth of advanced manufacturing, data centers, artificial intelligence and defense production has increased the nation's reliance on abundant and reliable electricity'. Back in April, in a presidential determination, Trump had designated transformers, transmission lines, conductors, substations, high-voltage circuit breakers and power control electronics as 'essential to the national defense'.

That regulatory pressure has also translated into concrete measures: in July, the Federal Communications Commission (FCC) added foreign-made power inverters —which help connect power sources for AI facilities— to its 'Covered List', the roster of equipment and services the US government considers an unacceptable national security risk. In addition, the Trump administration is reportedly drafting a ban on US imports of new Chinese optical transceivers. Neither the FCC nor the White House responded directly to CNBC's questions about that last measure.

On the industry side, there are already moves to reduce dependence. Hitachi Energy announced in September 2025 a $1 billion investment to expand production of critical grid infrastructure in the United States, including $457 million for a new large power transformer plant intended to serve demand from the AI boom. Siemens Energy, for its part, announced in February a $1 billion investment in US production of grid equipment and gas turbines for AI infrastructure and data center expansion.

In the optical field, any restrictions on Chinese companies could benefit US firms such as Lumentum and Coherent, both of which received a $2 billion investment each from Nvidia in March. Analysts warn, however, that scaling up manufacturing to cover the new demand will not be easy. Neil Shah, vice president of research at Counterpoint, noted in an August report that Western competitors such as Coherent and Lumentum have advanced photonic designs but currently lack the cleanroom capacity, automated packaging infrastructure and throughput scale needed to absorb Innolight's and Eoptolink's volume within a 12- to 24-month horizon.

The risk that restrictions will create shortages and drive up prices is already showing up in concrete figures: according to Wood Mackenzie research, power transformers and substations already face an estimated market shortfall of 15% and 8% respectively in 2026, and restrictions on the use of Chinese-made units will only aggravate these supply chain problems. Wood Mackenzie's Boucher explained in an August blog post that much of the impact will fall on data centers that have been using Chinese units precisely to minimize lead times, and that the above-100 MVA segment is where the shortage is already most acute — exactly the segment in which data centers have turned to Chinese manufacturers to manage those lead times.

Despite these cost and supply risks, the political will in Washington to reduce dependence on China in the power stack continues to grow, according to Iyer. The analyst draws a telling parallel with the 5G rollout: the US administration appears determined to 'de-risk' its AI infrastructure before Chinese technology becomes fully embedded in it, much as happened with 5G, where the subsequent removal of Chinese telecom equipment caused delays in network rollouts and higher costs for telecom companies.

From Beijing, a spokesperson for the Chinese embassy in the United Kingdom told CNBC that 'we oppose overstretching the concept of national security to target foreign companies' and expressed the hope that 'relevant countries will provide a fair, equitable and non-discriminatory business environment for Chinese companies'. A White House spokesperson, for his part, defended the administration's agenda, saying that 'reshoring manufacturing critical to our national and economic security has been a top priority for President Trump' and that the strategy of tax cuts, tariffs and deregulation 'is working', citing trillions of dollars in investment in key sectors ranging from steel to semiconductors and the auto industry.

The backdrop to this entire debate is the scale of the growth expected: according to S&P Global, US data center capacity will go from 62 GW in March 2026 to 152 GW in 2030, driven by high-density AI workloads. That accelerated expansion is precisely what, according to the analysts cited, has led operators to lean on Chinese suppliers able to deliver equipment quickly, and what now makes any attempt to disconnect from that supply chain a delicate exercise: reducing geopolitical risk without slowing down, or excessively driving up the cost of, the very race to build the AI infrastructure that Washington considers strategic.