As the United States and China vie for AI supremacy, reliance on China for key components used to power data centers has come under increasing scrutiny, raising the prospect of rising costs and worsening supply chain shortages for building AI.
U.S. hyperscalers are racing to build multibillion-dollar data centers for AI, and Chinese companies are supplying most of the components needed to develop these facilities. Analysts told CNBC these include transformers, switchgear, batteries and optical technology.
President Donald Trump signed an executive order last week declaring a national emergency, citing “extraordinary external threats” to the United States involving large-scale power system equipment produced overseas. The order gave the Department of Energy the authority to prohibit or impose conditions on certain transactions involving some components used in power grids and data centers.
Tensions between the US and China over AI are escalating as China’s AI models become more sophisticated and their adoption expands around the world. US companies produce cutting-edge chips and hardware used in AI, but China’s influence on the technology that powers data centers is considered a strategic vulnerability for the US.
“Chinese presence in the power stack of U.S. data centers has come under increasing scrutiny over the last eight months or so, which is a shift from the previous focus on just the computing stack,” Raveena Iyer, senior analyst at The Economist Group, told CNBC.
dependencies
Data center power stacks are critical to keeping your facility running with minimal downtime. This includes transformers that convert high-voltage power from the power grid into the lower-level power needed by servers and cooling equipment. Switchgear is a centralized system with switches, fuses, circuit breakers, and requires a battery as a backup power source.
Ben Boucher, senior supply chain analyst at Wood Mackenzie, told CNBC: “We’re seeing an increased reliance on imports from China in several critical areas as we build out our AI infrastructure.” “The most notable ones include substation transformers, which hyperscalers are typically installed on-site to step down the transmission voltage.”
“Medium-term exposures are concentrated in grid-connected equipment such as transformers, switchgear, and batteries,” Yuri Dvorkin, an associate professor at Johns Hopkins University, told CNBC.
“China’s share in certain categories of transformers and switchgear has reached nearly 30 percent, and it accounts for more than 40 percent of U.S. battery imports,” he added. “Our analysis also found exposures deeper upstream, including in copper, electrical steel sheets, and battery cathode materials.”
Optical technology, which uses light to transmit large amounts of data over fiber optic cables instead of electrical signals over copper wires, is another area where China has an advantage.
Companies such as Zhongji Innolight and Eoptolink lead global data center optical transceiver revenue, with Chinese companies together accounting for about two-thirds of global unit supply, according to research firm Counterpoint.
President Trump said in April that “transformers, power lines and conductors, substations, high-voltage circuit breakers, and power control electronics” are “essential to national defense” as part of his executive action.
reduce dependence
The US government is scrambling to reduce its dependence on Chinese suppliers as it rapidly develops the infrastructure needed to power AI.
“Since my first term, the threats to the United States regarding foreign supplies of electrical equipment for large power systems have become more serious,” President Trump said in a statement related to the executive order issued last week.
He added: “Rapid growth in advanced manufacturing, data centers, artificial intelligence, and defense production has increased the country’s dependence on abundant and reliable electricity, magnifying the impact of a successful attack or disruption to large power systems.”
CNBC contacted the British Embassy in the People’s Republic of China for comment on President Trump’s executive order and the country’s involvement in supplying components to build AI data centers.
In June, S&P Global announced that U.S. data center capacity is expected to grow from 62 million kW in March 2026 to 152 million kW by 2030, driven by dense AI workloads.
“Reviving manufacturing, which is critical to our national and economic security, is a top priority for President Trump, and his administration continues to implement a strong and agile policy of tax cuts, tariffs, and deregulation,” a White House spokesperson told CNBC. “Trillions of dollars in investment across key sectors from steel to semiconductors to autos is proof that the administration’s strategy is working.”
Hitachi Energy announced in September 2025 that it would invest $1 billion to expand production of critical grid infrastructure in the U.S., including $457 million for new large-scale substation equipment to meet demand from AI buildouts. siemens energy announced in February that it would invest $1 billion in U.S. production of grid and gas turbine equipment for AI infrastructure and data center expansion.
Foreign-made power inverters that help connect energy sources for AI facilities were added to the Federal Communications Commission’s (FCC) list in July. This list includes devices and services that the U.S. government has determined pose an unacceptable risk to national security.
The Trump administration is also reportedly drafting a proposal to ban U.S. imports of new Chinese-made optical transceivers that can transmit data at high speeds over fiber-optic cables between data centers. The FCC, which oversees the telecommunications industry, did not respond to requests for comment, and the White House did not address the report.
Any restrictions on Chinese optical technology companies could give a boost to U.S. companies like Lumentum and Coherent, both of which have received $2 billion in investments. Nvidia March — Analysts warn that scaling up manufacturing to meet new demand could be fraught with challenges.
“Western competitors like Coherent and Lumentum have advanced photonic designs, but currently lack the cleanroom capacity, automated packaging infrastructure, and yield scale needed to absorb Innolight and Eoptolink volumes within 12 to 24 months,” Neil Shah, vice president of research at Counterpoint, said in an August report.
Fallout from restrictions
The Economist Group’s Iyer said a ban on optical transceivers would force U.S. hyperscalers to find local alternatives, potentially raising costs.
Wood Mackenzie research has already estimated that transformers and substations will face a market shortage of 15% and 8% respectively in 2026, and restrictions on the use of Chinese-made units will further exacerbate supply chain challenges.
“The majority of the impact will be concentrated in data centers that use units in China to minimize lead times,” Wood Mackenzie’s Boucher said in an August blog.
“The 100 MVA+ segment is already where the shortage is most acute and is exactly where data centers are turning to Chinese manufacturers for lead time management,” he added.
However, there is growing political will in Washington to reduce dependence on China in the power structure, Iyer added. “The U.S. government appears to want to de-risk AI infrastructure before Chinese technology is fully integrated, much in the same way that late removal of Chinese telecommunications equipment during 5G rollouts led to network rollout delays and higher costs for carriers.”
