The data centers powering the artificial intelligence race in the United States are hiding a problem that Washington took a while to acknowledge out loud.
Behind the billions invested in AI infrastructure, there is a real and strategic dependence on equipment manufactured in China — and it has finally become a national security concern. 🚨
Transformers, electrical protection systems, batteries, and optical technology are some of the critical components that keep these data centers running, and a large share of them comes from Chinese suppliers. It is no exaggeration to say that without these components, much of the infrastructure running artificial intelligence models today simply would not exist in its current form — or would be far more expensive to build.
The numbers are pretty concrete. According to Yury Dvorkin, associate professor at Johns Hopkins University, China accounts for roughly 30% of certain categories of transformers and switchgear, and more than 40% of the batteries imported by the U.S. are of Chinese origin. On top of that, Chinese companies like Zhongji Innolight and Eoptolink control approximately two-thirds of the global supply of optical transceivers, according to research firm Counterpoint. When you put these numbers side by side, it is hard to ignore the real scale of this exposure — especially at a time when geopolitical tensions between Washington and Beijing remain high.
The situation became even more tense after President Donald Trump signed an executive order declaring a national emergency related to foreign threats to the country’s energy system. The measure authorized the Department of Energy to prohibit or impose conditions on certain transactions involving components used in the power grid and in data centers. With that, something the tech industry already knew but preferred not to put at the center of public debate came to the surface: the supply chain of major American data centers has deep roots in Chinese territory, and this represents a vulnerability that goes far beyond a simple trade issue.
As Laveena Iyer, senior analyst at The Economist Group, put it, scrutiny over the Chinese presence in the power layer of American data centers has only grown over the past eight months — a shift from the previous focus, which concentrated solely on the compute layer. But reducing this dependence is neither simple nor fast, and the consequences of a rushed transition could be felt directly in the costs and timelines of AI expansion over the coming years. 👇
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
What is at stake in the data center supply chain
To understand the depth of the problem, you need to look at what actually keeps a large-scale data center running. It is not just GPU chips or high-performance servers — those are the components that get all the attention. What many people do not realize is that the physical infrastructure behind all of it depends on far more basic equipment.
Transformers, for example, convert high-voltage electricity from the grid to the lower levels required by servers and cooling systems. Switchgear systems are centralized assemblies that include switches, fuses, and circuit breakers, while batteries provide backup power in case of failure. All of these items make up what experts call the power layer, or power stack, and it is precisely in this layer that the dependence on Chinese-made equipment becomes most visible and concerning.
Ben Boucher, senior supply chain analyst at Wood Mackenzie, pointed out that substation transformers are among the most notable points of this exposure, since large operators typically keep them on-site to step down transmission voltage. Dvorkin from Johns Hopkins adds that the medium-term exposure is concentrated right in the grid connectors: transformers, switchgear, and batteries. And it goes further — according to him, there is an even deeper exposure upstream in the chain, involving copper, electrical steel, and cathode materials for batteries.
The global supply chain was built over decades based on cost efficiency, and China positioned itself as a central supplier of heavy industrial components with a combination that is hard to beat: production capacity at scale, cheap labor, and technological dominance in certain product categories. Optical technology, which uses light through fiber cables instead of electrical signals through copper wires to transmit massive amounts of data, is another clear example. Chinese companies did not end up controlling two-thirds of the global market by accident — it was the result of years of investment in manufacturing capacity and price reductions that made Western competition unviable in many segments. The problem is that when this kind of supply concentration meets intense geopolitical tension, the risk stops being theoretical and becomes operational.
And data center operators know this. Major cloud companies invest hundreds of billions of dollars in physical infrastructure, and any disruption in the supply of critical components can delay entire projects by months or even years. If the dependence on Chinese-made equipment is abruptly restricted through tariffs or sanctions, this bottleneck could become even more severe, with a direct impact on the capacity expansion plans that underpin the growth of artificial intelligence.
Washington races to reduce the dependence
The obvious answer to this problem would be to diversify suppliers — and that is already happening, at least in the rhetoric. In a statement tied to the executive order, Trump said that since his first term, the threat related to the foreign supply of electrical equipment for the energy system has become even more acute. He added that the rapid growth of advanced manufacturing, data centers, artificial intelligence, and defense production has increased the country’s dependence on abundant and reliable electricity, amplifying the consequences of a successful attack or supply disruption.
On the other side, China is pushing back against the measures. A spokesperson for the Chinese embassy in the United Kingdom told CNBC that it opposes the excessive broadening of the national security concept to target foreign companies, and stated that it expects the countries involved to provide a fair and non-discriminatory business environment for Chinese companies.
The practical reality of rebuilding or expanding a supply chain for heavy industrial components is far more complicated than switching software vendors or changing cloud providers. Manufacturing high-voltage power transformers requires specialized machinery, specific raw materials, and skilled technical labor that cannot be created from scratch in just a few years. The U.S. has drastically reduced its domestic manufacturing capacity in this sector over the past several decades, and rebuilding it takes time and substantial investment — time that the fast pace of the artificial intelligence race does not typically have the patience to wait for.
Still, investments have started to show up. Hitachi Energy announced in September 2025 a $1 billion investment to expand critical grid infrastructure production in the U.S., including $457 million for a new large-scale transformer factory aimed at AI demand. Siemens Energy also announced in February a $1 billion investment in American production of grid equipment and gas turbines for AI infrastructure and data center expansion.
Geographic diversification of the chain, however, does not necessarily mean eliminating Chinese influence — many times it just means shifting that influence to other countries that, in turn, depend on inputs or technology that still pass through China. The battery case is a perfect example: even if the cells are assembled in another country, the active materials like lithium, cobalt, and refined graphite frequently go through processing on Chinese soil before reaching the assembly line. This makes the problem structurally more resistant than it appears at first glance. 🔍
The real impact on AI expansion plans
When it comes to data center expansion to support artificial intelligence workloads, the numbers are astronomical. According to S&P Global, data center capacity in the U.S. is expected to jump from 62 GW in March 2026 to 152 GW by 2030, driven by high-density AI workloads. All of this new capacity needs power, and power at data center scale means transformers, distribution systems, battery backup, and ultra-high-speed fiber optic connectivity — the very same components whose supply chain is under pressure.
The problem is already tangible. According to Wood Mackenzie research, power transformers and substations face an estimated market shortfall of 15% and 8%, respectively, in 2026. Restrictions on the use of Chinese-manufactured units are likely to make these challenges even worse. Boucher from Wood Mackenzie notes that most of the impact will be concentrated on data centers that had been using Chinese units to minimize lead times, with the segment above 100 MVA being precisely where the shortage is most acute.
Beyond transformers, the U.S. government is also targeting optical technology. Foreign-made power inverters were added in July to the FCC’s so-called Covered List, which includes equipment and services deemed an unacceptable risk to national security. And there are reports that the Trump administration may be drafting a ban on imports of new Chinese optical transceivers — components that enable ultra-high-speed data transmission between data centers.
Any restriction like this could, in theory, benefit American companies like Lumentum and Coherent — both received $2 billion in investment from Nvidia in March. But analysts warn about the limits of this transition. Neil Shah, vice president of research at Counterpoint, noted that Western competitors have advanced photonic designs but still lack the cleanroom capacity, automated packaging infrastructure, and yield scale needed to absorb the volume of Chinese firms Innolight and Eoptolink within a 12-to-24-month timeframe.
A ban on optical transceivers would mean that major American operators would have to seek local alternatives, potentially driving up costs, as Iyer from The Economist Group pointed out. And the financial impact of this is not abstract. When a data center project gets delayed because it cannot secure enough transformers, the cost is not just the equipment — it is the cost of idle capital, energy contracts already signed, construction crews waiting, and cloud service revenue that cannot be delivered.
Even so, the political will in Washington to reduce dependence on China in the power layer is growing. Iyer sums up the reasoning well: the U.S. administration appears willing to de-risk its AI infrastructure before Chinese technology is fully embedded in it — drawing a parallel to what happened with 5G deployment, when the later removal of Chinese telecom equipment led to delays and mounting costs for carriers.
What is taking shape in practice is a parallel race: on one side, the race for leadership in artificial intelligence; on the other, the race to ensure that the infrastructure supporting that leadership does not become a point of strategic vulnerability. And both races are happening at the same time, with resources, priorities, and timelines that frequently clash. The outcome of these decisions will shape not only the future of artificial intelligence but also the architecture of global technological power for the coming decades. 🌐
