The biggest names in American tech are writing checks that would make sovereign wealth funds blush. Microsoft, Amazon, Alphabet, and Meta have collectively committed over $650 billion in capital expenditure for data center expansion in 2026, with some analyst estimates pushing that figure closer to $750 billion.
There’s just one problem: a significant chunk of the hardware keeping these facilities humming comes from China.
The supply chain paradox
The US is simultaneously trying to outpace China in AI development and depending on Chinese manufacturers to supply the electrical infrastructure that makes AI data centers possible. Chinese firms supply roughly 30% of US transformers and switchgear, over 40% of batteries used in data center operations, and about two-thirds of global optical transceiver units.
The US currently faces an estimated 15% shortfall in power transformers and an 8% deficit in substations projected for 2026. These aren’t minor inconveniences. Power transformers are the backbone of electrical distribution, and without enough of them, data centers simply cannot connect to the grid at the scale hyperscalers need.
To put the power demands in perspective, a major facility linked to OpenAI is projected to consume 1.2 gigawatts upon completion in 2026. That’s roughly the output of a nuclear power plant, dedicated to a single data center campus.
A ban that could backfire
Washington is reportedly discussing a potential ban on Chinese optical transceiver imports, with implementation targeted for late 2026. Analysts warn that such a ban could significantly escalate costs and disrupt AI infrastructure builds across the board. Domestic alternatives exist, but they’re more expensive and lack the manufacturing scale to fill the gap overnight.
The irony is striking. US export controls already restrict advanced AI chips from reaching China, which has pushed Chinese firms to develop their own alternatives. Now the US finds itself on the other side of a similar dependency equation, needing Chinese electrical components to build the very data centers that train the AI models those chip restrictions are meant to protect.
What this means for the industry
Research firms including Counterpoint and Wood Mackenzie have flagged concerns about how rising operational costs from imported materials could compress profit margins for the tech giants driving AI infrastructure buildouts.
If import restrictions materialize, hyperscalers will likely be forced to pursue higher-cost domestic alternatives, accelerating investment in US manufacturing but straining financial resources in the near term. That transition period, the gap between cutting off Chinese supply and scaling domestic production, represents the most vulnerable window for American AI ambitions.