American technology companies are committing record sums to artificial intelligence infrastructure at the same time as Washington tightens restrictions on China's access to advanced chips. A report published by CryptoBriefing on 6 September 2026 argues that this build-out rests on a supply chain in which Chinese manufacturers hold significant positions, creating a dependency that runs in the opposite direction to US export controls.
The scale of the build-out
According to the report, Microsoft, Amazon, Alphabet and Meta have collectively committed more than 650 billion US dollars in capital expenditure for data centre expansion in 2026, with some analyst estimates placing the figure closer to 750 billion dollars. Power demand is rising in step. A major facility reported to be linked to OpenAI is projected to consume 1.2 gigawatts on completion in 2026, roughly the output of a nuclear power plant devoted to a single campus.
Where Chinese suppliers hold positions
The report states that Chinese firms supply roughly 30 per cent of US transformers and switchgear, more than 40 per cent of the batteries used in data centre operations, and about two-thirds of global optical transceiver units. These components sit on the electrical path between the grid and the server rack, rather than inside the AI models themselves, but without them the facilities cannot operate.
The United States faces an estimated 15 per cent shortfall in power transformers and an 8 per cent deficit in substations projected for 2026, according to the report. Without sufficient transformers, new facilities cannot connect to the grid at the scale the hyperscalers require.
A reported ban that could cut both ways
Washington is reportedly discussing a ban on imports of Chinese optical transceivers, with implementation targeted for late 2026. No such measure has been formally announced. Analysts cited in the report warned that a ban could significantly raise costs and disrupt AI infrastructure construction, because domestic alternatives exist but are more expensive and lack the manufacturing scale to replace Chinese supply quickly.
The dependency in context
The report frames the situation as a mirror image of existing US policy. American export controls restrict advanced AI chips from reaching China, which has pushed Chinese firms to develop domestic alternatives. The United States, the report argues, now faces a comparable dependency in electrical components, needed to build the very data centres that train the models those chip controls are designed to protect.
Research firms including Counterpoint and Wood Mackenzie have flagged concerns that rising costs for imported materials could compress profit margins for the companies funding the build-out, according to the report. If import restrictions do materialise, hyperscalers would likely turn to higher cost domestic suppliers, accelerating investment in US manufacturing while straining financial resources in the near term. The report describes the transition period between cutting Chinese supply and scaling domestic production as the most vulnerable window for American AI ambitions.
What is established and what is merely claimed
Established by the source: the four named hyperscalers have committed large 2026 capital budgets, reported in aggregate above 650 billion dollars; research firms Counterpoint and Wood Mackenzie have raised concerns about cost pressure from imported materials; and discussions in Washington about a possible ban on Chinese optical transceiver imports have been reported, not announced.
Merely claimed: the specific supply share figures of roughly 30 per cent for transformers and switchgear, over 40 per cent for batteries and two-thirds for optical transceivers, the 15 per cent transformer shortfall, and the 1.2 gigawatt projection for a facility described as linked to OpenAI all originate from this single report and the analyst estimates it cites. They have not been independently verified. Whether any ban on Chinese optical transceivers will actually be imposed remains unknown, and no US official is quoted confirming the discussions in the material supplied.