China detailed a five-year plan worth 2 trillion yuan, roughly $295 billion, to build national AI infrastructure connecting data centres across the country, targeting 80% of the technology from domestic companies and using national accelerators instead of imported ones.
The composition target reveals more than the figure. Eighty per cent domestic technology is industrial policy, with long timelines and explicit tolerance for inferior performance in the interim, which is only accepted when the objective is autonomy rather than efficiency.
The open technical question, raised by sector watchers, is which interconnect layer the country will adopt. Connecting data centres at national scale is a different problem from filling one data centre with accelerators, and it's where overall performance is decided.
For the global market, the short-term effect is more demand competing for the same already-tight manufacturing capacity. The medium-term effect is another pole of capacity with its own standards.
Private companies reached that conclusion in the same semester, buying equity in manufacturers and designing their own chips. Here the reasoning runs at state scale: anyone depending on a supplier subject to restrictions doesn't control their own calendar.