Will China overtake the US in the AI race? (In charts)

It examines whether China will overtake the US in the AI race by charting energy, data-centre capacity and deployment, and compute resources. China has massively expanded electricity generation (about 10,578 TWh in 2025 vs roughly 4,520 TWh in the US) and can build power lines and data centres much faster, with Western hubs like Ulanqab offering very cheap power (~$0.05/kWh), huge wind capacity, cold climates and rapid build timelines. However, only a tiny share of new Chinese electricity goes to data centres, data-centre usage and capacity remain lower than in the US (117 TWh and 24 GW in China vs 224 TWh and 56 GW in the US in 2025), and logistics (fiber, rack utilisation, GPU shortages) constrain growth. US projections for data-centre power rise steeply, and delivery bottlenecks in the US could delay but not negate expansion.

The compute and chip picture decisively favors the US. Measured in H100-equivalents the US has about 15 million vs China’s 2.8 million (half of which are rented or smuggled abroad), US AI spending dwarfs China’s ($764B vs $102B in 2026), and Nvidia outproduces Huawei by orders of magnitude. Foundry and tool advantages - TSMC’s node lead over SMIC and the EUV machine export ban - mean China lags in cutting-edge chips and faces multi-year timelines to close the gap. The conclusion: China’s energy and build advantages matter, but constrained access to top-tier compute, chips and investment make a near-term overtaking of the US in advanced AI unlikely.