Bernstein: China's AI Computing Power Expansion to Ignite Long-Term Demand for Grid Energy Storage
nashnova research
Bernstein's September report projects China's data-center power demand will rise from 202 TWh to 1,476 TWh by 2035 — a 24% CAGR — making grid infrastructure and energy storage a core long-term investment theme.
Power demand set to rise sevenfold — where does the number come from?
China's AI compute capacity is currently only about 15% of the US level, but domestic chips and infrastructure are scaling fast.
This means → closing that gap alone would generate massive incremental power demand.
Bernstein's projection: data-center electricity consumption goes from 202 TWh in 2025 to 1,476 TWh in 2035, a ten-year CAGR of 24%.
Why is Chinese AI so much cheaper than American AI?
China's leading large models score competitively on intelligence benchmarks, yet they are priced at a fraction of their US equivalents.
In plain terms = same "smartness," much lower price — creating a "more intelligence per dollar" edge.
That edge rests on three pillars: more efficient model architectures + lower infrastructure costs + cheaper electricity.
This reflects something broader: China's AI competitiveness is not just an algorithm story — electricity cost itself is a moat.
Where does all that power come from — and is it enough?
China's total generation capacity already exceeds twice that of the US; in 2025 alone it added more than 500 GW of new capacity.
Data-center electricity costs in AI hubs such as Inner Mongolia, Ningxia, and Gansu rank among the lowest in the world.
This means → cheap renewables plus cost-competitive nuclear keep AI operators' power bills far below those of overseas peers.
How is policy tying "compute" to "power"?
The "East Data, West Compute" programme — routing eastern computing demand to cheaper western power — along with green-data-center rules and the "compute-power coordination" framework, is driving joint expansion of compute and grid infrastructure.
New data centers in national-level compute hubs must source at least 80% of electricity from renewables.
Leading operators have pledged 100% renewable energy by 2030. In plain terms = no green power, no building permits.
How much will the grid and storage buildout cost?
Between 2026 and 2030, China's grid investment plan exceeds ¥5 trillion, focused on ultra-high-voltage lines that connect western clean energy to eastern compute clusters.
On the storage side, China holds roughly 80% of global lithium-battery capacity; 2026 storage installations are forecast to grow 95% year-on-year to 300 GWh.
This means → renewable-energy absorption needs plus AI power-reliability needs give storage growth a "twin engine."
Who benefits most — and what is the key variable?
Bernstein sees the highest investment value in grid infrastructure and energy storage.
Top picks: CATL (宁德时代) — the leading grid-scale storage system provider; Sungrow (阳光电源) — positioned across renewables, power conversion, storage, and grid integration.
This reflects a critical point: the endgame test for this AI investment thesis is not about chips — it is whether power infrastructure can keep pace with compute expansion.
市场有风险，内容仅供研究参考，不构成投资建议。