China AI power infrastructure, grid infrastructure and energy storage: China’s AI expansion could turn grid infrastructure and energy storage into major beneficiaries

Overview

This industry report examines how China’s power-cost advantage, expanding generation base and policy-led coordination of computing and electricity infrastructure could support AI deployment. Bernstein concludes that the resulting demand for transmission, renewable integration and battery storage creates a multi-year opportunity across the AI power value chain.

Core views

Bernstein’s starting point is that China’s AI ecosystem can compete increasingly on economics rather than frontier-model capability alone. The report argues that several Chinese large language models offer broadly comparable intelligence at materially lower output-token prices than leading US models. It attributes this advantage to more efficient model architectures, lower infrastructure costs and cheaper electricity. As AI use shifts from training a limited number of frontier models toward high-volume inference, the report views intelligence per dollar as a more important competitive measure. The report frames China’s compute catch-up as a substantial power and infrastructure challenge. China currently has about 5 ZFLOPS of AI compute, roughly 15% of the US level. Bernstein assumes US AI compute rises to 219 ZFLOPS by 2030 and 511 ZFLOPS by 2035, and models a scenario in which China reaches 59 ZFLOPS in 2030 and 511 ZFLOPS in 2035. That requires a 100-fold expansion from China’s current compute level. The analysis assumes China must rely mainly on domestic AI chips because US restrictions continue to constrain high-end chip supply. Chinese new-chip efficiency is assumed to improve from 0.4 TFLOPS/W currently to 3.2 TFLOPS/W in 2035, while US efficiency rises from 1.8 to 5.7 TFLOPS/W. Bernstein expects the efficiency gap to narrow but persist, meaning China needs more chips and more data-center capacity for equivalent compute. On that basis, China would need AI-dedicated data-center capacity of 214GW by 2035, a roughly 15-fold increase from 14GW currently; the report notes this would be 1.6 times projected US capacity in GW because of lower chip efficiency. China’s data-center electricity demand is projected to grow from 202TWh in 2025 to 622TWh in 2030 and 1,476TWh in 2035, a 24% CAGR. This adds 1,274TWh of demand and accounts for 28% of China’s total electricity-demand growth over the decade, while total national demand rises from 10,163TWh to 14,741TWh, a 3.8% CAGR. Data centers would represent about 10% of China’s electricity demand by 2035, versus 16% in the US. Bernstein notes uncertainty over how quickly announced compute capacity becomes grid-connected operating capacity: other industry estimates put 2030 data-center demand at 300-800TWh, compared with its 622TWh forecast. Bernstein believes China is better positioned than many markets to supply this load. It produces more than twice as much electricity as the US, added more than 500GW of generating capacity in 2025, and has average industrial electricity tariffs of about US$56/MWh. Major AI and cloud hubs such as Ulanqab, Guizhou, Qingyang and Zhongwei have tariffs around RMB0.33-0.37/kWh, with some Xinjiang projects near RMB0.32/kWh. China added 314GW of solar and 119GW of wind capacity in 2025, while renewables supplied 22% of generation, or about 2,300TWh. The report also highlights lower renewable-equipment costs and nuclear projects completed in 5-7 years at roughly US$2,000-2,500/kW, supporting a combination of low-cost intermittent generation and reliable low-carbon baseload power. Policy is central to the proposed buildout. The 2022 East Data, West Computing program directs latency-insensitive training and batch workloads toward western hubs with abundant renewable resources and lower land and power costs, while eastern clusters serve real-time inference closer to users. The 2024 Green Low-Carbon Data Center Action Plan requires new national-hub data centers to source more than 80% green power, with PUE below 1.25 for new large facilities and below 1.20 for national-hub projects. The newer Computing-Power Grid Coordination framework links computing capacity, renewable generation, storage and grid infrastructure; the report cites industry commentary indicating 15-20% storage ratios and 2-4 hours of duration for AI-focused data centers. Major cloud and data-center operators, including Alibaba Cloud and Tencent, have committed to 100% clean or green energy by 2030. Grid expansion and storage are therefore the report’s principal stock-market implications. China had 43 UHV transmission lines operating by mid-2026 and plans more than RMB5tn of grid investment in 2026-30, more than 420GW of west-to-east transmission capacity, 15 new UHV DC lines, and a grid able to integrate over 2.8TW of wind and solar by 2030. Bernstein argues that transmission is needed to prevent renewable curtailment and move low-cost western generation to computing demand. Battery storage is equally important for managing intermittency, peak regulation and frequency regulation. China holds an 80% share of the lithium-ion battery market and had more than 3,000GWh of manufacturing capacity at end-2025. Global BESS battery demand grew an estimated 85% year on year to 556GWh in 2025; Bernstein expects China to install about 300GWh this year, up 95% year on year, versus 70-80GWh in the US. Among its covered names, the report identifies CATL as a beneficiary through grid-scale storage leadership and Sungrow through power conversion, renewable integration, storage systems and grid infrastructure.

Analysis framework

Bernstein compares Chinese and US AI economics and compute capacity, then translates a China-to-US compute-parity scenario into chip-efficiency, data-center-capacity and electricity-demand requirements. It tests those requirements against China’s generation capacity, power costs, renewable and nuclear supply, policy framework, transmission plans and battery-storage capability before identifying exposed companies.

Methodology notes

• Industry AnalysisSupply-demand framework

  Power-demand and generating-capacity scenario analysis

  The report projects data-center electricity demand from AI-compute growth and compares it with China’s projected total power demand and capacity additions.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI-compute-to-power-infrastructure value-chain transmission

  The analysis links AI data-center expansion to renewable generation, UHV transmission, grid equipment and battery storage demand.

• Industry AnalysisCost curve analysis

  Relative power, infrastructure and renewable-cost comparison

  The report uses lower Chinese industrial tariffs, data-center tariffs and renewable and nuclear construction costs to explain China’s AI-service cost advantage.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• CATL (300750.CH; 3750.HK)

  Direct beneficiary of AI-driven power infrastructure and grid-scale storage demand.

  Strengths

  Leadership in battery technology and grid-scale storage systems.

  Risks

  The report’s demand case depends on AI compute expansion, renewable deployment and storage investment.

• Sungrow (300274.CH)

  Direct beneficiary through power conversion, renewable integration, energy storage systems and grid infrastructure.

  Strengths

  Exposure across renewables, power conversion, storage and grid integration.

  Risks

  The report’s demand case depends on AI compute expansion, renewable deployment and grid investment.

Key data

• China data-center electricity demand202TWh in 2025 to 1,476TWh in 203524% CAGR; Bernstein’s AI-compute-parity scenario.
• China AI compute5 ZFLOPS currently; 59 ZFLOPS in 2030; 511 ZFLOPS in 2035The 2035 level is assumed to match projected US compute capacity.
• China AI-dedicated data-center capacity214GW by 2035About 15 times current levels in the report’s scenario.
• China total electricity demand10,163TWh to 14,741TWh by 20353.8% CAGR; data centers account for 28% of incremental demand.
• Grid investment planMore than RMB5tn during 2026-30Includes over 420GW of west-to-east transmission capacity and 15 new UHV DC lines.
• China BESS installationsAround 300GWh this yearExpected growth of 95% year on year.

Impact & implications

Bernstein sees the AI buildout as reinforcing demand for transmission equipment, renewable integration and battery energy storage. It identifies CATL and Sungrow as the most direct covered beneficiaries of the required investment in power reliability, renewable integration and grid expansion.

Risks

• China’s ability to reach the modeled compute scale depends on progress in domestic advanced AI-chip design, manufacturing and efficiency.
• The timing of data-center demand is uncertain because announced compute capacity may not convert into grid-connected operating capacity as quickly as expected.
• Without sufficient transmission investment, renewable curtailment could rise and slow renewable deployment.

What to watch

• Progress in domestic Chinese AI-chip efficiency relative to the report’s 3.2 TFLOPS/W 2035 assumption.
• Conversion of announced data-center capacity into grid-connected operating capacity.
• Implementation of more than RMB5tn of planned grid investment, new UHV lines and west-to-east transmission expansion.
• Compliance with green-power, PUE and storage requirements for national computing hubs.
• China’s BESS installation pace and the expansion of renewable power procurement by cloud and data-center operators.