BofA Raises China AI Data Center Capex Forecast to $327 Billion by 2030
Miles Bennett
BofA Securities lifted its China AI data center capex forecast to $327 billion by 2030 at a 24% CAGR, arguing the AI compute race is now a race for power and materials — with copper, fiber, and transformers facing structural supply tightness.
$327 billion — what does this number really mean?
BofA raised China AI data center capex to $327 billion by 2030, a 24% CAGR, accounting for roughly 20% of global AI capex by then.
Global AI capex was simultaneously raised to above $1.7 trillion in 2030, up sharply from $260 billion in 2025.
This means → China is no bystander in global AI investment. By 2030, for every $5 spent on AI capex worldwide, $1 goes to China.
What is the real bottleneck in the AI race?
BofA's core thesis: the AI compute race is increasingly becoming a power-infrastructure race.
Global data center installed capacity is projected to grow from about 100 GW today to nearly 300 GW by 2030. Per-rack power is climbing from a legacy 10–15 kW, along Nvidia's roadmap, to 100–120 kW now — and could exceed 1 MW in next-gen systems. In plain terms = a single server rack's power draw is jumping from "one home air conditioner" to "a small office building."
China's data center capacity is expected to grow from 29 GW in 2025 to 77 GW in 2030, with power consumption rising from 121 TWh to 318 TWh — about 2.5% of national electricity use.
BofA also notes that chip restrictions are pushing some Chinese internet firms to deploy compute in Southeast Asia, so China's actual data center power draw may understate the true pace of AI demand growth.
Which of the five key materials is tightest?
Copper: China data-center-related copper demand is projected to rise from 341 kt in 2025 to 1,190 kt in 2030, a 28% CAGR. Global copper supply is expected to face a 491–754 kt shortfall in 2026–2027. This means → the copper squeeze is imminent, not a distant hypothetical.
PCB materials: Low-end copper foil is in oversupply, but AI servers are driving a surge in high-end copper foil demand. Conversion barriers are high and customer qualification cycles typically exceed one year — the gap will not close quickly.
Optical fiber: Demand is shifting rapidly from traditional telecom to AI data centers. Preforms — the core raw material for fiber, with long expansion cycles and high technical barriers — create a structural bottleneck. BofA rates Jiangsu Zhongtian Technology as a buy, projecting roughly 75% EPS CAGR for 2026–2027.
Uranium: Viewed as a core strategic resource for scalable, zero-carbon, stable baseload power in the AI era. In plain terms = for 24/7 uninterrupted data center power, nuclear is currently the best-qualified zero-carbon option. BofA forecasts uranium prices rising 47% / 29% year-on-year in 2026/27.
What structural advantages does China have in power supply?
China's commercial electricity prices are 30%–60% lower than the U.S./EU; effective reserve margins sit at around 30%; and transmission and distribution infrastructure averages under 20 years old (versus 40+ in the U.S. and Europe), with higher grid stability.
This means → China's grid is newer, cheaper, and has more headroom — exactly the three metrics AI data center siting values most.
China operates 46 ultra-high-voltage transmission corridors and a deep power-equipment manufacturing chain. Transformer exports are expected to grow 30% year-on-year in 2026; a global transformer shortage is projected to last at least until 2029, with high-voltage transformer lead times stretching to 3 years.
Dongfang Electric is China's only company capable of exporting medium-to-large gas turbines. Its G50 50 MW units have already been sold — 10 units — to a Canadian data center customer, with management planning to lift export capacity to 23 units by end-2027 and 45 units by end-2029.
Why is liquid cooling the next breakout segment?
Rack power density is now exceeding the ceiling for air cooling (roughly 40 kW per rack), and tightening energy-efficiency regulations in China are accelerating the shift. In plain terms = chips are getting so hot that blowing air is no longer enough — heat must be carried away directly by liquid.
BofA projects China's data center liquid-cooling penetration rising from 30% in 2025 to 70% in 2030, with liquid-cooling demand growing from 1.4 GW to 9.5 GW — a 47% CAGR.
The overall data center cooling market is expected to reach RMB 70 billion by 2030. Immersion cooling — submerging entire servers in coolant — is projected to grow from about 5% market share to roughly 17%, reaching RMB 16 billion.
Where is the opportunity in energy storage and power systems?
BofA forecasts global battery energy storage (BESS) additions at a roughly 23% CAGR from 2025 to 2030, with global AI data center BESS additions reaching 70 GWh by 2030.
China's AI data center power supply systems (UPS + HVDC + SST) market is projected to grow at about 25% CAGR. Nvidia is actively pushing the supply chain toward an 800 V DC architecture. This means → the power architecture itself is undergoing a generational shift, and suppliers that keep pace with the new standard will capture incremental orders.
Content is for reference only, not financial advice.