人工智能基础设施-中国(HA)：传统与转型领域的AI材料及电力

Traditional & transitional:AI materials &Power Industry Overview Global/China AI CAPEX of US$1.7tn/US$330bn by 2030 16 June 2026 Our global team has raised its 2030 global AI CAPEX forecast to US$1.7tn (vs US$260bnin2025). Per Bloomberg News, China may spend c.RMB2tn on AIDC CAPEX over the nextfive years. We expect China’s AIDC capex to grow from US$91bn in 2025 to US$330bnin 2030, driven by leading internet & cloud platforms, alongside govt. funding. In this 3rdAI infrastructure thematic note (see prior notes here#1,#2), we focus on traditionalareas transitioning into the AI value chain and where China is strong (such as poweringAI, AI metals, PCB materials, optical) to trigger re-rating through domestic substitutionand/or exports. EquityChinaAI Infrastructure Matty Zhao>>Research AnalystMerrill Lynch (Hong Kong)+852 3508 4001matty.zhao@bofa.com Yibing Xia>>Research AnalystMerrill Lynch (Hong Kong)+852 3508 8045yibing.xia@bofa.com AI materials:What materials feed into AIDC? We identified fivekey materials within the AIDC value chain that may face structuralshortages (Exh8)1. Copper:top pick in commodities. We expect DCs to a/c for 6% ofdemand in 2030 vs. 2% in 2025; powering AI will indirectly drive copper demand onpersistent supply tightness. BuyZijin/JXC/CMOC; 2.PCBmaterials:copper foil&glass fiberare oversupplied in low-end products, while PCB upgrades & difficulties inshifting capacity should result in shortages of high-end products and support price hikes.BuySinoma S&T;3. Optical fiberdemand is shifting from traditional telecom to AIDCto support high-speed, long-distance data transmission while supply mismatches haveled to shortages in high-end products. BuyZhongtian. 4.Minor materials(rare earths,tungsten, uranium) are small in volume but critical in value, supporting magnets,precision manufacturing, and power generation. BuyJL Mag/ CGN Mining. 5.LiquidCooling:we forecast a 36% 2025-30E CAGR in China due to high power intensity and20-50x higher heat transfer efficiency than air.BuyShenling/ AVC. Peter Wang>>Research AnalystMerrill Lynch (Hong Kong)+852 3508 7185peter.wang2@bofa.com Fiona Liang>>Research AnalystMerrill Lynch (Hong Kong) Sun Jung Lee>>Research AnalystMerrill Lynch (Seoul) Doris Kao>>Research AnalystMerrill Lynch (Taiwan) Gary Tsang>>Research AnalystMerrill Lynch (Hong Kong) Powering AI: Global AI power from 100GW to 300GW See Team Page for List of Analysts Owing to low costs (China tariffs 30-60% lower vs. the US/EU), high adequacy, young &well-developed grid, and a robust equipment chain, POWER represents a distinctcompetitive advantage for China in AI. We forecast cumulative DC power demand togrow from 100GW in 2025 to 300GW in 2030. Direct DC power consumption may a/c for3.7% of global power demand by 2030. Five themes (Exh9):1.globaltransformertightness: China grid capex growth & solid export demand for China and Korea. BuySieyuanon export growth,Jinpanfor its direct US factory & high AI exposure;LS/HDHyundaion Korea export (US preferred); 2. Gas Turbineshortages till 2030: forecastglobal gas turbine orders of c.120GW over next 3 years. BuyDongfang(generator)/Yingliu(blades); 3. Chinesediesel engineplayers are well penetrated the US as AIDCbackup power. BuyWeichai; 4)ESS:forecast 23%/27% 2025-30E CAGR for globalBESS/ AIDC-related BESS demand. BuySungrow/CATL; 5) forecast 25% 25-30E CAGRfor China AIDCpower supply systemmarket. BuyKehua(UPS)/Megmeet(PSU)/Delta. This research report provides general information only. No part of thisreport may be usedor reproduced or quoted in any manner whatsoever in Taiwan by the press or otherpersons without the express written consent of BofA Securities.>> Employed by a non-US affiliate of BofAS and is not registered/qualified as a research analyst AI capex drives materials and power Exhibit3:By end-2030, we think Nvidia’s roadmap could lead to >1.5megawatt racks for the Feynman era, nearly 100x higher than astandard server CPU rack at 10-15 kilowattsPower capacity per rack (kilowatts) across various generations of Nvidia Forecast global/China AIcapexto reach US$1.7tn+/US$327bn in 2030 platforms vs. a traditional server rack Data center power demand by region (GW) Exhibit7:Global ESSadditionalinstallation togrow from c.300GWh in2025 to 867GWh in 2030EGlobal BESS new installation (GWh) forecast Exhibit6:Forecast global gas turbine orders c.120GW for the nextthree yearsGlobal gas turbine order forecast (bars) and estimated production capacity (line) (GW, simple cycle basis) Exhibit8:What materials feed into AIDCWe present key AImaterials companies across metals, PCB materials, thermal materials, optical materials, and electronic/semiconductor materials. Executive summary China AI capex to reach US$330bn in 2030E Our global team estimates that AI-related capex will reach US$1.7tn+ by 2030E,supported by the shift toward accelerated computing, stiff competition among cloudplatforms, sovereign AI buildout, and growing enterp