转型投资：数据中心——高耗能、高耗水、高金属需求，5万亿美元机遇

Data centers: power-hungry, water-thirsty,metals-needy, $5tn opportunity Investment Strategy Weneed more energy than Japan, more water than NYData center sites have exploded, with 11,200+ globally today, while capacity is set to double to 200GW by 2030E. Meanwhile, AI-specific capacity has tripled in just 18months. This is a transition‑era resource shock, attracting $7tn of investment by 2030E,consuming more electricity than Japanannually, drinking New York’s entire annual wateruse, and accounting for up to 6% of global critical‑metal demand. There is not enough…and hyperscalers secured all reservesBy 2030, data centers could exceed 3% of global electricity demand, straining grids withlarge, inflexible loads as GPU power use grows c.30% annually. Hyperscalers haveresponded by moving upstream, driving 80% of top‑10 corporate clean‑energyprocurement, accelerating battery storage and locking in nearly half of SMRs capacity. Behindthe rise of AI lies an epic water bill A 100‑word AI prompt uses half a liter of water. Global data center water demand is setmore than double by 2030, with over a third of hyperscalers’capacity already in regionsunder severe water stress. A $58bn water infrastructure shortfall in the US is emergingas a critical bottleneck. And more than half of water sector stakeholders have yet tofactor AI capacity and advanced manufacturing into their resource planning. Critical metals,critical to delivery, critical bottleneckData centers are highly metal‑intensive, requiring up to 75 tons per MW, led by copper and aluminum, alongside tin and specialty metals, with cooling and back‑up poweraccounting for c.75% of metals intensity. Although metals make up <5% of data centerscapex, supply bottlenecks are delaying projects, with transformer lead times now up tofour years and prices more than 60% above pre‑2020 levels. Scarcity is the mother of innovation Advanced cooling technologies such as closed loop liquid, direct to chip and immersionsystems are scaling, alongside circular solutions that can reduce on-site cooling wateruse by up to 90%. In parallel,‘bridge‑to‑grid’strategies are evolving into longer‑durationhybrid models. Early deployments of long duration storage, especially iron air, arecreating viable alternatives that sidestep critical minerals constraints. Rising server rackdensities are accelerating a shift toward more integrated power, cooling andelectrical‑equipment solutions that help mitigate grid and resource limits. $5tn investment opportunity across 65+ stocksThe opportunity lies in the physical enablers. We identify $5.5tn across 67 Buy‑rated BofA-covered stocks across power generation, electrical equipment, metals, andwater/cooling solutions (Exhibit 3). We also note that the BofA custom baskets team hasa range of baskets on global AI ecosystem (Exhibit 2). Contents The Data Center transition opportunity: Exhibit1: A transition‑investment approach centered on the physical systems that enable AI growth.The AI Data Center investment opportunity is moving up the value chain into energy supply, electrical and water infrastructure, and strategic transition metals BofA Baskets on global AI ecosystemWe note that the BofA custom baskets team has a range of Exhibit2: BofA custom baskets on global AI ecosystemChina AI Supply Chain (MLAPAISC) AI Infrastructure (MLAIINFR), US Power Supercycle (MLUSPOWR), EU AI Power Surge (MLESURGE), EU Power Grids(MLEUGRID) and full range of BofA AI baskets of ecosystem. baskets on global AI ecosystem, such as China AI Supply Chain(MLAPAISC) AI Infrastructure (MLAIINFR), US Power Supercycle(MLUSPOWR), EU AI Power Surge (MLESURGE), EU Power Grids(MLEUGRID). Note: These are proprietary baskets that arestructured and maintained by the BofA Securities trading desk Exposure extends across 65+ names and $5tn Market cap We highlight a screen of 67 BofA Global Research-covered companies with a Buy ratingwith a combined market cap of c.$5.5trn, identified based on exposure to the AI datacenter build-out across power supply, electrical and water infrastructure, and strategictransition metals. Data center buildout. Did you know that: •AI‑focused data‑center capacity has 3x in 18 months across a global base of11,200+ facilities, over one‑third of which are in the US1•1GW of DC capacity costs c.$50bn, c.2/3 spent on chips and networking2•A typical DC creates 1,000-10,000 construction jobs but only 50-300permanent roles3•By 2030, global DC capacity could double to c.200GW with totalinvestments of $7trn4•In 2025, DCs accounted for 485 TWh or more than 1.5% of global electricityconsumption (+17% YoY)4•By 2027, a single AI server rack (the size of a large refrigerator) could have apeak power demand equivalent to 65 households4.•By 2030, DCs to use more electricity than Japan, driving >20% of demandgrowth in advanced economies, with renewables supplying 50% of increase4•US and European high‑voltage grid transmission projects often take 7–10years to permi