全球能源存储：数据中心电池密度上升 市场情绪降温

Global Energy Storage: Data center battery intensity rises assentiment cools Data centers are becoming more battery intense with average duration increasingto c. 5hrs as more are built behind the meter. Batteries are used in data centers forcentralized power (UPS), distributed power (BBU) and energy storage (ESS). While UPS andBBU demand is relatively short duration (<0.25hrs), ESS demand can be longer. With 24%of data centers under construction off grid and 39% of the pipeline off grid, battery energystorage is rising in duration with some hyperscalers having up to 8hours storage. Neil Beveridge, Ph.D.+852 2123 2648neil.beveridge@bernsteinsg.com Brian Ho, CFA+852 2123 2615brian.ho@bernsteinsg.com Assuming 103GW of data center capacity by 2030, cumulative battery capacity willrise to 277GWh.We assume US data center capacity will increase from 41GW to 103GWby 2030. We expect annual data center battery demand to increase 10x from 5GWh to66GWh. Cumulative ESS demand for data centers reached 11GWh in 2025 and will likelyincrease to 277GWh by 2030. This implies the battery /data center capacity GWh/GWmultiplier will rise from 0.3 to 2.7x. Newer projects have a multiplier of 4-8x, implyingfurther scope for battery demand to increase. Kelvin Yuan, Ph.D., CFA+852 2123 2612kelvin.yuan@bernsteinsg.com Increased data center demand raises our total battery demand for the US to486GWh by 2030 which represents 22% CAGR and a 15% increase on previousestimates.70% of incremental demand is forecast to be from ESS which will likely growfrom 49GWh (2025) to 266GWh by 2030. ESS capacity in the US is currently limited at36GWh but set to grow to 102GWh this year as more EV plants are converted to ESS. But while our base case has increased, it's also fair to say that so too has uncertaintyin long term growth.Increased data center capex and the success of lower cost Chinesemodels is creating near term uncertainty in the pace of AI data center build out. While thelonger term outlook looks positive, there is a risk that capex could slow if hyperscalersbecome less confident in returns from the AI business model. Upcoming mid-year resultsfrom the hyperscalers may provide further guidance. Profitability remains another key question, with margins (ex-subsidy) close tobreakeven.While demand looks set to grow by 20-25% CAGR though to 2030, marginsexcluding tax benefits remain negative or close to zero. While guidance is for Korean batterymaker margins to turn positive in 2H26, this remains still the subject of debate. While wethink that higher ESS utilization will raise margins, continued de-rating of EV demand in theUS YTD (-25%) may continue to put pressure on margins in the near term. Korean battery makers valuations have a better risk-reward (12x EV/EBITDA2027 and 5x for 2030) but we would wait for a better entry point.LGES and SDIare pricing in 22-26% revenue CAGR to end of decade (in line with industry growth) anda return to high single digit margins. Longer term we assume mid to high single digitgrowth and margins, which give c. 10% &30% upside from current levels for LGES and SDIrespectively. While these stocks look more attractive, we would wait for a better entry point. BERNSTEIN TICKER TABLE PRICE TARGET CHANGE / ESTIMATE CHANGE IN BOLD O - Outperform, M - Market-Perform, U - Underperform, NR - Not Rated, CS - Coverage SuspendedSource: Bloomberg, Bernstein estimates and analysis. INVESTMENT IMPLICATIONS Korean battery makers look well place to benefit from the growth in ESS demand in the US and the tariff wall & subsidies whichhave been put in place to protect them from Chinese competition. Data center demand in particular could be a material driverof growth with a 10x increase in cumulative demand through to the end of the decade. Even by 2030, the storage/capacitymultiplier of 2.7x could well below the 4-8x multiplier for new data centers which are increasingly being build off grid. With LFPcapacity set to expand significantly this year, Korean battery makers also look better placed to service this demand. SDI is moreexposed to the ESS trend than LGES, with almost half of battery sales set to come from ESS and relatively lower EV exposure.Valuations more attractive given the sell-off driving forward EV/EBITDA multiples to 12x and long term (2030) multiples to 5x.But with questions surrounding the sustainability of AI capex and continued uncertainty on 2H26 margins we still believe thereare better entry points ahead. We remain Market-Perform on LGES, LG Chem and SDI. Model updates: we make a series of mostly minor update based on our recent demand update. For the details of the modelupdates, please see page 14 onwards. VALUATION COMPS TABLE DETAILS DATA CENTER DEMAND IN US We expect cumulative U.S. data center power capacity to reach 103 GW by 2030, driving incremental battery demand to 69GWh in 2030 and cumulative battery installations to 277 GWh. Based on more than 20 announced U.S. data center projects, w