美国通信基础设施：数据中心——市场传闻...关于我们数据中心启动的顶级投资者反馈（3部分中的第2部分）

Data Centers: Word on the Street...top investor feedback on ourdata center initiations (part 2 of 3) We initiated coverage on US Communications Infrastructure just over a month ago, includingtwo Outperforms in the Data Center space (DLR, $218; EQIX, $1,128). Since then, we’vespoken with 70+ investors, the company, and our industry contacts. Because this sectoris less contentious than our other one (neoclouds), most of our conversations were morequestions than disagreements—cataloged below with detailed answers in the note Madison Rezaei+1 917 344 8622madison.rezaei@bernsteinsg.com Nancy Wu+1 917 344 8545nancy.wu@bernsteinsg.com Is it possible that you are dramatically understating global data center demand?Sure, but it’s upside for the data center names and marginal at that. The DLR + EQIX buildtrajectories would also have to shift, which is not planned, but could provide incrementalupside. What makes you confident in your hyperscale build numbers? How do they tie withother reports? We have triangulated using a number of sources and believe we are in range on owned +leased capacity - a total of 58GW in 2025 growing to 119GW by 2030 aimed at hyperscale. If you like enterprise collocation so much, why do you prefer DLR to EQIX?DLR is underpenetrated in enterprise relative to their scale and footprint. Coorespondingly,we believe they should have a faster enterprise revenue growth rate (18.6% relative to11.6% CAGR 25-30). How should we think about the play between development and stabilized assets,especially for DLR, but also for EQIX? This is a tricky one - the two have different risks, returns, and investor bases. While weprovide significant detail in the note, we generally feel that DLR especially has done a nicejob of delineating the asset classes with their JVs and closed-end fund. What are the appropriate multiple ranges for these guys? They look expensive…Datacentersarelooking expensive, but rightfully so! They’re great assets with strong tailwindsand have appreciated appropriately. Despite multiples being in the ~24-26x range today,we maintain our 25-26x forward P/AFFO multiples and could justify further expansion overtime. Are data centers facing competitive risk from space-based deployments?We strongly believe we’re years away from this threat becoming even semi-real, but it eventually could if some major tech and economic hurdles are overcome. DLR and EQIX areamong the safest of the data centers if that did play out. On the whole, we maintain our Outperforms, with DLR being our top pick, followedby EQIX. We like DLR’s enterprise colo ramp, particularly backed by AI tailwinds,and expect to see continued progress in 1Q earnings. EQIX has lots of headroomover the longer term as their builds complete in 2029-30. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We rateDLROutperform on a differentiated growth and de-risked development story, underpinned by a stable hyperscalerrevenue floor, underappreciated enterprise upside driven by global interconnection density (with strong growth in the <1MWsegment), and a capital-light development strategy that supports a durable valuation. Our $218 price target is based on 26x our2027E AFFO per share of $8.31. We rateEQIXOutperform on a long-term enterprise capacity reset, supported by its best-in-class interconnection ecosystemand metro footprint, with upside driven by a renewed build cadence and improving visibility into future capacity growth—albeitpartially offset by near-term capex intensity and a measured contribution from xScale. Our $1,128 price target is based on 25xour 2027E AFFO per share of $44.92. DETAILS IS IT POSSIBLE THAT YOU ARE DRAMATICALLY UNDERSTATING GLOBAL DATA CENTER DEMAND? Of course. Nobody knows what is going to happen with AI demand. We’re modeling AI demand at ~60 GW by 2030, but it couldeasily be twice that, particularly if the LLMs find their way towards profitability through means other than power efficiency (or themarket tolerates unprofitability for a longer time). Realistically for DLR and EQIX, even if we are wildly undercounting AI demand,it’s only moderate upside given their current build plans. We’ve written at length about what gives us confidence in our numbers (power efficiency, hyperscaler self-builds, hyperscalernetwork visibility). To stress test, we’ve sensitized our supply and demand model, assuming an incremental ~50GW of demandby 2030, with ~43GW of incremental supply. In this high case, we assume no easing of data center vacancy during the forecastperiod. If this is how the market evolves (it’s not our belief, but we’re here for you, AI bulls), our revenue numbers would be ~7% higherfor DLR by 2030 and ~10% higher for EQIX. Impact is muted because a) the companies would have to change their build plansto take advantage of this demand growth, b) DLR + EQIX represent a small portion of the industry and already have very highoccupancy rates, c) they are increasingly focused one enterprise versus hyper