数据中心：顶级数据中心市场的样子——包括DLR和EQIX的布局

Data Centers: What a top-tier data center market looks like—including the DLR and EQIX footprints We’re fielding a lot of questions on data center buildouts, at least in part driven by the dailyheadlines on power delays, NIMBYism, and supply chain constraints. Most of our answers boildown to “it depends on what market(s) you’re looking at - we feel pretty good about MajorMetros and the players building / buying there”. In this note, we do a comparison of markettypes and a deep dive into three of the top U.S. markets. Madison Rezaei+1 917 344 8622madison.rezaei@bernsteinsg.com Nancy Wu+1 917 344 8545nancy.wu@bernsteinsg.com We think of three data center market archetypes: Major Metro markets (typically consideredthe top 8-10 in North America), Minor Metro markets (sometimes thought of as Tiers2-4, smaller cities or the extreme outskirts of Major Metros), and Rural markets.The twopublic data center company footprints, DLR (OP, $232) and EQIX (OP, $1,222) areoverwhelmingly concentrated in Major and Minor Metros; they are among the mostinsulated from market noise. Major Metros are and will continue to be the most important markets for latency-sensitiveuse cases (including inferencing), as well as enterprise colocation. These markets arewell established and hard to build in. Power queues are long, build costs are high, andsupply chains are tight. Existing capacity and in-progress, permitted builds are valuable.We are seeing an increase in NIMBYism in these markets, particularly NoVa and Phoenix,in part driven by rising power costs. We believe the established builders will manage thisappropriately, but it is an emerging risk. After Major Metros, overflow goes into Minor Metros. This works where there is marginallyless latency sensitivity and/or marginally more price sensitivity. There has been significantgrowth in these markets over the last ~5 years, which we anticipate continuing. Finally, training, latency-insensitive, and/or very price sensitive workloads take place inRural markets. These markets tend to have larger facilities, more speculative builds, andless experienced developers than Major Metros. On the positive side, there is faster poweravailability and lower cost to build. On the negative side, there is not as clear visibility tosustained demand. In the U.S., the biggest data center market is Northern Virginia (NoVa), which has ~8GW ofsupply today, ~5GW under construction, and power queues reportedly in the seven-yearrange. Pricing in NoVa has only gone up over the last several years, currently around $217/KW/mo, with vacancy stable below 1%. While there has been some recent news around theDigital Gateway Project (QTS and Compass), this market is typically a well-oiled machine. This quarter, both DLR and EQIX spoke about large hyperscale-focused builds outside ofAtlanta, another top U.S. data center market growing quickly (~2GW today, ~3GW underconstruction). Atlanta pricing is $175/KW/mo today, with vacancy around 2%. We likethis strategy - market dynamics are favorable, both DLR and EQIX are savvy builders withstrong, experienced teams in the Major Metros, and relatively limited execution risk.Weremain positive on both DLR and EQIX, believing both companies will deliver ontheir build projections given their Major Metro market exposure. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We value DLR on a Price to Adjusted Funds from Operations (AFFO) per share multiple. Our $232 price target is based on 27xour 2027E AFFO per share of $8.60. We value EQIX on a Price to Adjusted Funds From Operations (AFFO) per share multiple. Our $1,222 price target is based on25x our 2027E AFFO per share of $48.63 DETAILS A DEEP DIVE INTO THE TOP U.S. MARKETS The three market archetypes are essentially based on how developers and customers trade off latency vs. power availabilityvs. cost—with urban markets optimizing for proximity, rural markets optimizing for scale and power, and secondary marketssitting in between as an equilibrium zone. We see a clear gradient across data center market tiers:Major Metrosare optimizedfor latency-sensitive inference and enterprise colocation, with the highest pricing ($120–150/kW hyperscale, $160–230colo), highest build costs ($12–15B per GW), tightest cap rates (≈4.5–6.5%), and the longest power delays (4–7 years), whileMinor Metrossit in the middle with slightly lower pricing, costs, and delays but higher yields; in contrast,Rural marketsareprimarily focused on large-scale AI training, offering the lowest costs ($7–9B per GW), fastest power availability (1–3 years),and cheapest hyperscale pricing ($70–120/kW), but also the highest cap rates (≈6.5–9%+) and minimal colocation presence—leading to a broader industry shift where growth and new capacity increasingly migrate from constrained urban hubs to lower-cost, power-rich rural regions. EXHIBIT 3:… while NoVA, Dallas/Ft. Worth, and the Chicago metro area have the most capacity today EXHIBIT 5:… as well as the greatest