How much of the AI data center expansion blueprint can actually be realized? Bernstein: Only one-third of the nearly 600 GW of planned capacity is credible.

As the AI investment boom continues to inflate U.S. data center construction plans, Wall Street is beginning to reassess how much of these massive expansion blueprints can actually be realized. In a research report released on November 10, Bernstein noted that the total scale of North American data center projects under development it tracks has reached 586 gigawatts (GW), but after a comprehensive assessment of construction progress, customer qualifications, developer strength, and regulatory risks, the firm believes only about 199 GW has relatively high credibility of being realized, accounting for approximately 34%. This means that about two-thirds of the planned capacity in publicly announced data center construction plans cannot yet be regarded as reliable future supply. However, Bernstein emphasized that this does not mean the remaining projects will all be canceled, but rather that many projects remain in early planning stages, with significant uncertainty regarding construction timelines, financing capacity, and final delivery scale. Bernstein noted that although changes in the financing environment for AI cloud computing companies, political election outcomes, and adjustments to investment plans by large technology companies over the past few months have continued to affect market expectations, data center construction itself has a long cycle, typically requiring two to three years, and even longer in regions with tight power supply. Therefore, short-term market fluctuations will not immediately change the construction landscape of the entire industry. Based on the latest data, the scale of projects tracked by Bernstein has increased from 411 GW in July to 488 GW in August, and further to 586 GW in October. During the same period, the risk-adjusted credible construction capacity was 136 GW, 170 GW, and 199 GW, respectively, corresponding to ratios of 33%, 35%, and 34%. This indicates that although the scale of data center planning continues to expand, the proportion of projects with genuinely high construction certainty has consistently remained at around one-third. The most important factor contributing to this gap is the development stage of the projects. Bernstein data shows that approximately 497 GW of data center capacity is still in the announced planning stage, accounting for about 85% of all projects under development, but the probability-weighted credible capacity of this portion is only 123 GW, corresponding to a construction credibility of about 25%. By comparison, projects already under construction total approximately 87 GW, with credible capacity reaching 75 GW and construction credibility as high as 86%. In other words, although projects already under construction account for only a small portion of total planned scale, they contribute about 38% of credible construction capacity. Bernstein believes that actual construction progress reflects the likelihood of projects ultimately being realized far better than corporate expansion announcements. In addition to construction progress, the credit quality of customers is also an important factor determining whether projects can be delivered. Data centers with long-term lease commitments from large technology companies are typically more likely to obtain financing and advance construction, while some projects converted from cryptocurrency mining operations, projects by emerging AI cloud service providers, and development plans lacking clear tenants face higher uncertainty. From the perspective of developer types, data center projects built by large cloud computing companies themselves have the highest credibility, at approximately 75%; professional data center colocation operators have credibility of approximately 62%. By comparison, projects by emerging AI cloud service providers have credibility of only about 29%, projects converted from cryptocurrency mining operations about 24%, and other types of developers have even lower average credibility of only about 11%. Bernstein noted that four large technology companiesAmazon.com, Inc. (AMZN.US), Alphabet Inc. Class C (GOOG.US, GOOGL.US), Microsoft Corporation (MSFT.US), and Meta (META.US)together contribute approximately 78 GW of credible construction capacity, accounting for about 39% of all credible projects. From the perspective of individual developers, Amazon.com, Inc. leads with approximately 33 GW of credible capacity, Alphabet Inc. Class C has approximately 20 GW, Meta approximately 15 GW, and Microsoft Corporation approximately 10 GW. This also means that although an increasing number of emerging companies are announcing massive AI data center investment plans, the incremental capacity truly expected to be delivered remains highly concentrated in the hands of large technology companies and professional operators with strong financial resources, clear customer demand, and mature construction experience. It is worth noting that data center projects also face increasingly strict local regulation. Bernstein pointed out that multiple major U.S. states for data center construction are tightening restrictions on grid interconnection, energy cost allocation, and project approval, with policy changes in Texas, Georgia, and Ohio particularly noteworthy. Texas is the largest data center planning market in the U.S., with approximately 150 GW of capacity under development, but Bernstein believes only about 43 to 44 GW of that has relatively high credibility. The state's grid interconnection and approval restrictions may delay projects that have not yet broken ground, though approximately 21 GW of projects have already entered the construction phase and are relatively less affected by the new restrictions. Bernstein believes that tightening regulation is more likely to lead to project delays or higher construction costs rather than large-scale cancellations. Especially for projects that have already secured customer support, completed financing, and entered the construction phase, the overall prospects for realization remain relatively solid. Additionally, some developers are attempting to reduce dependence on the public power grid by building their own power sources or directly pairing with dedicated power supply. Bernstein data shows that the planned capacity of such projects has increased from 185 GW in August to 231 GW in October, but credible capacity has only increased from 42 GW to 47 GW, accounting for approximately 20%. This indicates that although self-supplied power solutions are popular in the market, the related projects still face significant challenges in terms of funding, power infrastructure construction, and actual execution. In terms of investment opportunities, Bernstein continues to favor data center developers with mature operational experience and diversified project pipelines. The firm lists Equinix, Inc. (EQIX.US) and Digital Realty Trust, Inc. (DLR.US) as top picks in the data center sector, both rated "Outperform," with price targets of US$1,270 and US$226, respectively. In the industrial and electrical equipment sectors related to data center construction, Bernstein is also bullish on Quanta Services (PWR.US), Eaton Corp. Plc (ETN.US), Hubbell Incorporated (HUBB.US), Forgent (FPS.US), and Legence (LGN.US), believing these companies are poised to benefit from long-term demand driven by data center construction and power infrastructure investment. Overall, Bernstein believes that the AI data center construction boom is still ongoing, but investors need to distinguish between announced planned capacity and projects that are truly expected to become operational. As financing capacity, customer credit, and regulatory conditions gradually become key factors determining project progress, companies with actual construction experience, stable funding sources, and high-quality customer bases will be more likely to stand out in this round of AI infrastructure expansion.