The Zhitong Finance App learned that as the boom in artificial intelligence (AI) investment drives US data center construction plans to continue to expand, Wall Street is beginning to re-examine how many of these huge expansion plans can actually be implemented. Bernstein pointed out in the latest research report released on November 10 that the total scale of the North American data center projects to be built has reached 586 gigawatts (GW), but after a comprehensive assessment of construction progress, customer qualifications, developer strength, and regulatory risks, the bank believes that only about 199 gigawatts have a high landing reliability, accounting for about 34%.
This means that in the data center construction plan announced by the market, about two-thirds of the planned capacity cannot yet be considered a reliable future supply. However, Bernstein stressed that this does not mean that all remaining projects will be cancelled; rather, many projects are still in the early planning stage, and there is great uncertainty about construction time, financing capacity, and final delivery scale.
Bernstein pointed out that although changes in the financing environment for AI cloud computing companies in the past few months, political election results, and adjustments to investment plans of large technology companies have continuously affected market expectations, data center construction itself has a long cycle, usually two to three years, or even longer in regions where electricity supply is tight. Therefore, short-term market fluctuations will not immediately change the construction pattern of the entire industry.
Judging from the latest data, Bernstein's tracking project scale has increased from 411 gigawatts in July to 488 gigawatts in August, and further increased to 586 gigawatts in October. Over the same period, the risk-adjusted credible construction capacity was 136 GW, 170 GW, and 199 GW, respectively, with corresponding ratios of 33%, 35%, and 34%. This shows that although the scale of data center planning continues to expand, the proportion of projects that actually have high construction certainty has remained around one-third.
The most important factor contributing to this gap is the development stage the project is in. According to Bernstein's data, the current data center capacity of about 497 gigawatts is still in the announced planning stage, accounting for about 85% of all projects to be built, but the probable capacity of these projects after weighted probability is only 123 gigawatts, and the corresponding construction reliability is about 25%. In contrast, the scale of the project that has already started is about 87 gigawatts, the trusted capacity has reached 75 gigawatts, and the construction reliability is as high as 86%.
In other words, although the projects that have already started account for only a small portion of the total planned scale, they have contributed about 38% of the credible construction capacity. Bernstein believes that the actual construction progress is far more reflective of the possibility of the final implementation of the project than the expansion announcement issued by the company.
In addition to construction progress, the customer's credit quality is also an important factor in determining whether the project can be implemented. Data centers with long-term lease commitments from large technology companies are generally easier to obtain financing and advance construction; while some cryptocurrency mining enterprise transformation projects, emerging AI cloud service provider projects, and lack of clear tenant development plans face higher uncertainty.
In terms of developer types, data center projects built by large cloud computing companies themselves are the most reliable, reaching about 75%; the trustworthiness of professional data center hosting operators is about 62%. In contrast, the trustworthiness of emerging AI cloud service provider projects is only about 29%, projects transformed from cryptocurrency mining businesses are about 24%, and the average trustworthiness of other types of developers is even lower, only about 11%.
Bernstein pointed out that the four major technology companies, Amazon (AMZN.US), Google (GOOGG.US), Microsoft (MSFT.US), and Meta (META.US), together contributed about 78 gigawatts of trusted construction capacity, accounting for about 39% of all trusted projects. Looking at a single developer, Amazon leads with a trusted capacity of around 33 gigawatts, Google about 20 gigawatts, Meta about 15 gigawatts, and Microsoft about 10 gigawatts.
This also means that although more and more emerging companies are announcing large-scale AI data center investment plans, the new capacity that is actually expected to be delivered is still highly concentrated in the hands of large technology companies and professional operators with strong capital strength, clear customer needs, and mature construction experience.
Notably, data center projects are also facing increasingly stringent local regulations. Bernstein pointed out that several major US data center construction states are tightening restrictions on power grid access, energy cost sharing, and project approval. Among them, policy changes in Texas, Georgia, and Ohio are particularly noteworthy.
Texas is the largest data center planning market in the US, with about 150 gigawatts of capacity to be built, but Bernstein believes that only about 43 to 44 gigawatts are highly reliable. The state's grid access and approval restrictions may delay projects that have not yet started, but about 21 gigawatt projects have already entered the construction phase and are relatively less affected by the new restrictions.
Bernstein believes that tighter regulations may be more likely to delay projects or increase construction costs rather than large-scale cancellations. In particular, for projects that have received customer support, completed financing, and entered the construction phase, the overall prospects for implementation are still quite stable.
In addition, some developers are trying to reduce their dependence on public power grids by building their own power supplies or directly supporting them. According to Bernstein data, the planned capacity of such projects has increased from 185 gigawatts in August to 231 gigawatts in October, but the trusted capacity only increased from 42 gigawatts to 47 gigawatts, accounting for about 20%. This shows that although self-contained power solutions are sought after by the market, related projects still face great challenges in terms of funding, power construction, and actual implementation.
In terms of investment opportunities, Bernstein continues to be optimistic about data center developers with mature operating experience and diversified project reserves. The bank listed EQIX.US (EQIX.US) and Digital Real Estate Trust (DLR.US) as the preferred targets in the data center sector, and gave them “outperforming the market” ratings, with target prices of $1,270 and $226, respectively.
In the field of industrial and electrical equipment related to data center construction, Bernstein is also optimistic about Guangda Services (PWR.US), Eaton (ETN.US), Hubble Group (HUBB.US), Forgent (FPS.US), and Legence (LGN.US), believing that these companies are expected to benefit from long-term demand brought about by data center construction and power infrastructure investment.
Overall, Bernstein believes that the AI data center construction boom continues, but investors need to distinguish between the planned capacity announced by the company and the projects that are actually expected to be put into operation. As financing capacity, customer credit, and regulatory conditions gradually become key factors in determining project progress, enterprises with actual construction experience, stable funding sources, and a high-quality customer base will be more likely to stand out in this round of AI infrastructure expansion.