Morgan Stanley: AI financing and energy financing are converging, the era of "electricity + computing power" is coming.

With Microsoft Corporation (MSFT.US), Alphabet Inc. Class C (GOOG.US), Meta (META.US), Amazon.com, Inc. (AMZN.US), and Oracle Corporation constantly increasing their capital expenditure budgets, the scale of artificial intelligence (AI) infrastructure construction is reaching unprecedented levels. The market's focus is mainly on data centers, graphics processing unit (GPU) procurement, and massive financing needs. However, a recent research report released by Morgan Stanley points out that as electricity becomes a key bottleneck for AI expansion, the boundary between AI financing and energy financing is rapidly blurring. From data center financing to "power + computing power" financing A few weeks ago, Morgan Stanley discussed AI infrastructure financing issues, analyzing how tech companies can provide funding for large capital expenditures through stocks, bonds, and other channels. However, after further discussions with clients and industry chain companies, the bank is more convinced that the financing issue cannot be understood solely from within the tech industry. The reason is that the bottleneck of AI infrastructure construction is extending upstream. The data center itself is just the final carrier of computing power, and the electricity, power transmission networks, transformers, power generation equipment, engineering labor, and water resources that support the operation of data centers are becoming equally important constraints. In many regions, data centers are no longer marginal increments in the electricity grid load but one of the largest and fastest-growing sources of new electricity demand. However, the expansion cycle of power generation capacity, power transmission networks, and the supply chain of critical power equipment is much longer than data center construction. This means that even if tech companies have sufficient funds to build data centers, they may not be able to obtain sufficient electricity access in a timely manner. Therefore, AI infrastructure financing is shifting from simple data center financing to comprehensive "power + computing power" financing. Electricity bottlenecks are changing the logic of data center construction Several sets of data listed in the report reflect the severity of the constraints on power infrastructure. According to IndustrialSage data, the average delivery time of power transformers has reached 128 weeks, and the delivery time of power generators and transformers has reached 144 weeks, compared to the pre-COVID-19 normal delivery time of 12 to 16 weeks. In other words, the waiting time for some critical power equipment has extended from a few months to over two years. At the same time, the backlog of integrating the U.S. power grid is intensifying. According to data from the Berkeley Lab, as of early 2025, the scale of energy projects awaiting integration into the U.S. grid has exceeded twice the installed generating capacity in the U.S. Morgan Stanley believes that this backlog situation may be even more severe now. These bottlenecks are changing the behavior of data center developers. In the past, data center siting may have considered factors such as land costs, network connections, tax incentives, customer proximity, and operating costs; now, the availability of power is becoming one of the most important prerequisites. Developers have begun to prioritize locations where they can obtain power access more quickly and explore integrated solutions that combine power generation facilities with data centers more closely. AI financing and energy financing are converging With the continued strengthening of power constraints, the once clear boundary between AI infrastructure and energy infrastructure is rapidly disappearing. In the past, energy projects were mainly invested and built by utilities or independent power producers, while tech companies focused on servers, chips, and data centers. Today, this division of labor is being broken. In order to secure stable power supply, more and more AI companies are actively participating in energy asset investments, including signing long-term power purchase agreements, directly investing in power generation projects, and even acquiring related energy assets. The report specifically mentions two types of solutions: The first type is "off-grid power" solutions, including fuel cells, gas turbines, energy storage systems, etc. These solutions can bypass some grid restrictions and provide data centers with a more direct and flexible source of power. The second type is the "time-to-power" strategy, which means shortening the time to obtain power. For example, repurposing bitcoin mining facilities. Since mining facilities often already have significant power access capacity, they may be converted into AI data centers or related computing facilities. In Morgan Stanley's view, the financing needs of the AI and energy industries have begun to converge, and in the future, the two are likely to form a more closely integrated, interdependent capital ecosystem. Labor, water resources, and policy also present challenges In addition to electricity, Morgan Stanley also notes that AI infrastructure construction faces broader structural constraints. First is labor. The report points out that the U.S. is expected to face a shortage of about 300,000 electricians in the next decade. At the same time, over one-fifth of the existing electrician workforce is aged 55 or older, nearing retirement. Second is water resources. S&P analysis shows that 43% of data centers globally are located in areas with high water resource pressure. As some data center cooling systems consume water resources, expanding further in water-stressed areas in the future may face sustainability and regulatory pressures. Third is policy and local resistance. The report states that opposition to new data center projects by both parties in the U.S. is on the rise. New York state has proposed legislation to suspend new data center projects; Texas Governor Greg Abbott recently issued an order requiring regulators to ensure that the addition of new data centers does not raise costs for other users; and more data center projects are being rejected at the local level. More broadly, legislative bodies in 14 states in the U.S. are considering some form of measures to suspend data center construction. Shortage of computing power may strengthen the pricing power of "compute merchants" With multiple constraints, Morgan Stanley believes that the risk of structural imbalance between supply and demand of computing power is increasing. If AI demand continues to grow rapidly, and constraints such as electricity, equipment, labor, water resources, and policy approvals limit the expansion of computing power supply, then "scarcity" will become an important feature of the future computing power market. Enterprises with large-scale, reliable computing power resources will gain stronger pricing power. The report refers to these enterprises as "merchants of compute," which can be understood as "compute suppliers." Their core advantage is not only in having data centers or cloud services but also in controlling scarce, stable, deliverable computing capacity. Similar to power companies in the energy industry that control critical resources, "compute suppliers" may have stronger price negotiation capabilities in the industry chain. This article is reproduced from "Cai Lianshe", GMTEight editor: Jiang Yuanhua.