Morgan Stanley AI Energy Summit warns of "power crisis": High-risk period in 2027-2028 Turbine manufacturers and computing power service providers may benefit

On December 4, 2025, Morgan Stanley hosted the second "AI Energy Summit" in New York City, with core conclusions focusing on electricity shortages, iterative solutions, investment directions, and providing key references for industry development and capital layout. The summit focused on the energy support challenges and opportunities for AI development in North America, with 400 attendees, nearly 4 times the amount from the previous year, reflecting an explosive increase in attention to AI infrastructure. Key points included the possibility that American data center developers may face serious AI electricity shortages, the rise of off-grid solutions, more favorable power + data center transaction terms, political risks, labor and equipment shortages; the value of "time-to-power" solutions, and the expected growth in natural gas demand. The primary finding of the summit was that American data center developers will face severe electricity shortages. Morgan Stanley predicts that in the coming years, the electricity gap for American data center developers may reach 10-20%, with risks gradually becoming apparent starting in 2026, and entering a high-risk period in 2027-2028. This gap stems from the non-linear growth in AI computational power demand - current data center sizes have surged from 100-500MW 1.5 years ago to 1GW (with a park in Louisiana reaching 4GW), while traditional grid access capacity and power supply growth rates lag significantly behind. This, coupled with the political controversy caused by high electricity and water consumption at data centers, further exacerbates the pressure on power supply. To address the power shortage, off-grid solutions have become an industry consensus. As grid projects face controversies over rising electricity prices, complex approvals, and higher security deposits for grid access, data center developers continue to show a preference for off-grid solutions. Mainstream technology combinations include natural gas turbines, reciprocating engines, fuel cells with battery storage, which not only avoid grid controversies but also gradually transition from transitional power supply to permanent solutions. While improvement in power and data center development trade terms has been made, execution risks remain significant. Given excessive demand for such products and a worsening imbalance between supply and demand, power and data center developers are expected to obtain better terms (such as project risk allocation). This progress is positive as project execution faces many risks, including supply chain and labor, among other aspects. The bank believes that 2026 will be the "year of execution," and the performance of stocks related to "time to power" solutions will increasingly depend on the effectiveness of power and data center projects. In terms of investment direction, Morgan Stanley is bullish on the significant returns brought by the application of artificial intelligence and expects the scale of these returns to rapidly expand with the continued non-linear iteration of large language models (LLMs) by 2026. Many data show that even with sufficient deployment of data center infrastructure (which is highly challenging in the US), computational power demand will still surpass supply. The bank proposes the core strategy of "laying out AI deployment constraint areas," which specifically consists of three categories: 1) "time to power" solutions, including utility companies that quickly provide grid access, companies related to the transformation of Bitcoin mines into HPC data centers, turbine manufacturers like GEVernova and Siemens Energy, fuel cell companies like Bloom Energy, and natural gas industry service providers like EQT Corporation and Williams Companies; 2) "computational power service providers," including chip ecosystem companies like NVIDIA Corporation, data center REITs, and emerging cloud service providers like CoreWeave; 3) "bottleneck easing" areas, including engineering construction, transformers, and power transmission equipment. Bruker Corporation Field shared its views on AI power supply, planning to provide a complete integrated solution for data centers (including land, power, and chips), with opportunities centered in the North American and European markets. The company leverages its global real estate assets to gain a competitive advantage in this business. Electricity remains the core bottleneck in data center construction, despite constraints in labor, land, and other areas: when customers bring their own GPUs, electricity costs account for only 5% of total data center costs (Bruker Corporation Field believes that existing GPUs carry significant residual value). The real constraint is not from gas turbine original equipment manufacturers (OEMs) themselves, but from the encapsulation and integration required across the full value chain - turbine manufacturers rely on intermediaries to assemble complete solutions and deliver final power solution packages to customers. Williams (WMB.US) shared its views on indoor power solutions, with long-term contract terms of 10-15 years (Enterprise Value/EBITDA multiple during the construction phase is 5 times). Renewal depends on customer demand, but customers are expected to continue with the colocated power mode. The company believes that grid access is beneficial for all parties, where turbines can provide incremental megawatt-level power capacity as needed; alternatively, these turbines can also be relocated for use in their core business (compression stations). Ohio (the location of the company's first indoor power project, "Socrates") is a favorable region for indoor power generation: local legislation provides support (with a clear and stable approval process), enabling rapid construction and commissioning. Galaxy Digital (GLXY.US) shared its views on converting cryptocurrency facilities to data centers, and the company believes that concerns over CoreWeave's credit quality have been exaggerated by the pre-listing lender evaluations of CoreWeave. Considering that CoreWeave has reduced its reliance on Microsoft Corporation/OpenAI, diversified its business, its credit condition may have improved since then. Galaxy remains optimistic about the recent addition of power approvals for the Helios project. Galaxy has reserved a safety margin in its construction timeline/budget, which results in its capital expenditure per megawatt being slightly higher than other transactions already announced for converting cryptocurrency facilities into high-performance computing facilities.