U.S. Data Center Power Set to Nearly Double by 2027, According to Morgan Stanley

U.S. data center power capacity is projected to nearly double by 2027 due to the increasing demand for high-performance computing (HPC) to support Generative AI (GenAI), according to a recent note from Morgan Stanley.

The investment bank forecasts that U.S. data center power will rise from 40 gigawatts (GW) in 2024 to 79 GW by 2027, reflecting the escalating requirements of AI technologies.

Morgan Stanley’s analysis indicates that data center expansion in the U.S. and globally will accelerate in the coming years, primarily driven by the need for HPC resources for GenAI.

The bank’s estimates are grounded in a bottom-up model that assesses power demand based on expected quantities of GPUs and custom silicon.

On a global scale, power consumption for GenAI is expected to soar from 68 terawatt-hours (TWh) in 2024 to 378 TWh by 2027. This surge will significantly influence overall global data center power usage, which is projected to grow from 443 TWh in 2024 to 877 TWh by 2027.

Additionally, the growth of the data center market is reflected in the asset-backed securities (ABS) market. Morgan Stanley notes that data center ABS issuance is on track to reach a record $8 billion in 2024, increasing the total outstanding to $25 billion—a 37% year-over-year rise. If the current trend continues, the ABS market could reach $49 billion by 2027 as more operators tap into this funding option and the investor base expands.

The bank expects much of the data center growth to come from new constructions, which could cost between $30 million and $34 million per megawatt (MW) for complete setups, or $8 million to $12 million per MW when excluding GPUs and servers.

However, the timeline for bringing new data centers online is lengthening, particularly in states like California, where projects are now taking up to 125 months. This delay highlights the need for retrofitting existing data centers, such as those originally designed for bitcoin mining, to accommodate the immediate demand for AI and HPC data processing.