US data center power to nearly double by 2027: Morgan Stanley by Investing.com

Investing.com — U.S. data center power capacity is expected to double by 2027, driven by growing demand for high-performance computing (HPC) to support generative AI (GenAI), according to a note of Morgan Stanley on Monday.

The investment bank’s research predicts that US data center power will grow from 40 gigawatts (GW) in 2024 to 79 GW by 2027, a surge fueled by the increasing demands of AI technologies.

“We expect data center growth in the US and globally to accelerate rapidly in the coming years, driven primarily by the growth of high-performance computing (HPC) data centers for GenAI,” they wrote.

Morgan Stanley’s estimates are based on a bottom-up model that calculates power demand using designed volumes of GPUs and custom silicon.

Globally, the bank predicts that energy consumption for GenAI will grow from 68 terawatt-hours (TWh) in 2024 to 378 TWh by 2027. This growth will have a significant impact on total data center energy consumption at the globally, which is expected to grow from 443 TWh in 2024. to 877 TWh by 2027.

They add that the growth of the data center market is also reflected in the asset-backed securities (ABS) market.

Morgan Stanley explains that data center ABS issuance is on track to hit a record $8 billion in 2024, leading to $25 billion, up 37% year-over-year. By 2027, they say the ABS market could reach $49 billion if current trends continue, driven by more operators accessing this funding source and a growing investor base.

Morgan Stanley anticipates that much of this data center expansion will involve new builds, which could cost $30 to $34 million per megawatt (MW) for full setups, or $8 to $12 million per MW, excluding GPUs and servers.

However, they note that the time it takes to bring new data centers online is getting longer, especially in regions like California, where projects now take up to 125 months. This delay is seen as underscoring the importance of upgrading existing data centers, such as those used for bitcoin mining, to meet the immediate demand for AI and HPC data processing.