"A slow-starting race to build underground salt caverns could hamper the AI data center boom and weaken power delivery for the massive computing facilities that typically require 99.999% reliability. About half as much new gas storage is planned than will be needed in the future, industry sources estimate. Yes, you read that correctly, salt caverns. Manmade reservoirs thousands of feet below the surface are ideal storage structures for the volume of natural gas required to power AI data centers being built by hyperscalers"
"U.S. natural gas output is projected to spike another 15-25% from 2024 through 2030-and continue rising-because of a doubling of gas exports and a surge in domestic demand from the data center construction wave, ongoing electrification, and manufacturing onshoring. The lack of underground storage now threatens to become a bottleneck in the AI race against China. Without nearby gas storage facilities, customers are reliant on gas pipelines for their supplies."
""I don't want to be a bomb thrower or a Chicken Little, but it just seems like everybody in the data center world is in a great big giant hurry, and they haven't thought about all the things that can happen on the gas side," said Edmund Knolle, president of Gulf Coast Midstream Partners, which is developing a major salt cavern gas storage project near Houston slated to come online by the end of 2030."
Construction of underground salt cavern gas storage is lagging while AI data center development and gas exports drive U.S. natural gas demand sharply higher. Planned new storage is roughly half of projected need, leaving facilities dependent on pipelines that can fail from weather, landslides, or corrosion. New pipelines and power plants are being built even as gas-fired turbine shortages persist, but little new storage has been added in over a decade. The storage shortfall raises the risk of power interruptions for high-reliability data centers, increases utility cost volatility, and could become a strategic bottleneck.
Read at Fortune
Unable to calculate read time
Collection
[
|
...
]