Matthew Smith warns that the US faces a historic natural gas deficit by 2030, driven by LNG exports and AI data center demand, potentially exhausting working gas storage and causing unbounded price spikes. His bottom-up modeling shows that production and infrastructure cannot keep up, and the only sustainable long-term solution is large-scale nuclear power, which won't be ready until 2033-2034.
Summarized by Podsumo
*LNG exports are the primary driver*: US export capacity is set to rise from 15 Bcf/day to 35 Bcf/day by 2030, consuming an ever-larger share of domestic production.
*AI data centers add ~5 Bcf/day demand* (in the base case), but could double to 10-15 Bcf/day if all proposed projects are built, worsening the deficit.
*Production can't keep up*: The US can add only ~20 Bcf/day of new gas supply, while LNG alone requires 20 Bcf/day — leaving no room for AI or other demand growth.
*Infrastructure bottlenecks*: Processing, gathering, and interstate pipelines are all at or near capacity, and building new ones takes 3-5 years.
*The only long-term solution is large-scale nuclear*: AP1000 reactors are the only scalable option, but they won't come online until 2033-2034 at the earliest.
"The upside risk to prices becomes unbounded and convex."
— Matthew Smith
"Once you start to draw down working gas storage, the system becomes extremely tight and prices can spike to $10, $20, or more."
— Matthew Smith
"The die was cast long before AI compute came to the scene. LNG exports were already the primary demand driver."
— Matthew Smith