Have a look at this chart.

It shows 69.1 GW of nominal data-center IT capacity scheduled for delivery from August of this year through 2028.

One of the key problems is likely to be powering all of it.

The U.S. had roughly 35–44 GW of nominal data-center IT capacity in 2024 and is projected to reach 56–132 GW by 2030. That points to a potential tripling, and the chart shows that trend unfolding.

There is skepticism that the buildout will be completed on time at all, since shipping a broken app update is much easier than finishing a building and procuring all the chips and racks needed to run it. If that does happen, the next bottleneck is even tighter: power.

Increasingly, states are requiring data centers to supply their own power, and companies are responding by trying to co-generate, usually with natural gas turbines. The issue is that manufacturing capacity for power equipment is limited, and backlogs are now very large. That includes a range of components, especially transformers, which are particularly backlogged. For developers, that means the timeline is no longer just about finding land, financing projects, or installing servers; it also depends on utility interconnections and equipment lead times that can stretch well beyond a normal planning cycle.

If that hurdle is cleared, the next potential problem is fuel for the turbines. If 75% of this is powered by natural gas, that would amount to around 8.6 Bcf/d. The good news is that there is currently a surplus of gas. The bad news is that this surplus is already expected to be absorbed by LNG exports, which are set to rise sharply as new facilities come online in the 2025-2029 window. According to the EIA, LNG liquefaction capacity is on track to add about 13.9 Bcf/d between 2025 and 2029.

The latest EIA data puts U.S. production at 116.0 Bcf/d in 2027, so this additional data center demand would consume about 5% of that total and exceed the entire EIA increase forecast for 2026 and 2027.

Taken together, the U.S. would need to add 10 Bcf/d of production in two years. That can be done, and gas is plentiful, but it would require a price signal, and front-month gas at $2.69 is not providing one. The practical constraint here is less about whether gas exists in the ground and more about whether production, pipeline capacity, and power infrastructure can all scale quickly enough to meet competing demand from LNG and data centers.

If hyperscalers actually carry out the capex, a floor for gas near $5 and a range of $5–8 depending on weather is conceivable. It is an interesting area to watch, but it is difficult to trade because there is significant skepticism that the data centers will be built on time.