Why is SoftBank building a 10GW AI datacenter?

SoftBank is planning a massive AI datacenter on a former U.S. nuclear weapons site, with plans described as a 10GW server farm and additional 10GW of “new generation,” alongside about a $4.2B grid upgrade. The practical driver is the scale of power demand implied by training and serving AI models.

The key point is that AI compute growth is colliding with electrical grid constraints. Large datacenter projects increasingly bundle generation capacity and grid upgrades to avoid delays and to secure reliable power for continuous workloads. A site with existing industrial infrastructure can also reduce development timelines, but the biggest message in the reporting is about energy: without enough electricity and grid throughput, AI scaling can stall.

This matters beyond SoftBank. Across the industry, competition for power has become a gating factor for AI infrastructure—forcing operators to pursue new power sources, negotiate with utilities, and redesign project timing around grid availability. Even when chips and server equipment are available, power procurement and interconnection bottlenecks can determine when capacity comes online.

If SoftBank’s plan proceeds as outlined, it would represent a shift toward datacenters that behave more like energy projects—integrating grid upgrades and additional generation so they can support sustained AI workloads.

For companies deciding where to deploy or expand AI services, the signal is clear: securing power at the tens-of-gigawatts scale is quickly becoming one of the most important (and hardest) parts of “AI scaling” itself.