AI infrastructure is priced on signed capacity. What limits it sits further up the chain, from public consent to grid connection to energised power, and the gap now shows in public filings.
For most of the cloud era, building a data center started with land. The order has flipped: power now comes first, and land is chosen by what energy can reach it.
SPAN's XFRA program, with NVIDIA as launch partner, is one of the clearest signals yet that AI infrastructure will not be built exclusively inside massive centralized campuses.
Hyperscale datacenter campuses are land grabs with compute attached. The next era of compute will be smaller, faster, more distributed, and more honest about power.
The future of AI is not limited by algorithms or data. It is constrained by power. Compute is no longer just a software problem. It is an energy problem.
AI is not the bubble. Poorly aligned infrastructure is. Why speculative data center construction breaks down, and why energy-aligned compute wins.
Data center developers are bolting aircraft engines onto server farms to keep up with AI. That is not innovation. It is a signal the centralized model is broken.