The Most Important Land Grab of This Century?
The Demand For Intelligence Is Insatiable Part 3
The short version
Driven by the compute demands of frontier model scaling, hyperscalers and capital pools are buying up millions of acres. What looks like a standard real estate cycle on the surface is a preemptive lockup of the physical bottlenecks of the digital age.
For decades, datacentre site selection prioritised proximity to fibre trunks, low tax jurisdictions, and low latency to major metro centres. AI compute changed that equation overnight. Modern training and inference runs require mega-campuses spanning hundreds of acres, capable of supporting 100 megawatts to multiple gigawatts of continuous draw.
So the land is no longer priced as industrial real estate. It is priced as a synthetic utility asset. The parcel is merely the vehicle; the value sits in the energy deposits, the behind-the-meter generation potential, the water rights and the substations anchored to it.
Traditional cloud hubs were built on 10 to 30 megawatt incremental builds. Frontier campuses demand single sites of 500 megawatts to 2 gigawatts, which pushes developers into rural greenfield land near nuclear facilities, gas basins and dedicated high-voltage lines. And the buying runs five to ten years ahead of any construction, to lock competitors out of the pipeline.
Continue reading Read the full article on Substack →Covers the three power-sourcing strategies, the four constraints on the buildout, what each hyperscaler is actually doing, and why Peter Thiel bought into an Argentinian shale basin.
Part One — AI Went to Software First Because Code Has a Judge. Part Two — Demand for Which Tier? Cheap Tokens and Frontier Tokens Just Moved Opposite Ways.