1,489 new hyperscale facilities planned

Data centers are having an infrastructure moment — but this is really a power story. AI demand has turned hyperscale construction into a race for electricity, land, and permits, not just servers and buildings. That is the shift behind the “1,489 new facilities planned” line floating around the market. The important part is not the exact brag number. It’s that the industry has moved from “can we fill these buildings?” to “can we even energize them?” ### What’s a hyperscale facility, exactly? A hyperscale data center is the giant kind run or leased by cloud platforms like Amazon, Microsoft, Google, and Meta — the campuses built for massive computing loads, not a company server room in a back office. These sites keep getting bigger because AI training and inference need more power, denser racks, and more cooling than earlier cloud workloads did. That is why the sector’s growth now looks less like normal real estate expansion and more like utility-scale industrial buildout. (dcbyte.com) ### What changed in the last two years? ChatGPT’s launch in late 2022 became a forcing event. Since then, hyperscaler capex has surged, and operators have started locking up campuses years before delivery. DC Byte says the market has entered a phase defined by power, policy, and speed. JLL’s 2026 outlook lands in the same place — AI is now the main growth engine, and by 2030 it could account for half of all data-center workloads. (mckinsey.com) ### Why does everyone keep talking about gigawatts? Because megawatts are now the real currency. JLL expects nearly 100 GW of new global capacity between 2026 and 2030, roughly doubling the sector’s footprint. In North America alone, JLL tracks more than 35 GW under construction. More than 10 projects of 1 GW or larger are already being built — which would have sounded absurd just a few years ago. Basically, the industry is no longer adding buildings one by one. (srgresearch.com) It is assembling power portfolios. ### So is the bottleneck construction? Not mainly. The harder bottleneck is grid access. JLL says power — not location or even cost — has become the primary site-selection criterion because grid connections can take years. That flips the old model. Developers used to find the land, then line up utilities. Now they often chase power first and let geography follow. That is why frontier markets in Texas, Tennessee, Wisconsin, Ohio, and the U.S. (jll.com) Southeast are suddenly pulling in demand. ### Why are mature hubs slowing down? Because success clogged them. Northern Virginia, Frankfurt, and other core markets still have the customers, fiber, and ecosystem, but they also have the worst scarcity. DC Byte points to power shortages, planning friction, and competition for land. Vacancy in major hubs is below 1% in some cases, and preleasing can run 24 to 36 months before construction finishes. So hyperscalers are not abandoning big hubs — they are building around their constraints. (jll.com) ### What does this do to costs? It keeps them high and makes schedules shakier. JLL says more than half of projects faced delays in 2025, with equipment delivery and construction bottlenecks still hitting timelines. Higher rack densities and liquid-cooling requirements also make facilities more specialized and expensive. So the spending boom is real, but the catch is that every delay pushes capital costs, lease rates, and power costs higher. (dcbyte.com) ### Why does Texas keep showing up? Because Texas combines land, energy options, and a willingness to host giant campuses. JLL says Texas alone has 6.5 GW under construction and could overtake Virginia as the world’s largest data-center market by 2030. That does not mean every project gets built on time. It means Texas has become the clearest example of the new rule: the next winners are the places that can deliver power at scale, not just the places with the deepest internet history. (jll.com) ### What’s the bottom line? This is not a normal real-estate cycle. It is a utility-and-supply-chain race disguised as a construction boom. Demand looks durable. The question now is whether grids, permits, transformers, turbines, cooling gear, and labor can keep up. If they cannot, the next phase of AI will be limited less by chips than by the physical world needed to run them. (dcbyte.com) (jll.com)