Trillions locked into data centers that may never pay off, empty racks could become the next office tower ruins

Technology is moving faster than the buildings meant to contain it. Data centers take two to three years to construct, but demand can flip in half that time. JLL warns that without a precise read on future workloads and target customers, new facilities risk opening already obsolete https://www.jll.com/en-us/insights/why-data-centers-could-hit-obsolescence-sooner-than-you-think.

Power densities are exploding. Rack loads are forecast to rise from 36kW today to 50kW by 2027, and AI training clusters could draw 80 to 100kW per rack. That level of heat forces cooling systems that most sites are not designed to handle https://www.jll.com/en-us/insights/why-data-centers-could-hit-obsolescence-sooner-than-you-think.

By 2030, nearly $7 trillion will have been poured into data center infrastructure, with more than $4 trillion going directly into computing hardware alone https://www.mckinsey.com/industries/public-sector/our-insights/the-data-center-balance-how-us-states-can-navigate-the-opportunities-and-challenges. That is fixed capital chasing a market that mutates every six months.

Electricity costs are already rising. In PJM’s 2024 capacity auction, data centers accounted for over 60% of the price increase, representing $9.3 billion in extra grid charges now being passed on to customers https://www.axios.com/2025/08/04/electricity-costs-bills-data-centers-ai. Consumers never voted for this. They never had the choice. They just pay the bill.

If AI demand slows or collapses, these hyperscale campuses become stranded assets. They cannot be repurposed easily, and retrofitting for liquid cooling or alternative workloads is expensive. The result is empty shells, oversized and underused, while households are left paying for both the initial buildout and the fallout.

The market is not modeling this risk. The public is not being warned. The cycle looks like every other boom before the bust: malls, office towers, factories. The difference is scale. This is trillions in steel and silicon built for workloads that may not last as long as the concrete around them.