Table of Contents
Data center operators are confronting a growing generation of stranded assets with long-term financial and environmental consequences
As AI sets off an unprecedented outlay of capital, data centers face increasing pressure to minimize inefficiencies resulting from stranded assets.
According to early 2025 data, data center vacancy across North American has dropped to 1.9% with 83% of new capacity already pre-leased. Crusoe projects that AI alone will require $5.2 trillion in data center investment by 2030, $7.9 trillion in a more aggressive scenario.
While these numbers indicate a healthy and intensifying demand for infrastructure, experts worry that if demand for AI infrastructure contracts in the future, or a technological step change occurs, it will leave behind a trail of stranded assets and revenue sinkholes for operators.
Stranded data center assets have limited alternative use with inherently depressed liquidation value. The asset could be the data center facility itself — real estate, land, and all — the GPUs powering the servers that are bought on loan, and/or underutilized energy supplies.
The financial toll of stranded assets are gnarly. In Wisconsin, ratepayers owe over $1 billion on retired power plants. The sum is expected to climb even higher in 2027, as companies continue to upgrade their plants for higher rates of return.
The cost does not come just as short-term loss. Payment for stranded assets continues long after the assets stop generating revenues. For example, We Energies, a local investor-owned utility company in Wisconsin, is on the hook to pay for a shuttered plant on the books for 21 years after it was shut down.
The problem does not just originate from overcommitting to infrastructure. Often aging systems in data centers guzzle more power, adding to the energy footprint of the facilities. The shift to inference is another major driver for the growing number of stranded assets in data centers. Today, inference accounts for approximately one half of AI compute. That is expected to rise to two-thirds by the end of the year, according to Deloitte’s estimates. These workloads demand their own infrastructure, and that may not just be repurposed older systems.
“A campus built for 2023’s training workload may struggle to serve 2027’s, let alone retrofit cleanly into inference,” Narula wrote.
The solution is to repurpose assets to recoup the losses. However, that’s easier said than done. Repurposing purpose-built facilities like AI data centers that are outfitted with custom cooling, high-density power configs, and designs to accommodate rows of GPU-loaded servers into general-purpose computing facilities is capital-intensive, if possible at all.
“Some campuses will adapt. Some will be repriced for lower-margin workloads. A meaningful number will be stranded between two workloads they can’t quite fully serve,” wrote Rahul Narula, research fellow at The Strange Review, in an article.
Besides, such an endeavor can sink significant portions of the original costs as the hardware will lose value prematurely, not to mention, require additional costs of replacement.
How about discarding old assets to cut losses? Disposal of obsolete assets can generate electronic trash or e-waste, increasing risks of environmental contamination as a majority of them — circuit boards, lithium-ion batteries, copper cables — all end up in landfills.
Experts recommend a checklist of best practices to unlock adaptive reuse of assets. For repurposing existing facilities, confirming access to adequate megawatt capacity and future-proofing existing spaces can in some cases prove to be a sustainable approach.
There is also the concept of circular energy, an endeavor that is close to Crusoe Energy’s own mission. Reusing surplus energy instead of releasing it in the atmosphere as carbon dioxide, leads to more sustainable data centers. The company has developed an approach called Digital Renewable Optimization (DRO) to capture stranded or wasted energy from renewable sources like wind and solar which often happens as aging transmission systems fail to carry the entire load that grids produce. Crusoe colocates data centers at renewable power substations to utilize this overflow.
There are some other great examples of how companies are harvesting and reusing stranded energy in data centers today. For example, in Rjukan, Norway, a seafood company is using renewable energy from data centers to power a trout farm. In the U.K., surplus energy from a small local data center is being redirected to heat its swimming pool. Residential apartments in Sweden are receiving indoor heating from a data center. These are but a few examples.
Although it is unclear what incentives might best encourage sustainable practices, there are a few obvious ones at this point. Stranded assets — whether that’s physical assets or energy — tend to be inefficient and quickly turn into cost centers if not handled properly. They impact investor sentiments and scale back efforts to reach net-zero emissions and earn green credentials. All these make mitigation just as urgent as the infrastructure build itself.
