Nearly Half of $9 Trillion Data Center Plans May Never Materialize
nashnova research
A Columbia Business School study finds roughly 45% of planned U.S. data-center capacity may never materialize — $9.3 trillion in investment that could go unrealized. The AI infrastructure boom already dwarfs every prior buildout cycle, but an arms-race dynamic is pushing real completion risk onto investors.
How big is this buildout?
The U.S. has roughly 509 gigawatts of new data-center capacity in the planning pipeline — nine times its current operating capacity.
Under the baseline scenario, about 300 GW actually gets built, driving $10.3 trillion in cumulative investment by 2032 — roughly 3.6% of U.S. GDP per year.
This means → even counting only the projects that *do* get finished, this wave is larger than the railroad, highway, and telecom buildouts combined.
Why might nearly half of these projects stall?
The model uses a time-decay logic: projects slated for 2026 have a 90% completion probability; each additional year knocks off 5 percentage points.
In plain terms = the further out the timeline, the more things can go wrong — construction costs rise, AI compute architectures shift, and the developer itself may not survive.
The researcher pointed to Oracle as an example: "If Oracle plans to finish a project in 2032, who knows if Oracle will still be around?"
Where does the $9.3 trillion figure come from?
The study benchmarks a single 200 MW AI campus at roughly $8.2 billion, then multiplies by the capacity unlikely to be completed — yielding about $9.3 trillion.
Factor in rising construction costs, and the actual unrealized figure climbs higher still.
This reflects a pipeline so massive that the *failed portion alone* is an astronomical number.
What does the "arms race" framing mean for investors?
The researcher calls the current dynamic an "arms race" — every company is racing to build capacity, but not all of them will survive.
This means → a large share of projects falling through does not mean the AI investment thesis is overstated — but individual projects and the companies behind them face real completion risk.
Community opposition, construction delays, and outright project abandonment are all live risks — which projects actually cross the finish line will be the key test of long-term corporate value.
What does Oracle's trouble illustrate?
Oracle's stock fell last Thursday after reports it was seeking to sidestep liability for potential delays at a planned New Mexico data center.
Oracle later issued a statement saying the project remains on track, but the market had already reacted.
In plain terms = the gap between a planning announcement and an actual completed facility is exactly where investors are most exposed.
市场有风险，内容仅供研究参考，不构成投资建议。