Hyperscalers like AWS, Google, Microsoft, and Meta have spent roughly $930 billion on data center construction over six years. That figure exceeds the inflation-adjusted cost of the Interstate Highway System, which ran $620 billion over 37 years. Fin Moorhouse shared the comparison charts on Twitter, and they're genuinely startling. Paul Graham amplified the thread, sparking a debate about whether raw dollar amounts tell the real story.
When you account for GDP, the picture shifts. Government megaprojects consumed a far larger share of the economy when they were built, and Moorhouse posted GDP-adjusted data that makes today's spending look less dramatic. A $930 billion investment now means something different than $620 billion spread across the mid-20th century.
But what are we actually buying? Public infrastructure like highways gets amortized over 50+ years, while data centers face GPU replacement cycles of just 2-3 years. That creates intense pressure to generate returns fast. Commenters pointed out that railroads might be the better historical analog: privately funded infrastructure that reshaped society, built under similar commercial pressures.
The physical constraints are biting hard regardless. Large power transformers, essential for connecting data centers to the grid, have lead times of 2-4 years. Only a handful of global manufacturers like ABB and Siemens Energy produce them. Grid interconnection queues in major markets such as PJM have stretched to 5+ years. Grid strain has become a critical bottleneck for scaling infrastructure. The U.S. transformer fleet averages over 40 years old, and replacement needs compete directly with new data center demand. Regulatory bans are emerging as a direct response to these infrastructure gaps. Money can't solve this bottleneck. You can't build faster than the grid allows.