The surge of capital spending on AI infrastructure is massive, already absorbing about **5% of U.S. GDP**. That puts it on par with the fiber-optic and router buildout of the early 2000s (5.2%) and just below the housing-driven construction boom of 2005 (6.7%).
Today’s capital is being funneled into **self-improving, power-intensive systems** - AI models that learn, iterate, and compound knowledge, even as their hardware depreciates. See on the [[model layer and above]]
Past cycles suggest that investors often overspend in periods of technological euphoria. With AI, the likeliest outcome is similar: much of the direct investment may be overbuilt, while the **real value will accrue to more prudent players**—those who harness the infrastructure and capture derivative benefits rather than chasing the boom itself.
[[The AI Stack - Building Blocks]]
[[How AI Will Impact Energy Demand]]
[[Data Centre Energy Demand]]
[[The AI Timeline - Navigating the Road Ahead#Key Highlights]]
[[Themes shaping 2025#3. Modular Data Centers and Energy Infrastructure Battefield]]
[[model layer and above]]