Moore's Law promised efficiency, but AI demands scale. Mega-data centers from Microsoft, Google, Meta, and Amazon now consume electricity like countries. Training frontier models costs tens of millions and requires massive power, flipping the script on shrinking infrastructure.

The AI market has surged from $25 billion to over $200 billion, with data centers projected to use 10% of global power by 2030. This build-out strains energy grids and water supplies, facing physical limits on chip shrinkage and community pushback.

History warns of a repeat: the 1990s fiber bubble left billions in unused cables. With a few giants dominating, the risk is a tech bubble where underused facilities and sunk costs mirror past failures, questioning the sustainability of exponential growth.