AI's Real Bottleneck Is Electricity, Not Chips — and It's Worse in Latin America

The rapid expansion of AI infrastructure is being constrained by a global electricity shortage, not a lack of chips or investment. OpenAI's Stargate project alone consumes 1.2 gigawatts — equivalent to powering 313,000 American homes — and grid connection wait times in the US have grown from under 20 months in 2005 to 55 months by 2023. Industry leaders including Jensen Huang, Mark Zuckerberg, and Sam Altman have each publicly acknowledged that energy availability, not computing power, will be the primary limit on AI growth. Researchers project AI data centers could require 100 gigawatts by 2030, roughly equal to the entire generation capacity of Spain or Australia. Latin America faces an even steeper challenge, with aging grids, frequent outages, heavy reliance on hydroelectric power, and bureaucratic delays that compound an already severe infrastructure deficit.