AI Is Consuming More Power Than Countries

Training GPT-4 consumed roughly the same electricity as 1,000 US homes use in a year. GPT-6 is estimated to have consumed multiples of that. And AI inference — running queries at scale, billions of times daily — dwarfs training in total energy demand.

Over 100 gigawatts of AI data center capacity is currently planned or under construction globally. For context: that is roughly 10% of total US electricity generation capacity.

AI has become one of the fastest-growing energy consumers in human history. And the industry is mostly silent about it.

Data Centers Are Now Reshaping National Energy Strategies

Saudi Arabia, the UAE, and the US are all accelerating power grid investments specifically to attract AI infrastructure. The GCC is positioning itself as a global AI data center hub, leveraging cheap land, political stability, and significant energy capacity. AI infrastructure location decisions are now geopolitical decisions.

The Compute War Is Also an Energy War

The lab that controls the most compute wins the model race. But compute requires power, and power supply is physically constrained. OpenAI, Microsoft, and Google are all signing decade-long power purchase agreements with nuclear, solar, and gas providers. Controlling energy contracts is now a frontier AI strategy.

The Carbon Cost Is the Uncomfortable Truth the Industry Avoids

AI labs publish benchmark results, safety reports, and funding announcements. Very few publish energy consumption data. The gap between AI's sustainability claims and its actual energy footprint is significant, and growing. This will become a regulatory issue within 24 months.

Smaller Models Are the Efficiency Answer — But Not Yet the Scale Answer

Research into model compression and efficient architectures is advancing rapidly. Some models now achieve 80% of GPT-6 performance at 10% of the compute cost. But frontier capability still requires frontier compute. Efficiency gains are real. They are not yet fast enough to offset the overall demand surge.

The Regions That Solve the Energy Problem Will Win the AI Infrastructure Race

The next decade of AI leadership will not be determined only by model quality. It will be determined by who can power the compute at scale, reliably, at acceptable cost. This makes energy policy and AI policy inseparable.

Bottom line: AI is not just a software story. It is an infrastructure and energy story. Every company building on AI is implicitly dependent on a global power supply chain. The leaders who understand this are already planning for it. Most are not.

AI Series continues.