Video documentary explores how AI datacenters are reshaping global infrastructure and economics.

The datacenter industry is undergoing a seismic shift driven by the rise of AI technology, transforming infrastructure fundamentally. Meta's abrupt halt of a datacenter construction project in Temple, Texas, exemplifies the rapid evolution in the industry's demands and priorities. Traditional datacenters, which focused on data storage and distribution, are being displaced by AI datacenters designed to function as supercomputers. This transition reflects a broader strategic reorientation across the industry, influenced by developments like ChatGPT, as companies reassess their infrastructure requirements to remain competitive in the AI era.

The architectural differences between traditional and AI datacenters are substantial. Traditional datacenters depend heavily on location optimization and extensive fiber networks to reduce latency and enhance connectivity to end-users. AI datacenters, conversely, operate more flexibly in terms of location, functioning as closed systems for model training that prioritize computational efficiency over latency and bandwidth. This fundamental design shift means that AI datacenters can tolerate longer inference latencies and are not location-sensitive, allowing companies to decouple their operations from geographic constraints tied to user proximity.

The power and cooling demands of AI datacenters far exceed conventional infrastructure capabilities. A single AI rack can require up to 132 kilowatts, compared to traditional setups that typically operate at only 3 to 10 kilowatts. To achieve this density, AI datacenters are transitioning to liquid cooling systems, which absorb energy more effectively than air cooling and reduce the physical footprint required for heatsinks, allowing more GPUs per server blade. Liquid cooling also extends silicon lifespan and improves overall energy efficiency by running components at lower temperatures. Meanwhile, AI datacenters demand access to critical IT power exceeding 200 megawatts—far surpassing the typical 30-megawatt capacity of traditional facilities—requiring extensive power infrastructure including numerous transformers and backup systems designed more for speed than resilience.

Looking forward, the energy implications are staggering. The race for Artificial General Intelligence is driving unprecedented demand, with projections of 40-50 gigawatts of additional power required by next year. Major technology companies are increasingly investing in energy generation infrastructure to support their AI datacenter operations, including partnerships with nuclear power plants such as the Susquehanna Steam Electric Station. As AI datacenters continue to proliferate and scale, their energy consumption is expected to rival that of entire nations, marking a transformative trend in both computing and global energy consumption patterns.