Companies including OpenAI, Meta, xAI, and Microsoft are investing up to one trillion US dollars per year in AI data center infrastructure worldwide — including builds in Europe and Germany. A new episode of the c't uplink podcast examines how these AI facilities differ fundamentally from traditional data centers.
A single AI rack can consume up to 160 kilowatts, compared to 5–15 kW for a standard server cabinet. The episode explores how Nvidia, AMD, Google, and Amazon build their accelerators, what chiplets and high-bandwidth memory deliver, and why data transport costs more energy than the actual computation.
Cooling is a central challenge: enormous waste heat output raises questions about environmental impact. Initial studies show measurable temperature increases in adjacent residential areas. The discussion also examines whether the planet has sufficient resources for the scale of infrastructure the industry envisions.
