A technical paper published by NVIDIA last year may contain a unit error in its copper demand estimate, leading to an overstatement of the copper shortfall. In its paper introducing its own 800VDC high-voltage direct current architecture, NVIDIA pointed out that in a traditional architecture, a single 1-megawatt rack requires copper busbars weighing 200 kilograms, while a single 1-gigawatt rack requires up to 500,000 metric tons of copper.
McKinsey forecasts that global data center capacity demand will reach 219 gigawatts by 2030, which would imply that construction in this sector would require over 100 million metric tons of copper. This is a staggering figure compared to the global copper reserves, which stood at just 9.8 billion metric tons at the end of last year.
However, a recent investigation suggests that NVIDIA's report may have contained a very basic error, resulting in a significant overestimation of the copper shortfall. Returning to NVIDIA's report, if a 1-megawatt rack requires 200 kg of copper, and 1 gigawatt is equivalent to 1,000 megawatts, a simple conversion shows that a 1-gigawatt rack should require 200 metric tons of copper busbars, not the 500,000 metric tons listed.
Consulting firm Thunder Said Energy believes that the figure of 500,000 metric tons of copper was actually intended to be 500,000 pounds of copper, which equates to approximately 226 metric tons. Recalculating based on this correction, if total data center demand reaches 219 gigawatts, the copper required for power transmission and distribution would be about 44,000 metric tons, representing only 0.15% of annual copper production.
In other words, the artificial intelligence data center sector is not as copper-intensive as previously imagined.
Comments