AI-driven data center expansion is running into supply constraints across critical metals, semiconductor memory, and grid interconnection, according to multiple recent analyses.
The World Economic Forum projects that by 2035, data centers could account for 6.0% of global gallium use, 2.4% of germanium use, and 1.2% of tin use, according to Finimize. Gallium and germanium are essential to advanced chip manufacturing, and rising data center demand threatens to tighten already concentrated supply chains.
Grid access poses a separate chokepoint. US waits for grid connections averaged four years during 2018-2023, according to Finimize. That timeline complicates rapid buildout of power-hungry facilities.
High-bandwidth memory, a component critical to AI accelerator chips, faces its own squeeze. The IEA expects the high-bandwidth memory shortage to continue through at least the end of 2027, per its analysis of energy and AI, as reported by Medium. The constraint limits how quickly chip suppliers can deliver the hardware data centers need to train and run large AI models.