Compute (AI Capabilities)

Compute refers to the hardware resources required to train and run AI systems, including GPUs, TPUs, and specialized AI accelerators. The current generation of frontier AI models requires extraordinary amounts of computational power—training runs cost tens to hundreds of millions of dollars in compute alone. The significance of compute for AI governance stems from several unique properties: it is measurable (training runs can be quantified in FLOPs), concentrated (the global semiconductor supply chain depends on chokepoints like ASML, TSMC, and NVIDIA), and physical (unlike algorithms that can be copied infinitely, hardware must be manufactured and shipped).