The Next AI Infrastructure War: Power Shifts from Chip Nodes to Grid Capacity

From an industry cycle perspective, the bottleneck for AI computing power is undergoing a structural shift. When the single-card power consumption of NVIDIA's B200 GPU breaks through 1000W, and OpenAI's Stargate supercomputing cluster requires 5GW of electricity, a clear signal has emerged: the physical boundary for expanding computing power is no longer Moore's Law, but the grid's capacity. TechCrunch Disrupt 2026 specifically set up a Smart Systems Stage agenda to discuss the power challenges of AI infrastructure, which itself confirms an industry inflection point.