That graph features a completely-made-up superexponential AI growth curve coincidentally reaching 100% of US electricity generation in 2030, but even if you take the prediction up to 2026 at face value, it also agrees that AI is currently a small fraction of overall demand. Of course any individual data center site might struggle with insufficient local supply, but China would have the same problem if they wanted to build bigger data centers. (Production capacity is actually north of 25 000 TWh, but poorly utilized because of a mismatch in place and time between producers and consumers.)