Even if we assume that, it's not necessarily a problem or suggests that AI progress will slow anytime soon. We can afford to dedicate a lot more energy to AI improvement than we currently are. Recent multimodal models seem to suggest there is plenty of room for efficiency gains yet. We are still far from limitations of energy becoming a primary concern, if it ever does, as AI self-improvement will make its own algorithms more efficient and get better and better at finding outside resources to exploit.