We believe, as I’ve said, that each industry and vertical should use tailored, moderately-sized, or “right-sized” models. In journalism, for example, you might need models for automated transcription and summarization. These models are small enough to run locally on your laptop, are not power-intensive, and don’t need to memorize every Hayao Miyazaki film. The point is that smaller, tailored models, especially open-source ones, are very energy-efficient and easy to fine-tune for your specific needs. Because of Taiwan’s dominant position in advanced chip manufacturing, we can enable this kind of Edge AI to run in an extremely energy-efficient manner. So, while we face energy challenges like any other country, I do not believe AI training will be our primary source of energy consumption, because we are not taking that brute-force, hyper-scaling path.