Or will more efficient algorithms just mean we run even more AI models, increasing the demand for AI chips even more?

I mean, god willing, but it'll be just as likely that we'll blissfully consume 100 million token contexts in that case.