Periods of technological revolution are usually associated with overvaluation of and over-investment by innovating firms. This paper develops a model that explains this behavior in a frictionless rational setting. When fully rational innovating firms face uncertainty about the returns to scale of their production functions, over-investment emerges as the optimal way to learn about the returns to scale. The optimal learning is also shown to produce overvaluation. The model is also able to generate what an observer ex-post would identify as bubbles followed by over-correction, negative excess returns in early periods, negative auto-correlation in excess returns and market-to-book and size effects.