 Variational quantum algorithms, VQAs, are a leading candidate for useful applications of near-term quantum computing, but limitations due to unavoidable noise have not been clearly characterized. Here, the authors prove that local Pauli noise can cause vanishing gradients rendering VQAs untrainable. This article was authored by Sampson Wang, Enrico Fontana, Mcerizo, and others.