Quantum simulation is expected to be one of the key applications of future quantum computers. Product formulas, or Trotterization, are the oldest and, still today, an appealing method for quantum simulation. For an accurate product formula approximation in the spectral norm, the state-of-the-art gate complexity depends on the number of Hamiltonian terms and a certain 1-norm of its local terms. This work studies the concentration aspects of Trotter error: we prove that, typically, the Trotter error exhibits 2-norm (i.e., incoherent) scaling; the current estimate with 1-norm (i.e., coherent) scaling is for the worst cases. For k-local Hamiltonians and higher-order product formulas, we obtain gate count estimates for input states drawn from a 1-design ensemble (e.g., computational basis states). Our gate count depends on the number of Hamiltonian terms but replaces the 1-norm quantity by its analog in 2-norm, giving significant speedup for systems with large connectivity. Our results generalize to Hamiltonians with Fermionic terms and when the input state is drawn from a low-particle number subspace. Further, when the Hamiltonian itself has Gaussian coefficients (e.g., the SYK models), we show the stronger result that the 2-norm behavior persists even for the worst input state. Our main technical tool is a family of simple but versatile inequalities from non-commutative martingales called uniform smoothness. We use them to derive Hypercontractivity, namely p-norm estimates for low-degree polynomials, which implies concentration via Markov's inequality. In terms of optimality, we give examples that simultaneously match our p-norm bounds and the spectral norm bounds. Therefore, our improvement is due to asking a qualitatively different question from the spectral norm bounds. Our results give evidence that product formulas in practice may generically work much better than expected.