We investigate the dynamics of thermally activated shear localization in power law viscoelastic materials. A two-dimensional (2D) thermomechanical numerical model is applied that uses experimentally derived flow laws for rock. We consider viscous and viscoelastic rheologies and show that the numerical solutions for shear bands are mesh insensitive and energetically conservative. Deformation under long-term tectonic strain rates (10 −14 s −1 ) yields to shear localization on the scale of kilometres. Although viscous and viscoelastic models exhibit comparable shear zone thickness, the timing of shear localization and stress magnitudes are affected by viscoelasticity. Large values of shear modulus (10 11 Pa) promotes fast stress loading (within 1% strain) and localization (<10% strain), whereas lower values (5 × 10 9 Pa) delays localization (∼15% strain) and reduces the maximum effective stress by a factor two. High stress exponents (up to n = 10) focuses the deformation into narrow shear zones (∼1000 m) while maintaining a relatively high average stress level in the material outside the shear zone (stress drop of ∼60%). Conversely, Newtonian materials produce broad shear zones (∼6× larger than for n = 10) and exhibit a stronger thermal weakening (stress drops of ∼85%). Finally, we evaluate the thickness of shear zones for a wide range of strain rates (from 10 −14 s −1 to 10 10 s −1 ) using both numerical models and an analytical scaling law. Our results suggest that thermally activated shear localization may occur from the scale of kilometre down to the nanometre.