In confined plasmas, a localized fluctuation in a marginal or weakly damped region will propagate and generate an avalanche if it exceeds a threshold. In this letter, a new model for turbulence spreading based on subcritical instability in the turbulence intensity is introduced. We derive a quantitative threshold for spreading from a seed in a stable region, based on a competition between diffusion and nonlinear growth of the turbulence intensity. The model resolves issues with the established Fisher equation model for turbulence spreading, which is supercritical and cannot support the stationary coexistence of multiple turbulence levels. Implications for turbulence spreading are discussed, including the dynamics of ballistic penetration of turbulence into the stable zone. Tests of the theory are suggested.The paradigm of directed percolation 1 is ubiquitous in non-equilibrium statistical dynamics. Directed percolation is realized by the contamination and ultimate excitation of, say, a localized nonlinear oscillator by interaction with its neighbors 2 . For the contaminated region to expand, the interaction must be strong enough to excite neighbors against their damping. This process of expansion of excited regions by directed percolation is relevant to many problems in the spatiotemporal dynamics of turbulence 3 . These include, but are not limited to, the expansion (and ultimate overlap) of turbulent slugs at the onset of turbulence in high Reynolds number pipe flow turbulence 4,5 , the spreading of a turbulent patch 6 , and the Loitsyansky problem 7 for large-scale evolution of 3D turbulence. The theme of spreading-by-contamination 8 appears in the magnetic confinement physics phenomena of turbulence spreading 9-12 and avalanching 13-15 . These closely-related phenomena are of pragmatic interest in that they delocalize the flux-gradient relation which governs turbulent transport. This letter applies the physics of spreading-by-contamination to propose a mechanism for how turbulence penetrates stable regions.Indeed, experiments and simulations of magnetic fusion (MF) plasma strongly suggest that the transport can be nonlocal : fluxes cannot be determined by local parameters and their gradients, but are generally given by some integrated spatiotemporal relation that depends on global profile structure (not only local gradients) 16,17 . Such phenomena cannot be explained by diffusive or Fickian processes and instead depend on fast propagation dynamics of turbulence in the plasma. Turbulence spreading and avalanching, both of which involve radial self-propagation of turbulence on mesoscales, are two players which influence nonlocal transport. Turbulence spreading, arguably the more general phenomenon, is the result of nonlinear coupling between ballooning modes, which results in spatial scattering and nonlinear diffusion of the turbulence energy. Standard spreading problems are the spatiotemporal evolution of a single, initially localized spot of turbulence, and the penetration of turbua) Electronic mail: robin@ph...