We introduce a new method, which we call stochastic fusion, which takes an exclusion process and constructs an interacting particle systems in which more than one particle may occupy a lattice site. The construction only requires the existence of stationary measures of the original exclusion process on a finite lattice. If the original exclusion process satisfies Markov duality on a finite lattice, then the construction produces Markov duality functions (for some initial conditions) for the fused exclusion process. The stochastic fusion construction is based off of the Rogers-Pitman intertwining. In particular, we have results for three types of models: 1. For symmetric exclusion processes, the fused process and duality functions are inhomogeneous generalizations of those in \cite{GKRV}. The construction also allows a general class of open boundary conditions: as an application of the duality, we find the hydrodynamic limit and stationary measures of the generalized symmetric simple exclusion process SSEP$(m/2)$ on $\mathbb{Z}_+$ for open boundary conditions. 2. For the asymmetric simple exclusion process, the fused process and duality functions are inhomogeneous generalizations of those found in \cite{CGRS} for the ASEP$(q,j)$. As a by-product of the construction, we show that the multi-species ASEP$(q,j)$ preserves $q$-exchangeable measures, and use this to find new duality functions for the ASEP, ASEP$(q,j)$ and $q$-Boson. Additionally, the construction leads to duality for ASEP with open boundary conditions. As an application of the latter duality, we find the hydrodynamic limit. 3. For dynamic models, we fuse the dynamic ASEP from \cite{BorodinDyn}, and produce a dynamic and inhomogeneous version of ASEP$(q,j)$. We also apply stochastic fusion to IRF models and compare them to previously found models. We include an appendix co-authored with Amol Aggarwal.