A multi-area SIR (susceptible-infectious-recovered) epidemiology model is embedded in a optimization model with vaccination allocation decisions across areas. A more contagious variant emerges after a random number of person-infectious days. After several approximations, the resulting nonconvex program is solved using a sequence of linear program approximations. The model uses COVID-19 data to identify scenarios under which it might be in a high-income nation's own interest to donate vaccine doses to another country before its own population has been fully vaccinated, postponing the variant. Generally, donating vaccine greatly decreases total predicted deaths, with a modest or no increase in donor country deaths. The model uses global data and up to 10 areas, representing groups of countries. It could be used with more areas and country-specific data.