From an economic, political and social standpoint, one of the most evident and visible features of today’s European Union as a supranational regional organization is its heterogeneity, where disparity seems to be the common denominator. This leads to the interest for measuring the territorial economic cohesion of the EU. From an eminently economic perspective, and working with the GDP per capita of the EU NUTS-2 regions for the period 2003–2021, this paper aims to provide evidence of a lack of territorial economic cohesion through a beta and sigma convergence methodology by applying cross-sectional and spatial panel data analysis. The findings show that the speed of convergence depends mainly on the level of economic development, its cycles and the heterogeneity of the, which implies conditional convergence. Less developed regions show higher convergence speeds, which are also accentuated during recession periods. Greater heterogeneity among the regions also increases the convergence speed, while accentuating in the less developed regions. In general terms, the results reveal convergence speeds of the entire NUTS-2 regions between 7 and 11 per cent (much higher than 2 per cent under absolute convergence). Likewise, when considering spatial dependence, a reduction in convergence speeds between approximately 3 and 8 per cent is detected. Finally, the 29 vulnerable regions have been identified, with economic development and growth below the EU average mean, emphasizing the need to take the concerns of territorial economic cohesion into account.