The Paris climate goals and the Glasgow Climate Pact require anthropogenic carbon dioxide (CO2) emissions to decline to net zero by mid-century. This will require overcoming carbon lock-in throughout the energy system. Previous studies have focused on 'committed emissions' from capital investments in energy-consuming infrastructure, or potential (committed and uncommitted) emissions from fossil fuel reserves. Here we make the first bottom-up assessment of committed CO2 emissions from fossil fuel-producing infrastructure, defined as existing and under-construction oil and gas fields and coal mines. We use a commercial model of the world's 25 000 oil and gas fields and build a new dataset on coal mines in the nine largest coal-producing countries. Our central estimate of committed emissions is 936 Gt CO2, comprising 47% from coal, 35% from oil and 18% from gas. We find that staying within a 1.5 °C carbon budget (50% probability) implies leaving almost 40% of 'developed reserves' of fossil fuels unextracted. The finding that developed reserves substantially exceed the 1.5 °C carbon budget is robust to a Monte Carlo analysis of reserves data limitations, carbon budget uncertainties and oil prices. This study contributes to growing scholarship on the relevance of fossil fuel supply to climate mitigation. Going beyond recent warnings by the International Energy Agency, our results suggest that staying below 1.5 °C may require governments and companies not only to cease licensing and development of new fields and mines, but also to prematurely decommission a significant portion of those already developed.