Photosynthesis is the largest flux in the terrestrial carbon cycle. For decades, eddy covariance networks monitor ecosystem-level photosynthesis (or gross primary production, GPP) at globally distributed sites, which have been upscaled using satellite data to understand the spatiotemporal dynamics of terrestrial photosynthesis. However, existing upscaled GPP products often overlook the response of photosynthesis to a rising atmospheric CO₂, known as CO₂ fertilization, which is responsible for a large proportion of the historic increase in the terrestrial carbon sink and changes in the interactions between carbon, water, and energy cycles. In this work, we developed a global GPP product considering CO₂ fertilization by upscaling eddy covariance measurements with satellite data, machine learning models, and biophysical constraints. Our product provides global monthly GPP maps at 0.05-degree resolution from 1982 to 2020. The CO₂ fertilization effect was implemented in two ways: a hybrid approach combining machine learning with biophysical theory, and a data-driven approach relying on machine learning alone. The spatial and temporal dynamics of our GPP estimations were evaluated against other products from remote sensing, upscaling, and dynamic global vegetation models. Our upscaled GPP product provides useful constraints of the terrestrial carbon cycle, especially the short-term and long-term temporal dynamics of global photosynthesis, necessary for better understanding vegetation-climate interactions.