Rapid artificial intelligence deployment increases near-term pressure on global carbon budgets

Limiting warming to 1.5 degrees Celsius depends on cumulative carbon dioxide emissions, not only on whether annual emissions eventually balance. Artificial intelligence is increasingly promoted as a tool for reducing emissions, but its supporting digital infrastructure produces emissions before many system-level benefits are realized. Here, we evaluate this timing mismatch using a probabilistic numerical cumulative carbon accounting model calibrated to International Energy Agency artificial-intelligence and energy scenarios through 2035. The model combines operational emissions, embodied emissions, and delayed system-level savings. Across 10,000 Monte Carlo realizations, the accelerated Lift-Off pathway yields a median cumulative carbon debt of 2.85 gigatonnes of carbon dioxide before annual savings exceed annual infrastructure-related emissions in late 2031. Across scenarios, the carbon imbalance varies with deployment speed, grid decarbonization, and the coupling between infrastructure growth and mitigation-relevant applications. These results indicate that rapid artificial-intelligence deployment can increase near-term pressure on the remaining 1.5 degrees Celsius carbon budget. Rapid expansion of AI infrastructure causes immediate carbon emissions through rising electricity demand, despite its potential to guide emission reductions, as suggested by a cumulative carbon accounting framework under AI-related scenarios