Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Table 1.

Description of AI electricity mix assumptions (rows) and AI deployment scenarios (columns).

More »

Table 1 Expand

Fig 1.

Methodological flow of this work.

Modeling framework for estimating AI’s carbon footprint across scenarios.

More »

Fig 1 Expand

Fig 2.

Illustration of projected trends in AI computational and energy demand under different scaling scenarios from 2024 to 2050.

Projected AI model growth and energy demand under three scaling scenarios (2024-2050). Panel (a): computational complexity per model (FLOPs). Panel (b): annual model deployments. Panel (c): total computational demand. Panel (d): projected power consumption (TWh/year) for training and inference. Curves show median projections; shaded bands denote 90% confidence intervals. Scenarios include baseline, fewer larger models, and more smaller models.

More »

Fig 2 Expand

Fig 3.

Projected trajectory of AI’s share of global electricity consumption under different scaling scenarios (2024-2050).

Projected share of global electricity demand from AI (2024-2050) under baseline, fewer larger models, and more smaller models scenarios. Curves show median forecasts; shaded bands indicate 90% confidence intervals.

More »

Fig 3 Expand

Fig 4.

Projected AI carbon footprint by electricity mix and AI energy demand scenarios for 2030 and 2050.

Projected AI carbon footprint in 2030 and 2050 under different electricity mix and model scaling scenarios. Panel (a) presents Business-as-usual (BAU) outcomes; panel (b) shows Energy Target (ET) projections. Each includes three deployment strategies: baseline, fewer larger models, and more smaller models. Emissions are disaggregated by lifecycle phase: training, usage, manufacturing, and transportation.

More »

Fig 4 Expand

Fig 5.

Projected AI adoption rates by country group, stratified by global inequality-adjusted human development index (IHDI), 2024-2050.

Projected AI adoption rates (2024-2050) across five country groups stratified by inequality-adjusted human development index (IHDI). Group A includes the top 20% of countries by human development, while Groups B-E represent progressively lower levels. Adoption follows logistic growth, with faster uptake in higher-IHDI countries.

More »

Fig 5 Expand

Fig 6.

Global distribution of AI user shares and corresponding carbon burden across selected countries for 2030 and 2050.

Panels (a) and (b) depict each country’s share of global carbon footprint of AI (green) among the countries analyzed for 2030 and 2050 respectively. In all panels, each country’s share of global AI users is shown in blue, with the size of each bubble corresponding to a larger share.

More »

Fig 6 Expand