AI and the Net-Zero Journey: Energy Demand, Emissions, and the Potential for Transition

TL;DR

This paper investigates whether AI will advance or hinder the net-zero transition by 2035, focusing on data-center energy demand and CO2 emissions. It develops a probabilistic, scenario-informed emissions framework that integrates Shell energy security perspectives, IEA cases, and AI-specific consumption trajectories, using Monte Carlo simulations to capture uncertainty. In the near term, data-center growth and AI workloads are projected to raise electricity use and emissions, but in the long term AI is expected to enable substantial emissions reductions through optimization across energy systems, transportation, and industry, aided by technologies like CCS, modular reactors, and renewables. The work highlights the importance of green hardware, energy-aware AI design, and policy and investment strategies to ensure AI contributes positively to the net-zero agenda.

Abstract

Thanks to the availability of massive amounts of data, computing resources, and advanced algorithms, AI has entered nearly every sector. This has sparked significant investment and interest, particularly in building data centers with the necessary hardware and software to develop and operate AI models and AI-based workflows. In this technical review article, we present energy consumption scenarios of data centers and impact on GHG emissions, considering both near-term projections (up to 2030) and long-term outlook (2035 and beyond). We address the quintessential question of whether AI will have a net positive, neutral, or negative impact on CO2 emissions by 2035. Additionally, we discuss AI's potential to automate, create efficient and disruptive workflows across various fields related to energy production, supply and consumption. In the near-term scenario, the growing demand for AI will likely strain computing resources, lead to increase in electricity consumption and therefore associated CO2 emissions. This is due to the power-hungry nature of big data centers and the requirements for training and running of large and complex AI models, as well as the penetration of AI assistant search and applications for public use. However, the long-term outlook could be more promising. AI has the potential to be a game-changer in CO2 reduction. Its ability to further automate and optimize processes across industries, from energy production to logistics, could significantly decrease our carbon footprint. This positive impact is anticipated to outweigh the initial emissions bump, creating value for businesses and society in areas where traditional solutions have fallen short. In essence, AI might cause some initial growing pains for the environment, but it has the potential to support climate mitigation efforts.

Figures (8)

• Figure 1: The figure illustrates the growth of foundational models between 2021 and 2024, and the revenue growth of high-density GPU markets. Data courtesy from ecosystemgraph
• Figure 2: The figure illustrates the decrease in energy consumption per GFLOPS over the years (left) and the number of FLOPS required to train different models (right)
• Figure 3: The figure illustrates the electricity consumption of several prominent large language models published in recent years. Although the authors usually do not provide these statistics directly, some do. Many factors influence electricity consumption, but typically, if they disclose the hardware details—such as the number of GPU hours used and the types of GPUs—it's possible to calculate the overall power consumption using the GPUs' wattage. However, this calculation does not account for other components like CPUs, data storage, and cooling technology. Nonetheless, it gives an indication of the resources required to train these models.
• Figure 4: Electricity generation and associated CO2 emissions between 2020-2060 under the three Shell Energy Security scenarios. Of those, only Horizon is reaching Net Zero by the 2050s. [Courtesy of Shell_Sce_2025].
• Figure 5: a) Four scenarios for electricity demand by data centers globally between 2020-2035 (reproduced from iea2025); b) Plausible scenarios for electricity usage by AI applications between 2020 and 2035. (reproduced from paccouandwijnhoven2024).