AI’s climate challenges and the potential opportunities

Artificial intelligence has rapidly become the emblem of technological progress, promising breakthroughs across industries from energy optimisation to climate risk prediction.

Yet behind this promise lies a reality that is far more complex, especially from an environmental and social standpoint. While AI may contribute to climate action and efficiency improvements, the current narrative tends to obscure significant environmental risks and systemic uncertainties.

If these challenges are not addressed, AI could inadvertently become a driver of ecological degradation and social disruption rather than a tool for sustainable progress.

Here we examine the environmental challenges linked to AI, focusing on energy consumption, emissions trends, and infrastructural pressures.

The hidden and rising energy footprint

Although AI currently represents only a small percentage of global emissions, the concern lies not in its absolute numbers but in the pace of exponential growth.

Electricity demand for data centres is among the only drivers of greenhouse gas emissions growth. Data centres, road transport and aviation are all projected to increase their direct and indirect emissions by 2030.¹

Historically US data centres’ electricity consumption was 1% to 2% of national use but has now surged to 4% to 5%.²

The International Energy Agency³ estimates that data centres currently emit around 180 metric tons of carbon dioxide, representing 0.5% of global fuel‑combustion emissions.

These emissions could grow by nearly 80% by 2030, even in conservative scenarios. In the US, electricity demand from data centres could surpass that of aluminium, steel, cement, chemical production, and all other energy‑intensive industries combined, by 2030.

These trends raise a fundamental question: Can AI continue scaling without overwhelming national and global energy systems or undermining national climate commitments?

Overbuilding power infrastructure

Despite high uncertainty over future AI-driven electricity demand, the projections are all pointing to substantial growth.

AI-driven electricity demand could double by 2030, with total future projections ranging from 700 to 1700 terawatt-hours by 2035.⁴

These projections have triggered a significant wave of new power generation projects to meet anticipated data centre demand. In the US, gas-fired capacity has tripled between 2024 and 2025, with 252 gigawatts of pre-construction and under-construction power plants announced.⁵

The rapid expansion of AI data centre capacity is already causing localised electricity pressures, especially in the US. According to Bloomberg⁶, wholesale electricity costs increased by as much as 267% more per month than five years ago in areas near data centres.

But some stakeholders are starting to question the reality of the future electricity needs, pointing to a potential over-investment in unnecessary fossil fuel-based power infrastructure and under-investment in renewables.

For example, the utility industry in the south-east US is planning for 50% more AI-related electricity demand than the tech sector itself projects.⁷

AI growth, if uncoordinated, may therefore create economic inefficiencies, electricity price inequality and systemic grid vulnerability, as well as increased stranded asset risk.

Uncertain climate benefits

AI applications could bring massive optimisations, potentially leading to emissions reductions.

Examples of such applications range from enhanced methane leak detection in oil and gas operations to power production and energy consumption optimisation in industry, transport and buildings.

According to the IEA⁸, these optimisations, if materialised, could mean emission reductions three-to-four times the size of total data centre emissions estimations by 2035. 

AI could also help to better understand climate change and its potential impacts. The benefits of AI for climate change adaptation range from faster monitoring of iceberg melting and improved deforestation mapping to more accurate prediction of climate-related disasters.⁹

But while AI presents a long list of potential climate benefits, there is currently limited momentum guaranteeing the widespread adoption of climate-oriented AI applications.

This means that many projected emissions reductions remain hypothetical, relying on assumptions about future deployment, usage patterns, and technological maturity.

The IEA warns that even potentially transformative applications may have a “marginal” aggregate impact by 2035 if enabling conditions are not created.¹⁰

In addition, efficiency gains achieved through AI may not necessarily translate into lower emissions, as these improvements – and the innovations they enable – can create new sources of demand. 

The so-called rebound effect, also known as Jevons paradox, describes situations where efficiency gains ultimately lead to increased overall consumption.

AI‑enabled efficiency must not be assumed to translate linearly into emissions reductions. Without proper regulation and behavioural incentives, improved performance can simply stimulate higher demand.

From risks to opportunities: Regional and sector focus

Despite the real environmental risks posed by AI, we do not believe its development should be avoided or slowed down.

A real-world case from Asia already demonstrates how AI can turn energy challenges into opportunities.

The 24/7 operation of AI data centres, combined with AI-driven grid optimisation, is helping smooth demand fluctuations and support higher renewable penetration.

In China’s western provinces, large-scale AI data centres are being co-located with renewable energy bases under the East Data West Computing initiative.

These facilities are actively absorbing curtailed solar and wind electricity that would otherwise go to waste. China’s East Data West Computing project is a leading example, while India is piloting similar geographic arbitrage in its solar-rich states.

AI is helping reduce net environmental pressure through rapid algorithm optimisation, which significantly lowers energy consumption per task.

For example, Chinese AI company DeepSeek has achieved substantial efficiency gains by design – activating only a small fraction of its model for each query and using lighter-weight computation – reportedly cutting training compute by around 42.5% versus a comparable conventional model.

This surge in AI-driven power demand is also creating positive financial returns.

Data centres and their dedicated power systems are evolving from traditional tech/real estate assets into quasi-infrastructure assets, supported by long-term power purchase agreements from hyperscalers. This is unlocking large-scale capital into renewables, energy storage, and transmission across the region.

Beyond these benefits, AI is becoming a new growth engine for national competitiveness in Asia. The ongoing Middle East conflict has intensified global energy security concerns and accelerated the search for alternative energy sources.

Asia’s dominant position in clean energy manufacturing – particularly China’s leadership in solar and batteries – allows the region to meet its own surging power demand while enabling emerging markets to leapfrog traditional fossil-heavy development paths.

 

In short, AI presents genuine risks, but in Asia it is also becoming a powerful catalyst for positive environmental outcomes, attractive financial returns in sustainable infrastructure, and long-term national competitiveness. [11][12][13]

Looking forward

Ultimately, AI will continue to develop and evolve, but we believe that to ensure it becomes part of the solution, rather than a new environmental burden, several conditions must be established:

• Massive investments in renewable energy, nuclear, grid flexibility, and energy‑efficient models
• Prioritisation of, and investment in, AI applications with high societal value
• Incentives for sustainable compute practices
• Regulatory frameworks to limit rebound effects and guide responsible deployment

AI holds the potential to support climate action and drive industrial efficiencies, but this potential is neither guaranteed nor automatically realised.

As currently deployed, AI is associated with rising energy demand; increasing emissions; heavy water consumption; critical mineral pressures; and profound social uncertainties.

Without strong governance, AI’s uncontrolled expansion could undermine climate goals and deepen environmental challenges.

Recognising AI’s environmental challenges is not pessimism, it is responsible realism. By asking the inconvenient questions today, we give ourselves a chance to align AI with a sustainable future.