AI and Clean Energy: How Governments Can Unlock the Power of the “Twin Transitions”

Contributors: Adrian Gonzalez (IRENA), Binu Parthan (IRENA), Bohan Liu (IRENA), Mamadou Goundiam (IRENA), Oisin Commane (Masdar) Toyo Kawabata (IRENA) 

Artificial intelligence presents both a growing challenge and an immense opportunity for the global energy transition. Both AI and energy systems are in the midst of profound transformation. On the energy side, the world is moving from a centralised, fossil-fuel-based system to one that is decentralised, digitalised and powered by clean energy. On the AI side, it is shifting from an era of narrow, niche applications to one in which AI is a general-purpose technology and a foundational piece of digital infrastructure that will underpin economic competitiveness, governance and public-service delivery. This shift is characterised by exponential growth in computing power, model complexity and deployment scale – all of which are driving unprecedented demand for energy and raising concerns about grid stability, energy availability and sustainability. While new, more efficient computing chips and AI models such as DeepSeek bring into question the scale of such projections, the technology’s broader trajectory points to significant and sustained energy implications.

At the same time, AI is emerging as a powerful enabler of the energy transition: helping optimise energy systems, manage increasingly complex grids, improve forecasting, and accelerate improved efficiency and deployment of renewables. Countries that can strategically align AI growth with clean-energy objectives will position themselves at the forefront of economic and technological leadership in the decades ahead. This is true not only for frontier, high-income countries but also particularly for fast-growing emerging-market economies, where surging energy demand, rapid urbanisation and digital transformation intersect, creating a unique opportunity to leapfrog traditional infrastructure models. The twin AI and energy transitions are therefore not just a technical or climate issue – they are a matter of national and global strategic importance.

To successfully navigate the twin transitions and harness the full potential of the AI era, governments must work with the AI industry to identify the “win-win” opportunities for both sides.

This requires creating the right policy and regulatory settings to make a country attractive for AI investment. But this investment must also help to deliver the aims of the government – delivering clean-energy infrastructure, driving energy-efficient AI deployment, and utilising AI to improve the energy system and make it faster, easier and cheaper to meet climate objectives.

This will require governments to understand the core components of the AI and energy nexus:

• Both their own strategic drivers and those of AI companies, to understand where they align and where they don’t.

• The physical infrastructure needed to build compute capacity.

• The right market settings to incentivise co-investments in AI and energy technology and harness the opportunities for AI in the energy system.

• International collaboration opportunities and strategies.

• The software investments necessary to minimise energy demand.

• Frameworks for harnessing the co-benefits.

Actions that will lead the twin transitions in the right direction include: putting in place joined-up strategies that align AI and energy policy; setting clear market conditions and regulatory frameworks that attract AI investment while driving clean-energy deployment; accelerating innovation in energy-efficient AI hardware and software to reduce long-term energy demand; and ensuring emerging markets are equipped with the infrastructure, investment and institutional support needed to participate in – and benefit from – the AI era. At the global level, coordinated governance and standards will be critical to avoiding a race to the bottom and ensuring AI infrastructure supports climate, development and energy-access goals.

With AI advancing rapidly, and energy and AI policy typically siloed within governments, leaders are struggling to navigate the growing interplay between AI and its energy demands. At stake are not just national prosperity and the ability to deliver secure, affordable and clean energy, but also technological competitiveness and geopolitical influence in a world increasingly shaped by both energy and AI capabilities. Leaders are also grappling with how to design the right policies and regulations to unlock opportunities within both energy and AI while ensuring long-term resilience, sustainability and strategic autonomy.

This paper sets out the key opportunities and challenges that leaders must confront as they navigate this nexus. Its aim is to help shape a global policy agenda that advances the twin transitions and captures the broader benefits of both elements, while also providing a framework to support leaders in accelerating action – delivering smartly and strategically on this emerging priority. COP30’s Action Agenda features an item addressing these topics, underscoring the importance of prioritising this conversation now.

The convergence of the AI revolution and the energy transition is transforming global economies. These two powerful trends are deeply interconne­­­­cted and each could be a driver of the other’s success: AI has the potential to accelerate the energy transition, while investments in clean technology can help expand and accelerate the development and application of AI.

The Growing Importance of Energy for AI

Computing has always required power. In the early days, computers were highly inefficient, consuming vast amounts of energy for relatively modest computational output. While computing has become more efficient over time, demand has also increased. Until now, improvements in computer-system performance have been accompanied by advancements in efficiency, mitigating unit-level energy requirements. However, the widespread adoption of computer systems and digital services has contributed to a rise in their energy consumption over the past decade, particularly in data centres, communication networks, the cryptocurrency and blockchain sector, and other digital spheres.

Today, the AI revolution is now significantly contributing to rising power demand. While data centres are becoming more efficient – the average Power Usage Effectiveness (PUE)^[_] score for data centres was 2.5 in 2007, compared to 1.57 in 2021, with some data centres scoring as low as 1.1 today^[_] – the demand for computing is growing more rapidly. Since 2012, the computational power used to train the largest AI models has grown 100-million-fold.^[_] Ten years ago, nearly all data centres used less than 10 megawatts (MW) of power. Today, large data centres can require 100 MW of power or more.^[_]

Even with the efficiency advances seen in recently launched AI systems like DeepSeek’s, overall power demands will likely continue to rise as AI workloads scale and its use becomes more widespread. The combination of larger model architectures, increasing demand for AI to think and respond instantly and at scale, and expanding global adoption of AI-driven services, suggests that the energy footprint of AI will remain a critical challenge for the foreseeable future.

For instance, while the industry continues to innovate on chip efficiency, AI-specific chips are consuming more absolute power. While Nvidia’s new GB200 NVL72 AI system can train and run AI models more efficiently, it consumes much more power in an absolute sense, using 120 kilowatts (kW) per rack (compared to 5 to 10 kW of power for a rack in a typical non-AI-specific data centre).^[_] Even if future chips and rack configurations are more computationally efficient (as is likely), they will still consume much larger amounts of power. Some projections suggests that AI-training facilities capable of accommodating 1 to 5 gigawatts (GW) of power demand are feasible by 2030.^[_]

AI’s demand for power is now increasing at such a breakneck speed that energy systems and utilities cannot keep up. Countries such as Ireland, Singapore and the Netherlands have instituted moratoriums on the construction of new data centres in certain regions to limit strains on the grid.^[_]

While AI is estimated to account for less than one-fifth of overall data-centre energy demand today, this share is likely to grow quickly over the next few years, with the International Energy Agency (IEA) projecting that global electricity demand from data centres could double between 2022 and 2026.^[_] One industry analyst projects that data-centre energy use could triple between 2023 and 2030, with AI accounting for 90 per cent of the growth.^[_] In some countries with heavy concentration of tech activity, such as Ireland, data centres and new tech could account for up to a quarter of a country’s electricity use.^[_] In the United States, data-centre power needs are expected to triple by 2030, rising from 3 to 4 per cent of total US power demand today to 11 to 12 per cent by the end of the decade.^[_]

This is an emerging challenge for the world: generating enough energy and connecting new AI infrastructure to the grid at a pace and scale that enables rapid AI growth, in addition to addressing the increasing energy demand resulting from the electrification and wider economic development of economies around the world.

The AI Challenge – and Opportunity – for Climate Action

The exponential growth in AI-driven energy demand has led to considerable concern that AI could undermine global climate ambitions. With the widespread adoption of AI and other AI-era digital technologies such as cloud computing and the Internet of Things, the demand for data processing will continue to surge. The proliferation and application of these technologies will require more powerful computing capabilities and larger storage capacity, driving increased electricity consumption.

This is not an unjustified concern. The IEA’s global projections for coal demand were adjusted upwards in late 2024, suggesting that the world won’t start seeing a reduction in coal demand for a few more years yet.^[_] Meanwhile, within the United States, the global front-runner for AI data-centre build-out, there is a growing demand for new gas generation to meet increasing power-consumption requirements.

The fear is that, if AI’s energy need is met by fossil fuels, or if newly added clean power is used primarily to support AI infrastructure rather than decarbonising the broader economy, global emissions-reduction efforts could be slowed rather than accelerated.

But this outcome is not inevitable. In fact, driven by corporate climate commitments – particularly among the major AI hyperscalers like Microsoft, Google, Amazon and Oracle – the AI boom is becoming a powerful accelerator for clean-energy investment. These companies have set some of the most ambitious climate goals in the private sector. Google and Microsoft, for example, are aiming to power their operations with 100 per cent carbon-free energy, 24 hours a day, seven days a week by 2030. Amazon has pledged to reach net zero by 2040.

To meet these targets, hyperscalers are increasingly investing in innovative clean-energy technologies such as advanced battery storage, small modular reactors (SMRs) and geothermal energy, and are now among the largest corporate purchasers of renewable-energy contracts. Amazon, Google, Meta, Microsoft and Apple accounted for more than 45 gigawatts (GW) of corporate renewable purchases worldwide in 2023,^[_] which is more than half of the global corporate renewables market. Microsoft has agreed to back $10 billion in renewable-electricity projects with Brookfield Asset Management over five years, marking the world’s largest corporate purchase of renewable energy.^[_] The AI hyperscalers are also driving investments into nuclear power with landmark commitments such as Microsoft’s agreement to purchase power to reopen the Pennsylvanian nuclear reactor Three Mile Island Unit 1 and Google’s deal with SMR developer Kairos Power to support early demand for new nuclear technologies. Hyperscalers are also creating similar agreements to drive investments into other more novel clean technologies, like advanced geothermal, novel long-duration energy storage or nuclear fusion.

In total, tech companies, corporations and utilities are projected to spend nearly $1 trillion on capital expenditures in the coming years to support AI growth.^[_] As a result, AI infrastructure, while often seen as a risk to net zero, is also becoming a major demand signal for clean power, helping scale and commercialise the next generation of zero-carbon energy solutions.

This represents a significant opportunity. For individual countries, there are opportunities to unlock vast volumes of investment in essential 21st-century infrastructure. For the world, having a group of corporate buyers who are willing to be offtakers of clean power sources at a cost premium offers an opportunity to drive innovation in clean energy, accelerate clean-energy investments in more challenging markets and power AI-driven development.

It is also possible, with the right government policy settings, that hyperscaler power demand could drive additional investments in clean power beyond those required to fulfil AI’s direct need, directly helping to accelerate domestic decarbonisation.

In addition to the energy-investment opportunities offered by AI, its use in the energy system is helping to lower cost, enhance efficiency and speed up decarbonisation in increasingly numerous ways. These include improved weather forecasting, optimised grid operations, enhanced energy-storage management, predictive maintenance of infrastructure and improved demand-response mechanisms. AI is rapidly emerging as an answer to many of the challenges around the efficient integration of intermittent and decentralised renewable sources.

Co-investments in compute and clean energy can also be harnessed to drive research and development. The UK’s first “AI Growth Zone” – a place-based initiative designed to accelerate the development, deployment and diffusion of AI by concentrating public and private investment in high-potential regions – will be at the UK Atomic Energy Agency’s Culham campus, with the potential for some of the associated compute investment to help accelerate progress in fusion development. Based on this model, governments and industry could decide to set compute capacity aside for research institutions or local start-ups, to develop AI solutions to local climate challenges.

AI has the potential to be an accelerant of clean energy, but countries across the world will need to implement the right policy conditions to harness this opportunity and unlock the positive loop. As AI and digital solutions evolve at speed, often outpacing institutional and regulatory capacity, policymakers will need to stay closely attuned to technological developments. This means not only assessing the near-term risks and benefits of emerging applications, but actively tracking where the twin transitions are headed and adapting policies and regulations to steer innovation towards resilient, sustainable outcomes.

Unlocking the Twin Transitions in Emerging Markets and Developing Countries

Currently, a significant portion of AI investment is concentrated in just two countries. The US and China together host nearly half of the world’s graphics processing unit (GPU)-enabled cloud regions, and the US alone hosts multiple regions equipped with NVIDIA’s advanced Hopper 100 (H100) GPUs, essential for training state-of-the-art AI models. In contrast, most emerging markets and developing countries (EMDCs) lack this infrastructure, resulting in “compute deserts” – countries that have limited or no public cloud AI compute access, which hinders local innovation and AI governance.

The countries that find themselves on the wrong side of this new AI and compute divide face limited access to the compute power needed to unlock AI’s economic-development potential, with ripple effects for developing the skills and capacity needed in an increasingly AI-driven global economy. The divide also raises important concerns around data security, sovereignty and privacy, as some countries may be unable or unwilling to rely on foreign infrastructure for processing sensitive data – further reinforcing the need for locally accessible, trusted compute infrastructure.

A core barrier to compute investment in EMDCs is energy. In many of these countries, energy systems are often less developed, with less resilient generation, transmission and distribution capacity; many communities still lack access to energy and in lots of cases utilities can be insolvent. Wider macroeconomic conditions can also mean that financing energy projects is more expensive and often considered too risky by traditional investors.

This means that rather than harnessing the positive loop of the AI and energy twin transitions, many EMDCs can find it harder than developed countries to attract the necessary investment in both AI and energy infrastructure .

The result of this emerging trend could be an increasingly entrenched divide – with developed countries able to build compute rapidly and at scale, while EMDCs are left behind without the ability to build the compute required to power their AI futures, or to attract the investment boom the global data-centre industry represents.

This would demonstrate the loss of a significant opportunity to drive investment and development into parts of the world where they are most needed. AI power demand could be a strong demand signal and offtaker of power, creating investable clean-energy projects and building domestic skills and capabilities that can unlock further investment. Furthermore, a lack of compute capacity and data availability within EMDCs could slow down the development of AI-enabled solutions tailored to EMDC-specific decarbonisation challenges. There is already a shortfall in the skills and capacity required to develop and use AI solutions in developing countries and regions – especially those that are climate-vulnerable – and capacity-building support is necessary to address this.

Unlocking the positive loop between AI and clean energy to drive global climate action and economic development should be a core priority for political leaders in every country around the world today.

Despite growing familiarity with both the AI and energy transitions, most countries still lack a joined-up strategy for navigating their intersection. Energy and digital policy are often developed in silos, and existing governance models are too slow or fragmented to keep pace with the scale, speed and complexity of technological change. Meanwhile, competition for compute, clean power and talent is intensifying — threatening to widen global divides and lock in inefficient or unsustainable trajectories. A new playbook is required to show how every country in the world can unlock the opportunities and mitigate the risks of the twin transitions.

The AI era is already transforming the energy system. Enabled by strong strategy, policy and delivery structures, there are opportunities to align the forces of the AI revolution and the energy transition to deliver growth, prosperity and accelerated climate action.

To help leaders navigate this new challenge, the playbook must consider all the core components of making the twin transitions a success.

Understanding the Strategic Drivers

To unlock the opportunities of the twin AI-energy transition, it is essential to understand the strategic goals of both governments and AI companies.

Aligning these goals will create win-win solutions for AI growth and clean energy – and ultimately drive wider climate progress. Achieving this alignment requires identifying the key drivers behind governments and AI companies’ actions and decision-making, as well as clarifying their high-level priorities. Answers to these questions will shape how governments and companies navigate and address the other key components of the AI-energy nexus (infrastructure planning, investment strategies, international collaboration, software advancements and driving co-benefits).

Infrastructure Requirements

The successful rollout of data centres and clean-energy infrastructure relies on the availability, alignment and speed of critical infrastructure development. AI can also support efforts towards predictive maintenance and reducing downtime of physical energy infrastructure and grids, detecting safety risks and identifying when infrastructure components will likely fail. However, misalignment between data-centre energy demands, grid capacity and broader infrastructure systems can create bottlenecks, strain energy networks and slow the adoption of clean energy. These challenges are further compounded by fragmented planning, where energy, compute and industrial infrastructure are often developed in isolation.

To unlock wider economic opportunities, countries must take a systemic and holistic approach to infrastructure planning, ensuring energy and compute infrastructure are integrated and coordinated with other industries and sectors. Critically, this also includes integration with water infrastructure – although a detailed exploration of water-related challenges is beyond the scope of this paper. Countries that can achieve this effectively and at speed will be better positioned to attract investment, strengthen energy systems and drive broader economic growth – all while supporting the clean-energy transition.

Investment and Markets

Unlocking the opportunities presented by the twin transitions requires the right investment and market settings. Large AI hyperscalers are increasingly becoming critical offtakers of renewable energy to meet their net-zero targets, creating a significant opportunity to drive clean-energy deployment. However, structural barriers such as high financing costs, regulatory hurdles, unreliable energy systems and the underpreparation of associated infrastructure, such as telecommunications and the internet, can limit these investments, particularly in EMDCs where the potential benefits are significant.

At the same time, the breakneck pace of AI development is creating pressure, prompting some companies to prioritise speed and scale over sustainability and clean-energy commitments, despite AI’s long-term potential to support decarbonisation and help address climate challenges. Aligning investment structures with energy goals is therefore essential to ensuring that the expansion of AI infrastructure becomes a catalyst – not a competitor – for clean-energy access. Understanding the right enabling conditions for investment is crucial to unlocking these opportunities and delivering equitable global benefits.

International Collaboration

International collaboration is critical to addressing the global-scale challenges of AI growth and clean-energy deployment. Without coordination, siloed data-centre development can lead to inefficiencies and a harmful race to the bottom, where countries compete by offering unsustainable incentives. This approach risks undermining long-term energy-system stability, sustainability goals and economic resilience, as short-term gains take precedence over cooperative solutions.

However, greater collaboration can unlock mutually beneficial outcomes. Coordinated cross-border agreements on AI infrastructure and energy systems can reduce costs, improve access to compute and clean energy, and drive innovation and economic growth. By supporting “cooperative competition”, countries can avoid undercutting each other while harnessing shared opportunities to build integrated, low-carbon digital infrastructure that benefits regions and economies more broadly.

Impacts and Co-Benefits

While data centres generate substantial energy demand and attract investment, their economic and social benefits are often limited compared to other energy-intensive industries, as they provide fewer direct jobs or immediate value to local communities. This imbalance can lead to concerns regarding their overall contribution to regional economies. However, these limitations persist because current planning and policies often fail to align AI infrastructure with broader societal goals or prioritise opportunities for local engagement.

If managed strategically, AI investments can unlock significant co-benefits: improving energy reliability through grid reinforcements, enabling greater integration of clean-energy sources, and providing access to AI resources for education, research and innovation. Realising these benefits requires intentional policies and incentives that can ensure AI infrastructure delivers meaningful, long-term value beyond its immediate operational needs.

Software and Chips

The rapid growth of AI workloads and the increasing complexity of models are driving AI’s energy demand. While there are significant opportunities to reduce this energy impact through software optimisation, energy-efficient AI design and hardware advancements in AI-specific chips, these practices are not yet widespread. Current incentives prioritise performance and scale over energy efficiency, which means that making both AI model development and chip design greener and leaner is often deprioritised. A lack of clear measurement frameworks and accountability mechanisms further limits the visibility into energy consumption, slowing progress toward more sustainable software practices. By addressing these challenges, AI developers, chipmakers and policymakers can unlock the economic and environmental benefits of energy-efficient models and chips, driving innovation that reduces costs.

AI’s rapid expansion is reshaping the global energy landscape, creating both a pressing challenge and a transformative opportunity. Without strategic intervention, AI’s soaring energy demands risk straining grids, delaying clean-energy deployment and deepening energy inequalities. However, if harnessed effectively, AI can become a catalyst for clean energy innovation – accelerating decarbonisation, optimising grid efficiency and driving investment in renewables and advanced energy-transition technologies.

AI can quickly detect faults and disruptions in the power grid, enabling a “self-healing” grid. This means that if a fault occurs, AI can help reroute the electricity flow to minimise the impact of the failure. In California, for instance, AI-based systems are used to detect faults such as tree branches touching power lines or equipment malfunctions. These systems can automatically identify the problem and isolate it, preventing widespread outages.

This is only the start. Done right, the twin transitions can complement one another to deliver transformational global benefits. But first, there are critical issues and questions that must be addressed to ensure AI supports rather than undermines the energy transition. From aligning investment incentives to strengthening policy frameworks and global cooperation, leaders must act now to shape AI’s role in the energy system. The choices made today will define whether AI drives sustainability and economic resilience or exacerbates existing challenges.

With COP30 on the horizon, and AI and climate part of its Action Agenda, the twin AI and clean-energy transitions are emerging as a defining priority. By considering both the possible benefits and the potential risks, this paper aims to lay the groundwork to ensure that AI’s future is inclusive and powered by clean energy.