AI’s Energy Hunger Puts Power Systems at Centre of Development Strategy

A new UNDP report argues that the global AI race is increasingly becoming an energy race, with access to abundant, affordable and low-carbon electricity emerging as a decisive factor in national competitiveness. The report warns that developing countries could either use renewable energy potential to build Green AI ecosystems or be left supplying land, minerals, power and data while higher-value gains accrue elsewhere.

A new report by the United Nations Development Programme (UNDP) states that the future of artificial intelligence (AI) will be shaped not only by algorithms, data or chips, but by access to clean, cheap and reliable electricity.

The report, "The Energy Sprint of the AI Race: A Green Window of Opportunity for Developing Countries?", prepared by UNDP in collaboration with the Technology and Industrialisation for Development Centre at the University of Oxford, warns that AI and the global energy transition can no longer be treated as separate policy tracks. Both depend on electricity grids, water, land, minerals, finance, regulation and institutional capacity.

The report cites projections that electricity consumption by data centres could more than double by 2030, reaching 945 terawatt hours. AI could add US$15.7 trillion to global output by 2030, while less than 10 percent of that value is expected to accrue to developing economies. It suggests that many countries, especially in Africa and Small Island Developing States, may have a more realistic entry point through modular, distributed and energy-efficient systems that match local demand and reduce pressure on fragile grids.

Why AI is becoming an energy and development issue

Modern AI relies on semiconductor fabrication plants, high-performance computing clusters, hyperscale data centres, cooling systems, stable power, water and critical minerals. These requirements link AI directly to energy policy, industrial strategy and climate governance. Countries with strong renewable energy potential may gain bargaining power as AI firms look for low-carbon power and reliable infrastructure, but renewable potential alone is not enough. The report warns that solar, wind or geothermal resources must be matched by stable grids, predictable regulation, industrial capability and domestic demand.

Without those conditions, developing countries may host energy-intensive infrastructure without building broader economic value. The report calls this risk a possible "extractive digital" pattern, where countries provide land, power, minerals and data but remain dependent on foreign cloud platforms, imported technology and limited local linkages.

Mineral-rich economies have often struggled to move beyond extraction. The report suggests a similar risk in the AI era unless countries move up the critical minerals-energy-compute value chain through refining, processing, component manufacturing, storage systems and local AI-energy innovation.

The Green AI opportunity and its limits

The report defines Green AI as a policy approach that aligns AI development with clean energy, environmental sustainability and human development. Practically, this means powering AI with cleaner electricity, reducing water, land and mineral pressures, using AI to improve energy and climate systems, and ensuring that AI infrastructure creates local value rather than new dependencies.

AI can support grid optimization, demand forecasting, renewable integration, predictive maintenance, energy efficiency and climate-risk modelling. These applications could help utilities, firms, farmers and public agencies improve services and reduce losses, but the report also cautions that AI does not diffuse automatically from global innovation hubs into local economies.

To scale adoption, countries need trusted local use cases, affordable services, skilled users, accessible data, institutional buyers and financing models that allow experimentation without excessive risk. This is crucial in the energy sector, where utilities and regulators often face high pressure to maintain reliability and may be cautious about untested technologies.

The report's startup mapping illustrates both promise and fragility. It analyses 88 Green AI startups across Brazil, Chile, India, Malaysia, Mexico, Nigeria and South Africa. These ventures work across areas including AI for energy, AI for climate, energy storage, grid infrastructure, climate finance, sectoral applications, deeptech materials and AI/compute infrastructure. Activity is concentrated in AI for energy, grid-edge intelligence and AI-enabled infrastructure.

However, commercial traction remains concentrated in a smaller subset of firms. The report identifies barriers including limited early adopters, thin venture capital ecosystems, weak demand, regulatory uncertainty, gaps in technical and business skills, fragmented public governance and misalignment between innovation agencies, energy regulators and digital ministries. This suggests that the Green AI window could close or narrow if startups cannot move beyond pilots.

Who is affected and what is at stake

The report treats AI-energy integration as a planning and governance issue, not just a technology issue. Ministries responsible for energy, digital affairs, industry, education, finance and environment will need to coordinate policies that are often managed separately. Utilities and regulators may face new demand from data centres and AI infrastructure, but they could also use AI to improve system performance. Their choices on procurement, grid upgrades, data access and testing rules will influence whether local AI-energy firms can scale.

Citizens and small enterprises are affected through electricity reliability, tariffs, service delivery and digital inclusion. If AI infrastructure strains already fragile grids, households and businesses could face higher costs or poorer reliability. If it is designed as public-interest infrastructure, the report suggests integrated AI-energy systems could reduce losses, stabilize tariffs and expand access.

Startups and domestic firms could benefit if governments create markets through procurement, regulatory sandboxes, concessional finance, utility partnerships, open energy data and testing facilities. But they may struggle if public institutions remain risk-averse or if large foreign platforms dominate the market.

International actors, including development banks and bilateral partners, also have a role. The report calls for concessional finance, guarantees, regional infrastructure planning and technical support whilst raising broader geopolitical stakes, as energy, minerals, data and compute become part of a single strategic domain.

Key challenges, trade-offs and policy uncertainties

The biggest risk is that AI expansion increases energy demand faster than grids can handle, pushing countries toward fossil-fuel-based or temporary high-emission power solutions. Water use, land use and mineral demand add further environmental pressure, particularly in water-stressed or ecologically sensitive regions.

Another major risk is over-investment in infrastructure before domestic demand and capabilities exist. The report suggests that fiscally constrained countries may be better served by moving early in applications, skills and institutions, while moving more cautiously on large physical compute infrastructure.

A third tension concerns sovereignty and dependency. Cloud-based services can offer lower-cost access to AI capabilities, but they may also deepen reliance on foreign providers. Building domestic compute can increase control, but it requires large capital spending, reliable energy, technical talent and regulation.

The report's policy roadmap calls for five broad actions:

1. building local AI-energy innovation ecosystems
2. making Green AI the default approach
3. moving up the minerals-energy-compute value chain
4. aligning AI, energy and industrial strategies
5. supporting Green AI startups through targeted entrepreneurship, finance and market-creation policy

What comes next

The next developments to monitor are national AI strategies that explicitly link compute infrastructure with renewable energy, grid planning and water standards. Equally important will be data-centre licensing rules, renewable procurement requirements, public procurement policies for AI-energy tools, and regulatory sandboxes for utilities and startups.

Another key signal will be whether development finance institutions create instruments suited to Green AI infrastructure, including concessional finance, guarantees, digital development bonds or regional green bonds. Regional power pools and cross-border power purchase agreements may also become more important as countries seek to create investable clean-power and compute corridors.

The most important test will be whether developing countries move from hosting AI-related infrastructure to capturing value from it, which means jobs, local supplier networks, public-service applications, startup growth, better energy systems and stronger domestic technical capacity.