The perception of AI as “virtual” obscures its enormous industrial footprint and energy consumption. The next frontier of AI competitiveness will depend not only on scaling compute but on reducing energy per computation. In an extract from a new book Angel Melguizo and Victor Muñoz explains whyenergy efficiencies will decide the digital race and makes the case for believing in a greener future for AI.
Artificial intelligence (AI) is not just a technological shift; it is an energy transition, an industrial transformation and a geopolitical rebalancing. The next phase of AI development will not be defined by model size or algorithmic breakthroughs, but by infrastructure – who builds it, who powers it and under what rules.
In our new book, Green Code: AI+Energy, we call for transforming the AI sprint into a marathon: fast-paced, yet grounded in long-term vision, powered by high-quality data, clean energy, talent and purpose. We need a global coalition of the willing that includes not only governments, but also scientific institutions and selected private-sector actors.
Below are eleven principles that frame the economic logic of sustainable AI.
Energy will determine the digital winners. Affordable, reliable and clean energy will determine the winners and losers of the AI race. Current consumption trajectories are financially and physically unsustainable.
AI is electricity intensive. Training models, running large-scale inference and scaling enterprise deployment all require vast compute capacity. That compute capacity, in turn, requires energy at industrial scale. Independent projections show that in the next decade AI electricity consumption could be 14 times as much as it is today.
Electricity prices, grid reliability and regulatory stability are now strategic variables in AI competitiveness. Regions with constrained grids or volatile pricing will struggle to scale data centre capacity while jurisdictions that align digital strategy with energy planning will attract investment. The AI race is no longer just about talent and capital. As OpenAI’s chief executive recently testified, it is about kilowatt-hours.
Efficiency and clean energy are sound economics. Investing in efficiency, scaling clean energy, and strengthening transparency regarding AI-related energy consumption is not only the right thing to do – it is also sound business and good policy. Efficiency lowers operating costs. Clean energy reduces regulatory risk and enhances energy security. Transparency builds investor confidence and policy credibility.
The cloud is physical. A simple prompt travels through fiber cables and satellites to massive data centers packed with graphics processing units, cooling systems and backup power – consuming real electricity, water and materials along the way. Behind every instant AI answer lies a global industrial chain of chips, energy, training data and geopolitics.
Each AI interaction activates telecommunications infrastructure, high-density servers, advanced semiconductors and cooling systems that run continuously. Data centres convert electricity into computation – and heat. Cooling can account for a significant share of energy consumption. Backup systems, storage arrays and networking equipment add to the load.
The perception of AI as “virtual” obscures its industrial footprint. AI runs on grids, supply chains and hardware ecosystems that are deeply material and globally interconnected.
AI needs smarter design, not just more power. Neuromorphic chips, optical computing, superconductors, AI-specific application-specific integrated circuits (ASICs) and AI factories are reshaping the path to low-energy AI. Brain-inspired hardware processes information asynchronously, activating only when needed. Optical systems use photons instead of electrons, reducing heat losses. Superconducting materials promise near-zero resistance computation under specific conditions. AI factories consolidate compute, data, and energy planning into strategic hubs.
The next frontier of AI competitiveness will depend not only on scaling compute but on reducing energy per computation. Smarter architectures can bend the energy curve.
The “green quantum advantage” will take time. Green quantum advantage aims for quantum computers to outperform classical ones in energy efficiency for specific tasks, leveraging entanglement to solve optimisation and simulation challenges with far less power. However, this advantage is not universal and not yet mature. Quantum systems remain technically demanding and capital-intensive. For particular problem classes – complex optimisation, molecular simulation, logistics – quantum approaches may eventually deliver lower energy per solution. Together with tensor-network methods and specialised green chips, quantum research represents a long-term bet on structural efficiency improvements.
Nuclear is back, for now. AI’s energy appetite is driving a nuclear comeback. Amazon, Microsoft, Google, Meta and Oracle are securing gigawatt-scale nuclear agreements and investing in small modular reactors (the so-called “SMR”), advanced fission and even fusion pilots to power data centres with stable, carbon-free baseload electricity.
From America’s ambitious expansion targets to France’s and China’s rapid reactor expansion, nuclear energy is returning to the centre of energy security, decarbonisation, and industrial policy. Latin America is entering its nuclear moment. Argentina’s CAREM SMR, Brazil’s Angra III, Mexico’s Laguna Verde, and emerging plans in Colombia and El Salvador signal renewed diversification efforts.
“Green AI” does exist, but it might take yet more time and money. Green AI does exist, potentially following a Kuznets-type curve linking AI development, renewable energy deployment and environmental impact. In early phases, AI expansion increases energy demand and emissions. Over time, as renewable penetration grows and hardware efficiency improves, environmental intensity per task may decline.
However, achieving this transition requires substantial investment in grids, generation capacity, storage and research. Estimates point to a substantial increase in annual AI investments required to get to the green side of AI ($820 billion v $153 billion in 2023 in leading 23 economies). Recent projections that big tech could invest up to $1.1 trillion in 2027 mostly in data centres confirm this estimate. Most middle-income countries, and even some high-income countries and global companies, will find it unaffordable to accelerate both AI infrastructure and energy transformation simultaneously.
Some companies are leading the innovating towards Green AI. Most Green AI initiatives today are being driven by the private sector. Some companies are designing renewable-powered data centres, implementing circular hardware strategies and improving emissions reporting. Hyperscale operators are optimising cooling systems, deploying liquid immersion technologies and integrating renewable purchase agreements. Semiconductor firms are improving energy performance per chip generation. Yet private leadership alone is insufficient. Without public frameworks that ensure transparency, competition and grid expansion, isolated corporate initiatives may not scale equitably.
Progress towards green AI is flourishing – the challenge is scaling up. From DNA-structured chips to natural ventilation systems in data centres and bio-inspired wind turbine designs, nature-inspired energy solutions are tangible and expanding. But none of these startups have scaled globally or reached public markets. Market failures such as information asymmetries, limited competition and co-ordination failures continue to constrain economic and social returns. Early-stage energy-compute innovations often face high upfront costs and uncertain demand signals, capping their scaling and replication.
Defining Industrial Policy 4.0. A new green window of opportunity is open for regions endowed with natural resources, digitally skilled young populations, favourable climates and strong connectivity. This calls for an “Industrial Policy 4.0” approach integrating digital infrastructure, clean energy strategy and workforce development. Cold climates reduce cooling costs. Abundant renewables lower marginal electricity prices. Fiber connectivity reduces latency. Skilled labor enables innovation ecosystems. Strategic co-ordination between energy and finance ministries, digital regulators and industrial agencies is essential.
Latin America can succeed in the AI global value chain. Latin America has a unique opportunity to integrate into global AI value chains by establishing green data and computing centers and fostering industrial AI applications in sectors such as energy, finance and mobility.
Renewable-powered computing hubs, a regional network of green infrastructure funds, and regulatory modernisation can position countries as attractive destinations for AI infrastructure. Industrial AI applications – optimising grids, improving logistics and enhancing agricultural productivity – can simultaneously reduce domestic energy intensity and increase competitiveness. Our estimates show that a digital infrastructure push plus upskilling and reskilling could accelerate Latin American economic growth 1.3 percentage points and create nearly five million expert IT jobs.
Turning the AI sprint into a marathon requires focusing on infrastructure and energy
Infrastructure requires energy and energy requires policy action. The future of AI should not be decided only by algorithms, but by grids, reactors, chips, policies and coalitions. In a best-case scenario, AI’s energy consumption will multiply eight times. And it will do so in a financially and environmentally sustainable way. Powered by an increasing share of clean and renewable energies, benefiting from energy savings and governed with principles and purposes.
For centuries, humanity believed it was the center of the universe. Nearly five centuries ago, Nicolaus Copernicus challenged this belief, arguing that the Earth revolves around the Sun – and that we are neither the center nor alone. This spirit inspires our ambitious yet humble call to build sustainable AI. We must think about others and future generations. We are not the center of the digital system, but part of a much broader energy, ecological and social network. And we must act decisively, because the path of AI is not predetermined. We can shape it, and what we do will mark the decades ahead. We are in charge.
This blog is based on El Codigo Verde: IA+Energia
This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.
Image credit: Khampaeng Studio provided by Shutterstock.