The Energy Cost of Intelligence: Can the World Power the AI Revolution?

The Energy Cost of Intelligence: Can the World Power the AI Revolution? Artificial Intelligence, Quantum Computing and the Emerging Global Energy Challenge Clean Energy and Water Technologies Pty Ltd (CEWT) Introduction: A New Industrial Revolution Human civilisation has experienced several great technological revolutions, each transforming how energy is produced and consumed. The Industrial Revolution was powered by coal and steam. The twentieth century brought petroleum, natural gas and widespread electrification. The digital revolution introduced computers, telecommunications and the internet. We are now entering another transformation: the industrialisation of intelligence itself. Artificial intelligence is rapidly evolving from a specialised computational tool into fundamental infrastructure supporting scientific research, manufacturing, healthcare, transportation, financial systems and everyday activities. Alongside AI, advances in quantum computing, robotics and autonomous systems promise to expand the boundaries of computation. Can the world's energy infrastructure expand rapidly enough to support the computational demands of the twenty-first century? The answer depends not only on electricity generation, but also on integrating storage, cooling, water management and carbon recycling. 1. From Energy-Intensive Industry to Energy-Intensive Intelligence Traditionally, industrial energy consumption has been associated with physical processes: extracting minerals, manufacturing steel, producing chemicals, transporting goods and heating buildings. Artificial intelligence introduces another form of demand. Computational work requires physical infrastructure. Every calculation involves electrical energy, and nearly all the electricity consumed by computing equipment ultimately appears as heat. Modern AI infrastructure requires specialised processors, high-speed networks, memory systems, electrical distribution and cooling. Many facilities operate continuously. A 100 MW data centre operating at full load for 8,760 hours consumes 876,000 MWh, or 0.876 TWh per year. A 1 GW computing campus consumes 8.76 TWh annually at continuous full load. Hundreds of such installations would represent an extraordinary concentration of demand. 2. The Scale of the Emerging Electricity Demand According to the International Energy Agency’s Energy and AI report, global data-centre electricity consumption was approximately 415 TWh in 2024 and could reach around 945 TWh by 2030. These projections already include AI computing; AI must not be counted as an additional, separate load. Computing growth coincides with electrification of transport, industry, heating and hydrogen production. Electricity demand and total primary energy consumption must nevertheless be distinguished: electrification can displace inefficient combustion and reduce primary energy requirements while increasing electricity consumption. The critical constraint may increasingly become the availability of firm electricity rather than computing hardware. 3. The Paradox of Computational Efficiency Semiconductor technology has delivered extraordinary improvements in computational efficiency. Modern processors perform calculations using far less energy than earlier generations, and specialised AI accelerators continue that trend. Yet improvements in efficiency do not necessarily reduce total consumption. When the cost of computation decreases, demand for computational services may rise—a pattern associated with the Jevons paradox. More efficient AI could encourage wider use in science, transportation, industrial automation and continuous decision-making. Total computation could grow faster than energy use per computation falls. 4. Quantum Computing: The Next Frontier Quantum computing uses quantum states represented by qubits rather than conventional binary bits. Potential applications include molecular simulation, materials science, chemistry, cryptography and selected optimisation problems. Some architectures, including superconducting quantum processors, require temperatures near absolute zero, dilution refrigeration and extensive control electronics. Others have different infrastructure requirements. It is premature to conclude that quantum computing will necessarily increase global electricity demand dramatically: some problems may eventually be solved more efficiently than on classical supercomputers. Its net energy impact remains uncertain, but quantum computing could enable entirely new classes of computational work. 5. Electricity Is Only Part of the Problem A data centre converts electrical energy into computational work, with heat as an unavoidable consequence. Almost all electricity supplied eventually becomes heat. A facility consuming 100 MW must therefore reject heat approaching 100 MW in steady operation, although the distribution among equipment and cooling systems varies. High-density AI installations increasingly use direct liquid cooling. Waste heat can potentially be recovered, but its usefulness depends on temperature, heat-transfer conditions and the needs of nearby consumers. Low-temperature cooling water may support district heating or heat-pump integration, but is generally unsuitable for direct industrial steam generation. The aim is to maximise useful heat recovery without excessive additional electricity consumption. 6. The Emerging Water Challenge Many cooling systems consume water through evaporation. Requirements depend on climate, technology, operating temperature, water quality and the balance between evaporative and mechanical cooling. There is also indirect water consumption associated with electricity generation. Data-centre design should consider treated wastewater reuse, closed-loop liquid cooling, dry and hybrid cooling, and desalination where appropriate. In resource-constrained locations such as Singapore, integrated energy and water planning is especially important. 7. Why Renewable Electricity Alone May Not Solve the Problem Solar and wind will be essential contributors, but annual generation is not the same as continuous availability. A 20 MW data centre operating continuously requires 175.2 GWh annually (20 MW × 8,760 hours). At a 25% annual capacity factor, 80 MW of solar capacity would theoretically generate an equivalent annual amount of electricity. That would not guarantee continuous 20 MW delivery. Storage, overbuilding, interconnection, dispatchable generation and demand flexibility may be required. Their optimum combination depends on hourly generation profiles, reliability requirements, storage duration and economics. The AI revolution requires not merely abundant renewable energy but reliable and economically sustainable electricity infrastructure. 8. Carbon Recycling Technology: An Integrated Energy Approach CEWT is investigating Carbon Recycling Technology (CRT), in which carbon acts as a recyclable energy carrier. Methane generates electricity, its carbon dioxide is recovered, and hydrogen is used to regenerate methane by catalytic methanation: CO₂ + 4H₂ → CH₄ + 2H₂O Methanation releases heat, but hydrogen production requires substantial external energy. Carbon recycling cannot generate net energy independently. Its potential value is the integration of renewable hydrogen, dispatchable methane generation, CO₂ recovery and useful heat utilisation. Overall greenhouse gas performance must include methane leakage, capture efficiency, hydrogen production and external electricity supply. The objective is not to circumvent thermodynamics but to improve the utilisation of available energy within a controlled industrial system. 9. The CEWT 20 MW CRT Trigeneration Demonstration CEWT is developing the engineering basis for a proposed 20 MW CRT trigeneration demonstration serving data-centre infrastructure in Australia’s Latrobe Valley. It compares three configurations: renewable electricity with battery storage; renewable electricity with battery and hydrogen storage; and renewable electricity integrated with CRT recycled-methane generation and heat recovery. The objective is a consistent comparison of continuous delivery, reliability, capital and operating costs, cooling, water and carbon performance. Methanation heat may support steam generation, feed preheating or absorption refrigeration, depending on temperature and availability. The proposed trigeneration outputs are electricity, useful thermal energy and cooling. Whether the recovered heat matches data-centre cooling demand must be demonstrated. Surplus thermal energy may serve other uses, subject to heat quality, distribution costs and nearby customers. The project remains at the technology-integration and evaluation stage, pending complete mass and energy balances, vendor information and techno-economic assessment. 10. A New Measure of Data-Centre Efficiency Power Usage Effectiveness (PUE) compares total facility electricity use with computing equipment electricity use. It is useful, but does not capture the carbon intensity or cost of electricity, recovered thermal energy, water demand or overall system reliability. Future evaluation should also consider lifecycle carbon intensity, Water Usage Effectiveness, energy availability, useful heat recovery, cost per unit of computation and useful computing output per MW. No single measure captures every AI application. The guiding objective is to maximise useful intelligence delivered per unit of energy consumed. 11. The Future Energy System Must Be Integrated Electricity generation, cooling, water supply and waste management have often been designed separately. AI expansion encourages integrated planning of renewable generation, storage, dispatchable power, thermal recovery, water recycling and carbon management. Chemical process engineering offers a familiar analogy: an efficient reactor does not necessarily produce an efficient plant. Overall performance depends on reactions, separations, heat integration, compression, equipment reliability and optimisation. The same is true of data-centre energy infrastructure. Optimising individual equipment matters; optimising the complete system is essential. 12. Will Energy Become the Ultimate Constraint on Artificial Intelligence? AI algorithms and semiconductor architectures are evolving rapidly, while major transmission networks, power stations and supporting infrastructure can take years to plan, finance, approve and build. This creates a potential mismatch between computing deployment and the development of firm electricity supply. Access to reliable power, suitable land, cooling resources and infrastructure could become as strategically important as access to advanced chips. Regions offering affordable, dependable, low-carbon electricity may gain an advantage in attracting computing investment. Electricity infrastructure could increasingly become a matter of national technological competitiveness. 13. Conclusion: Intelligence Has a Physical Cost Artificial intelligence is often discussed as though it exists independently of the physical world. In reality, every computational operation depends on equipment, electricity, materials and heat rejection. Quantum computing is no exception. The challenge is not necessarily that total global energy consumption will multiply several times. More immediately, computing electricity demand could grow much faster than the infrastructure needed to support it. For CEWT, this is the broader engineering context for Carbon Recycling Technology. Its proposed contribution is to investigate whether renewable energy, carbon recycling, continuous generation and heat recovery can be combined into an economically viable, environmentally responsible system. Success must be demonstrated through engineering evidence. For two centuries, industrial progress has largely reflected our ability to harness energy for physical work. In coming decades, progress may increasingly reflect our ability to harness energy for intellectual work. Both remain governed by thermodynamics. Intelligence may advance exponentially. Energy infrastructure cannot be assumed to follow the same trajectory. Final Reflection Humanity is investing enormous resources to create machines capable of reasoning, learning and solving complex problems. Yet one of the most important problems these machines must help solve is how to supply the energy required for their own operation. Perhaps the ultimate achievement of artificial intelligence will not simply be greater computational power. It may be the development of a civilisation capable of using energy, materials and natural resources with far greater intelligence than we do today. Reference International Energy Agency (2025), Energy and AI. https://www.iea.org/reports/energy-and-ai Clean Energy and Water Technologies Pty Ltd | Carbon Recycling Technology — Integrated Energy Systems for a Sustainable Future