Artificial intelligence (AI) is scaling rapidly across industries, embedded in everything from content generation to predictive maintenance, autonomous vehicles to smart buildings. But beneath the surface lies an urgent, often overlooked, reality: there is no AI without energy.
According to the International Energy Agency’s special report, Energy and AI, AI is emerging as a “general-purpose technology” that will soon become as central as electricity itself. Yet its deployment is creating a surge in electricity demand, infrastructure requirements, and systemic vulnerabilities.
Some key themes run through the report:
1. AI is driving exponential growth in electricity demand
AI adoption is already putting significant pressure on global energy systems. In 2024, data centres consumed around 415 TWh of electricity, accounting for 1.5% of global demand. That number is expected to more than double to 945 TWh by 2030, and rise to 1,200 TWh by 2035 in the IEA’s Base Case — an amount comparable to Japan’s entire electricity consumption today .
What’s driving this?
- AI workloads are the largest single factor. Accelerated servers (GPUs, TPUs, etc.) needed for model training and inference are growing at 30% annually.
- The largest AI data centres under construction will consume as much electricity as 2 million households .
- In the United States alone, data centres are projected to represent nearly half of all electricity demand growth to 2030, surpassing even the steel and cement industries combined .
This scale of demand requires not just digital planning but energy foresight. Organisations deploying or relying on large-scale AI must now consider themselves part of the energy sector, and plan accordingly.
2. New energy infrastructure will determine AI’s speed and scale
The IEA projects that powering AI will require a rapid, multi-source expansion of the global electricity supply. By 2035:
- Renewables will supply around 50% of the added demand (450 TWh).
- Natural gas will expand by 175 TWh, largely in the US.
- Nuclear will contribute a similar share, especially in China and the US, including small modular reactors (SMRs) expected from 2030.
- Geothermal and battery storage will play growing niche roles.
- In the US, investment in data centre-related energy infrastructure will require over $480 billion between 2025 and 2030 – a massive new capital requirement .
Yet clean energy matching remains a challenge. Many technology firms procure renewable energy via Power Purchase Agreements (PPAs), but these often rely on annual matching, which doesn’t ensure real-time clean power use.
- A data centre with “100% annual solar matching” may still draw fossil power for 55–65% of its operating hours without onsite storage.
- Real-time (hourly) matching is still rare and expensive – adding up to $200/MWh to costs .
Clean supply isn’t just about buying renewables. Real resilience and emissions compliance require hourly-matched, low-carbon electricity, proactive siting, and collaboration with grid operators.
3. Grid bottlenecks threaten AI rollout
Grid congestion is becoming a critical barrier to AI expansion:
- Around 20% of planned data centre projects globally are at risk of delay due to electricity connection issues.
- Lead times for transformers, turbines and high-voltage cables have doubled in the last three years.
- In advanced economies, building a new transmission line can take 4–8 years.
- Half of new US data centres are planned in already congested regional clusters, increasing the risk of localised grid instability .
Meanwhile, regulators are only beginning to incorporate data centre energy loads into long-term planning – and are often underestimating the urgency.
4. AI can optimise – and decarbonise – the energy system
While AI demands a lot of energy, it can also improve energy system efficiency, offering powerful opportunities for net emissions reduction. The IEA finds that broad deployment of current AI tools could reduce global energy sector emissions by 1.4 gigatonnes (Gt) of CO₂ annually by 2035, roughly 5% of total energy emissions .
Key areas of impact:
- Oil and gas: AI is reducing exploration errors, preventing methane leaks, and cutting downtime — translating to lower upstream emissions.
- Electricity grids: AI-driven forecasting can cut renewable energy curtailment and improve load balancing.
- Transmission: AI-powered grid monitoring can unlock 175 GW of transmission capacity without building new lines — more than the total data centre load growth to 2030.
- Buildings: Smart AI controls for HVAC and lighting could reduce electricity use by 300 TWh, equal to all of Australia and New Zealand’s annual generation.
- Industry: AI process optimisation could reduce global industrial energy use by more than Mexico’s total energy demand.
Leaders should be considering how to adopt AI tools not only to enhance operational efficiency, but to offset AI’s own emissions footprint. For energy-intensive sectors, this becomes a compliance and cost-saving strategy.
5. Rebound effects and security risks complicate the picture
AI is a double-edged sword. Its gains can be undone by unintended consequences:
Rebound effects
- Autonomous vehicles may reduce emissions per mile but increase total mileage.
- Lower production costs from AI-enhanced oil extraction could reduce prices and increase consumption — a $10 drop in oil prices could add emissions equivalent to 20 million new cars .
Cybersecurity
- Cyberattacks on energy systems have tripled since 2019, with AI making them faster and more sophisticated.
- AI is also improving cyberdefence, including anomaly detection and adaptive response systems .
Mineral supply chain vulnerabilities
• Data centre chips rely on rare materials like gallium, of which 99% of refined supply currently comes from China. Gallium demand from AI could reach 10% of today’s global supply by 2030 .
Water use
• AI data centres use significant water for cooling – up to 1.5 litres per kWh – raising concerns in water-stressed regions.
Therefore, AI deployment must include impact forecasting, not just benefit modelling. Plan for rebound effects, secure your supply chains, and engage early in water and cyber risk mitigation.
6. Emerging economies risk falling behind – but could leap ahead
Emerging and developing economies (EMDEs) represent half of the world’s internet users but host less than 10% of global data centre capacity .
Barriers include:
- Unreliable power (up to 700 hours/year of outages in some regions)
- Limited digital infrastructure
- Low AI literacy in firms and governments
However, EMDEs also have opportunities to leapfrog legacy systems. Examples already emerging include:
- India: Smart building energy systems
- Morocco: AI for industrial process optimisation
- Sub-Saharan Africa: AI-based grid loss detection and fraud prevention.
AI leaders with operations in EMDEs should invest in local energy reliability, support digital infrastructure, and explore AI-based localisation strategies for both service delivery and innovation.
7. The AI talent gap in energy is growing
Despite its transformative potential, the energy sector is struggling to attract AI talent:
- AI expertise in utilities and oil & gas is 40% lower than in finance, tech or healthcare.
- Salaries for AI roles in energy are up to 30% lower than in tech, contributing to a talent drain.
- Most firms lack internal digital capacity to assess, procure and integrate AI systems .
The IEA highlights that missing expertise is the single most cited reason for firms not adopting AI, surpassing concerns over cost, utility, or ethics .
Digital capacity is now a core operational competency, not a peripheral skill. Leaders must invest in workforce development, cross-sector hiring, and public-private partnerships to close the gap.
Strategic priorities for AI leaders
In light of these findings, executive teams implementing or scaling AI should:
1. Treat energy planning as a core AI requirement – secure long-term power supply, including renewables and flexible load options.
2. Engage with regulators and utilities now – avoid delays and align energy supply with organisational AI ambitions.
3. Shift to hourly-matched clean energy contracts – annual offsets will not be sufficient for compliance or reputational resilience.
4. Use AI to decarbonise your broader operations- apply in manufacturing, logistics, buildings, and grid interaction.
5. Develop AI capabilities internally – build digital fluency in leadership and operational teams.
6. Incorporate AI into enterprise risk management – plan for cyber threats, mineral dependencies, water use, and rebound effects.
7. Advance equity and localisation in emerging markets – build energy reliability and talent pipelines in regions with long-term growth.
AI’s rise will not be determined solely by model performance or compute power. It will be shaped by how organisations navigate the energy, infrastructure, security and human capital landscapes around it. The IEA’s Energy and AI report makes one thing clear: AI cannot scale without energy – and energy will be reshaped by AI.