AI Energy Consumption Forecast to Dominate Data Center Electricity by 2028

Artificial intelligence has quietly become one of the fastest-growing consumers of electricity worldwide. According to recent forecasts from Lawrence Berkeley National Laboratory, AI is projected to consume over half of all data center electricity by 2028—a dramatic shift that transforms AI energy consumption from a niche technical concern into a major infrastructure and policy challenge.

This transformation reflects the broader integration of AI into everyday products and services. As AI inference scales across applications from chatbots to autonomous vehicles, understanding where and how energy is consumed has become crucial for utilities, policymakers, and technology companies alike.

The Growing Scale of AI Energy Consumption

Current research from multiple authoritative sources paints a concerning picture of AI’s energy trajectory. The MIT Technology Review’s 2025 analysis reveals that AI-specific servers within US data centers consumed an estimated 53-76 terawatt-hours (TWh) in 2024, with projections suggesting this could rise to 165-326 TWh by 2028.

Perhaps most significantly, inference—the process of running trained AI models to generate responses—now accounts for 80-90% of AI computing energy usage. This shift represents a fundamental change from the early days of AI development, when model training was the primary energy consumer.

Breaking Down AI Energy Usage

The energy intensity of individual AI tasks varies dramatically based on model size and complexity:

• Text generation: Simple text responses using smaller models like Llama 3.1 8B consume approximately 114 joules per response, while larger models like Llama 3.1 405B require about 6,706 joules per response
• Image generation: Creating a 1024×1024 image with Stable Diffusion 3 Medium uses approximately 2,282 joules, with higher-quality settings consuming up to 4,402 joules
• Video generation: Low-quality short video outputs require around 109,000 joules, while higher-quality 5-second videos can consume up to 3.4 million joules

Global Context and Projections

The International Energy Agency (IEA) provides crucial context for understanding AI’s energy impact within the broader data center ecosystem. Their 2025 analysis estimates that global data centers consumed approximately 415 TWh in 2024, representing about 1.5% of global electricity consumption. Under their base case scenario, this could reach 945 TWh by 2030—nearly 3% of global electricity consumption.

Currently, AI accounts for roughly 5-15% of data center power usage, but this share could grow to 35-50% by 2030. This rapid growth creates significant regional concentration effects, with countries like Ireland seeing data centers consume 21% of national electricity, potentially rising to 32% by 2026.

The Transparency Challenge

One of the most significant obstacles to managing AI energy consumption is the lack of transparency from major technology companies. MIT Technology Review highlights a critical information gap: most major “closed” AI model providers do not disclose sufficient information to reliably estimate their total energy use or carbon footprint.

This opacity makes it difficult for researchers, policymakers, and even customers to understand the environmental impact of AI services they use daily. Without standardized measurement and reporting, comparing the efficiency of different AI models or platforms becomes nearly impossible.

Efficiency Improvements and Best Practices

Despite the concerning growth projections, there are promising developments in AI energy efficiency. Google Cloud’s 2025 methodology for measuring environmental impact shows significant improvements, with energy per median prompt falling by 33 times over a 12-month period, and total carbon footprint decreasing by 44 times.

Key efficiency strategies include:

• Improving algorithm efficiency through better model architectures and inference techniques
• Increasing utilization of graphics processing units and other computer systems to reduce idle compute time
• Using demand-aware model placement to keep AI inference closer to users or on lower-carbon parts of the power grid
• Optimizing serving systems to reduce idle capacity in hyperscale data centers

Implications for Infrastructure and Policy

The rapid growth of AI energy consumption creates several critical challenges for infrastructure planning and environmental policy. Unlike data centers, which can be built in a few years, electricity infrastructure takes much longer to develop. This mismatch requires proactive coordination between new data center development and long-term grid planning.

Data center electricity carbon intensity is currently 48% higher than the US average, highlighting the need for cleaner energy sources. The IEA estimates that nearly 60% of the global power supply mix for data centers comes from fossil fuels, underscoring the importance of accelerating procurement of carbon-free energy sources.

Water Usage and Environmental Impact

Beyond electricity consumption, AI data centers present significant water usage challenges. Cooling systems, essential for maintaining optimal operating temperatures, can consume large amounts of water—sometimes potable water in water-stressed regions. This creates additional environmental considerations beyond carbon emissions.

Responsible water stewardship programs, including recycling and replenishment commitments, become crucial as AI infrastructure expands globally.

The Path Forward: Recommendations for Sustainable AI

Addressing AI’s growing energy footprint requires coordinated action across multiple stakeholders. Key recommendations include:

1. Treat AI as a trackable energy sector: Define AI workloads explicitly within data center energy reporting and develop standardized metrics for AI-specific energy use
2. Improve measurement and transparency: Require tech companies to disclose energy usage and carbon footprint of AI-based products
3. Plan power supply ahead of demand: Coordinate new data centers with long-term grid planning to avoid relying on fossil fuels
4. Improve electricity mix quality: Track actual grid energy mix and accelerate procurement of renewable energy sources
5. Integrate AI energy policies into broader regulation: Include AI-specific requirements in existing environmental and energy regulations

Conclusion

The projection that AI will consume over half of data center electricity by 2028 represents a watershed moment for the technology industry and environmental policy. As AI becomes increasingly integrated into daily life, its energy consumption is no longer a marginal concern but a central challenge for sustainable development.

The path forward requires unprecedented cooperation between technology companies, utilities, policymakers, and researchers. By implementing standardized measurement practices, improving efficiency, and transitioning to cleaner energy sources, the AI industry can continue to innovate while minimizing its environmental impact. The decisions made today about AI energy consumption will shape not only the future of technology but also the trajectory of global efforts to address climate change.

References

Source: AI Energy Consumption: Statistics from Key Sources [2026] – AI Multiple Research