AI Is Consuming More Electricity. Can the Power Grid Keep Up?

AI applications are becoming more powerful, but their growth depends on something that is often overlooked: physical computing infrastructure.

Every AI-powered application relies on hardware, data centres, networking, storage and cooling systems. As more users interact with AI models and businesses integrate them into their workflows, the infrastructure behind those services must scale.

That creates an important engineering and energy challenge: how can we scale AI computing without putting excessive pressure on electricity infrastructure?

## 1. AI workloads are increasing data centre demand

AI data centres support two major categories of computing:

• Training: Processing large datasets to develop or improve AI models.
• Inference: Running trained models to generate responses, analyse information, create content or perform other tasks.

Both require computing resources, but their workloads differ. Training can involve intensive computing over extended periods, while inference demand can increase as more users and applications adopt AI.

The International Energy Agency (IEA) projects that global data centre electricity consumption could rise from approximately 485 TWh in 2025 to 950 TWh in 2030.

It also projects that electricity consumption from AI-focused data centres will roughly triple between 2025 and 2030.

These figures should be interpreted carefully: the 485–950 TWh projection covers all data centres, not just AI facilities.

2. Why scaling AI is also an infrastructure problem

Adding more GPUs and servers does not automatically mean a data centre can operate at full capacity.

Large facilities need electricity infrastructure capable of supporting their power requirements, including:

• High-capacity electrical connections
• Transformers and substations
• Transmission and distribution networks
• Backup power systems
• Cooling infrastructure

A region may have sufficient electricity generation but still lack the local transmission or distribution capacity required by a new data centre.

This makes location planning, grid connections and infrastructure investment important parts of AI expansion.

3. Cooling is a major part of data centre engineering

High-performance processors generate heat. As computing density increases, removing that heat becomes a critical design requirement.

Data centres use several cooling approaches:

• Air-based cooling
• Liquid cooling
• Direct-to-chip cooling
• Advanced airflow and temperature management

Some cooling systems also consume water. Actual water requirements vary according to the cooling design, climate and facility operations.

For data centre engineers, the challenge is to balance computing performance, energy efficiency, equipment reliability and responsible water use.

4. Will data centre expansion affect electricity bills?

The impact depends on local electricity markets, infrastructure investment and regulatory arrangements.

If the cost of expanding electricity infrastructure is passed on to existing customers, households could face additional costs. Some large data centre projects, however, include specific agreements about infrastructure expenses and demand management.

For that reason, the financial and regulatory structure of each project matters. It is not accurate to assume that every data centre will affect household bills in the same way.

5. OpenAI's Grid Protectors Program introduces flexible demand

On November 2, 2026, OpenAI announced its Grid Protectors Program.

The initiative aims to help future computing campuses support electricity-grid reliability.

One important concept is demand flexibility: a facility could reduce its electricity consumption during periods when the grid is under pressure.

Potential approaches include:

• Reducing or shifting suitable computing workloads
• Using on-site power generation
• Deploying battery storage
• Improving power-management controls
• Coordinating with utilities and grid operators

This approach is important because data centres are not just electricity consumers. With suitable infrastructure and operating policies, they may also be able to provide flexibility to the power system.

The actual results will depend on implementation, facility design and the amount of demand that can be reduced without disrupting essential workloads.

A related example: OpenAI and Georgia Power

An earlier agreement involving OpenAI's planned Effingham County, Georgia, data centre provides a more specific example.

Announced on July 22, 2026, the project is expected to require approximately 3,200 MW of power in phases.

The agreement described by Georgia Power includes up to 1,000 MW of flexible demand response, allowing the facility's electricity demand to be reduced during certain periods.

OpenAI also stated that it would pay the infrastructure and electric-service costs required for the project. Georgia Power announced regulatory approval of the contract on August 26, 2026.

This project-specific agreement should be distinguished from the broader Grid Protectors Program announced in November 2026.

6. Renewable energy needs supporting infrastructure

Solar and wind power can help supply cleaner electricity to data centres.

However, renewable generation varies with weather and time of day. Data centres generally require reliable electricity throughout their operating hours.

Supporting technologies and infrastructure include:

• Battery energy storage
• Improved transmission capacity
• Flexible electricity demand
• A diverse energy mix
• Better coordination between utilities and large electricity users

Renewable energy procurement is useful, but companies should also explain how their commitments translate into additional clean generation and how electricity supply is matched with actual demand.

7. Efficiency improvements can reduce computing requirements

There are several opportunities to improve the efficiency of AI infrastructure.

Approach

• Efficient AI processors
• Smaller specialised models
• Optimised model architectures
• Improved cooling
• Workload scheduling
• Battery storage

Potential benefit

• More computing performance per unit of electricity
• Less computing for suitable tasks
• Reduced processing requirements
• Lower energy used for heat management
• More flexible electricity consumption
• Better management of electricity demand

Efficiency does not necessarily mean total electricity consumption will decline. Lower computing costs can encourage more AI usage, potentially offsetting some efficiency gains.

This is why efficiency improvements need to be evaluated alongside overall demand growth.

As AI technology continues to advance, its capabilities are expanding beyond traditional chatbots and text generation. Google's Genie 4 is an example of this shift towards more interactive and immersive digital experiences. Read our detailed article: Google Genie 4 Explained: Is AI Moving Towards Interactive Worlds?.

8. What should India consider?

India's expanding digital economy creates opportunities for AI infrastructure investment.

However, future data centre projects will need careful planning around:

• Electricity generation and grid capacity
• Transmission and distribution infrastructure
• Water availability
• Cooling technology
• Renewable energy integration
• Long-term operating costs

A sustainable approach requires more than installing powerful computing hardware. It requires coordination between technology companies, electricity providers, policymakers and local communities.

Conclusion: AI needs an energy strategy, not just better chips

The growth of AI is changing the requirements of digital infrastructure.

More powerful models and larger workloads require computing capacity, but that capacity depends on electricity, cooling, reliable grid connections and responsible resource management.

Efficient processors, specialised models, renewable energy, storage and flexible demand can all help.

The next phase of AI development will therefore involve more than software engineering and model performance. It will also require better energy planning and infrastructure engineering.

The future of AI will depend on how efficiently we can power it.

References

• IEA – Key Questions on Energy and AI
• IEA – Artificial Intelligence and Energy
• OpenAI – Grid Protectors Program
• OpenAI – Effingham County AI Infrastructure
• Georgia Power – OpenAI Effingham County Project
• Georgia Power – Contract Approval

This article is for informational purposes. Energy projections are estimates and may change as technology and infrastructure develop.