Why Businesses Are Asking How Much Energy AI Really Uses

Ofgem notes that around 140 proposed AI and data centre projects have requested nearly 50 gigawatts of grid connection capacity. This already exceeds the current national peak of about 45 gigawatts.

This scale of investment is seen worldwide. MIT also highlights moves such as Meta and Microsoft pursuing nuclear power options, OpenAI and partners planning five gigawatt Stargate class centres, and Google budgeting about 75 billion dollar on AI infrastructure in 2025.

Electricity, carbon, and grid impacts have therefore become mainstream business issues whenever AI adoption is discussed.

Quick answer: should we worry about how much energy ChatGPT uses?

In isolation, no. A single ChatGPT query uses only a small amount of electricity. However, at scale and across an organisation, the cumulative energy use becomes worth paying attention to.

Every ChatGPT query triggers computation on specialised chips that produce heat and require cooling, all of which consumes a significant amount of energy. However, there is no single figure that applies to every query. Closed models (AI models like ChatGPT that do not publish details on hardware loads or routing) note that energy use varies with prompt length, model size, data centre location and time of day.

A typical ChatGPT query uses more energy than a basic web search but far less than streaming HD video. The exact amount depends on the efficiency of the data centre and the performance of the hardware serving the request.

How ChatGPT works

Where the energy from ChatGPT goes

The energy behind Large Language Models (LLMs) comes from two very different processes. Training requires weeks of concentrated computation across thousands of GPUs or TPUs and is extremely energy intensive. “Inference” is the process that runs when someone types a query. Although smaller in cost than training, inference still uses power dense chips that draw significant electricity and need substantial cooling.

According to the International Energy Agency, severs account for about 60% of electricity use in a modern data centre, while cooling can range from about 7% in highly optimised hyperscale facilities to more than 30% in less efficient sites.

Compared with pre-AI cloud computing, AI systems rely on dense clusters of accelerators that often draw power in short bursts. This creates new challenges for grid infrastructure and electricity networks.

Where ChatGPT’s energy comes from: data centres, clouds, electricity mix

AI runs in physical data centres, mainly owned by large cloud providers. These facilities use a combination of grid electricity and renewable sources. Data centres often operate on grids with higher carbo intensity than national averages because they require continuous power and are usually located where land and connectivity are available.

Businesses will recognise several familiar pressures. Electricity prices, network charges, and wider energy market conditions all shape how data centres manage their consumption and plan new capacity. As AI workloads increase, these factors influence the overall cost and carbon profile of the platforms businesses rely on.

How much energy a single ChatGPT query uses

Recent analysis from Epoch AI estimates that a typical GPT 4o text query uses roughly 0.3 watt hours or about 1,080 joules. This is about ten times lower than earlier 2023 estimates and similar to the electricity used by a 10-watt LED bulb running for around two minutes. Sam Altman, founder and CEO of OpenAI, has also stated that modern ChatGPT queries use about 0.3 watt hours, similar to a web search from earlier eras

Energy use varies by model size, prompt complexity, output length, hardware, and data centre efficiency. Expert comparisons show that generating a single AI image can require more than 2,200 joules, and producing short AI video clips can reach millions of joules depending on complexity and duration.

For a business example, if someone ran 15 text queries, generated 10 images, and created three five second videos, the total energy use could reach a few kilowatt hours. This is small on its own but scales quickly across an organisation, just as small tariff differences add up across a large site portfolio.

Does ChatGPT increase greenhouse gas emissions?

The carbon impact of ChatGPT depends entirely on the electricity powering the data centre. Cleaner grids reduce emissions and fossil heavy grids increase them. In the United States, data centres draw about 56% of their electricity from fossil fuels and have a carbon intensity about 48% above the national average.

In the UK, the Carbon Price Support will remain equivalent to 18 pounds per tonne of CO2 in 2026 to 2027. This keeps pressure on fossil fuel generators and influences electricity prices for high demand facilities such as data centres.

Why AI energy use matters for businesses

Electricity cost exposure is rising as data centre demand increases. Ofgem warns that proposed AI and data centre projects could require grid capacity greater than current national peak demand.

For businesses, this matters because any increase in network strain can influence future commercial tariffs, connection times, and local grid investment. As more digital services depend on power hungry infrastructure, energy pricing and availability become directly tied to the cost and reliability of the AI platforms organisations rely on.

Understanding this shift helps businesses make better choices about where and how they use AI, particularly when reviewing contracts, forecasting costs, or planning long term sustainability goals.

Will AI become more efficient or more energy intensive?

AI hardware, cooling systems, and data centre design are becoming more efficient each year. New chip architectures, liquid cooling, and renewable power procurement all contribute to improved performance per watt.

However, demand is rising quickly. Experts predict that AI agents, voice interfaces, reasoning models, and video generation are increasing the amount of computation required per user. As a result, total electricity demand from AI may continue to grow even as individual chips become more efficient.

Policy and energy market conditions will continue to shape how and where data centres expand, but predicting precise future costs or regulations is not possible.

Conclusion: what businesses should understand about the rise of AI

A single ChatGPT query uses a modest amount of electricity, roughly 0.3 watt hours according to recent analysis. At scale, however, the cumulative impact becomes material, especially across large organisations.

Businesses should approach AI use as part of their broader sustainability, procurement, and digital strategy. Choose platforms that offer any available information about their data centre efficiency, energy sources, or carbon footprint. Even high-level transparency helps businesses understand the environmental impact of the AI tools they use.

With UK data centre electricity demand forecast to grow fivefold by 2030, AI energy considerations are now part of routine corporate planning rather than a niche technical issue.