AI with Care

AI has become part of everyday work at DANREC, supporting our internal HELIX systems, AI assistants, reporting tools and daily office tasks. Alongside these opportunities comes a responsibility to understand the resources the technology requires.

Developing and using AI requires servers that consume electricity and need cooling. Depending on the data centre’s cooling technology and electricity supply, this also involves water consumption and CO₂ emissions. The European Commission identifies these as significant environmental impacts of data centres.

In its 2025 report Energy and AI, the International Energy Agency projected that global electricity consumption by data centres would more than double between 2024 and 2030, with AI as the largest driver of growth. This covers data centres overall, including services other than AI.

Understanding Our Own Consumption

At DANREC, our IT specialist, Paw, has estimated the impact of our current AI use and explored what running AI on our own equipment could mean.

Today, we primarily use services from OpenAI and Anthropic, the company behind Claude. Our internal systems are hosted locally, while AI processing takes place in external data centres through API connections.

Paw’s calculations estimate that our combined use—covering seven office employees and our internal AI systems—accounts for approximately 73–74 kWh of electricity, 9–30 kg of CO₂ emissions and 80 litres of water annually for external AI processing.

These figures provide an initial basis for assessing our choices. They are estimates of a defined part of our AI use, rather than a measurement of its complete environmental footprint.

Would Local AI Be a Better Choice?

Paw has also examined a possible investment in an NVIDIA DGX Spark, which would allow compatible AI models to run locally.

The estimated annual electricity consumption ranges from approximately 193 kWh if left idle around the clock to 2,100 kWh at continuous maximum load. Actual consumption would depend on operating hours and workload. Under the assumptions used, a dedicated local computer would therefore likely increase electricity consumption at our current usage levels.<br/>

There are other reasons to consider local AI. It could allow sensitive information to be processed on site, provide greater control over models and development, and reduce our dependence on external providers, their pricing and service changes.

At our current usage levels, an investment in local AI equipment would therefore primarily support digital independence, data ownership and technological control, rather than deliver an immediate environmental improvement.