Hanna Oredsson: Sweden needs a new energy measure for AI

As data centers increasingly demand more power, it’s not enough to just measure how efficiently electricity is used – we also need to measure what it creates. That’s why the energy debate needs to be complemented with new productivity metrics that show how much AI benefit we get per watt. Now, Hanna Oredsson, sales manager Secure Power at Schneider Electric Sweden, that the focus must shift from energy consumption to token-based value creation.

Sweden has good access to fossil-free electricity, a cold climate and stable conditions that make us attractive for data center establishment. At the same time, competition for electricity is increasing rapidly, as industry and transport are electrified and digitalization accelerates. When electricity is expensive and in demand, the crucial question becomes what we actually get out of each kilowatt hour.

Data centers require stable power 24/7, which places special demands on the electricity grid and makes their impact greater than many other businesses. A new report from the Confederation of Swedish Enterprise highlights this as a key challenge: data centers are seen as both an enabler of innovation and a growing burden on the electricity system. For Sweden to become a leading AI nation, we need to get a better handle on the actual energy use of data centers.

Swedish data centers are among the most energy-efficient in the world and are often measured by PUE (power usage effectiveness), which shows how efficiently a data center delivers energy to IT equipment in relation to total energy use. However, this metric says nothing about what the data center actually generates.

When individual AI facilities can require as much power as a small city, it’s not enough to know that energy is being used efficiently – we need to understand which benefit which is created per kilowatt hour.

Here the concept becomes “tokens per watt” relevant. AI models process data in small units called tokens, which are both the technical building blocks and the basis for how AI services are priced. By measuring how many tokens are generated per watt used, we get a concrete measure of productivity. It links energy use to actual output and makes it possible to compare different data centers, different AI applications, and vastly different ways of using electricity. Above all, it gives us a productivity metric to discuss value.

The Swedish energy debate has already started to move in that direction. Municipalities, authorities and businesses are increasingly asking the question of what we get in return for the megawatt. At the same time, it is not just a question of competition for electricity. Data centers can also contribute to the energy system, for example by delivering waste heat to district heating networks, acting as a stable and predictable electricity user and in some cases helping to balance the load on the electricity grid. The report from the Confederation of Swedish Enterprise highlights the need to better understand how electricity is used and what value different investments create.

When we can link AI use to energy consumption, it also becomes possible to calculate the economy and climate impact in a more concrete way. How much does an AI service cost in energy? How much value is created per unit of emissions? Which applications actually contribute to societal benefit? This is crucial in a situation where data centers not only affect the energy system but also Sweden's competitiveness.

As data and computing power become strategic resources, control over digital infrastructure becomes increasingly important. Where data centers are established ultimately affects innovation, security and digital sovereignty, which is a central aspect of Sweden's long-term competitiveness and preparedness.

Tokens per watt does not replace traditional metrics like PUE, but it complements them by showing what we actually get in return for the energy. In an era where both sustainability and energy distribution are high on the agenda, this can become an important metric. Not only for technology companies, but for politicians, authorities and investors.

Data centers are no longer just server rooms. They are factories for digital value creation. And just like any other factory, they must be measured by their productivity.

About Schneider Electric

Schneider Electric is a global leader in energy technology, driving efficiency and sustainability by electrifying, automating and digitizing industries, businesses and homes. The company's technology enables buildings, data centres, factories, infrastructure and power grids to function as open and interconnected ecosystems, strengthening performance, resilience and sustainability. Its portfolio includes intelligent devices, software-defined architectures, AI-powered systems, digital services and expert consulting. With 160,000 employees and a network of 1 million partners in over 100 countries, it is ranked Schneider Electric recurring as one of the world's most sustainable companies.