The rapid spread of artificial intelligence is having a measurable impact on electricity consumption, especially inside data centers. Nagy‑Gál Levente, sustainability and energy efficiency advisor at Schneider Electric, told the Nagy AI‑sztori podcast that AI’s power needs must be considered together with the future of buildings, industrial sites, data centers and the power grid. The podcast’s regular expert guest was Aczél Petra, communication researcher and professor at Széchenyi István University.
Scale in numbers
The conversation included concrete figures to illustrate the scale. Training the GPT‑3 model required roughly 1.3 gigawatt‑hours (GWh) of electricity. According to the episode, that energy use corresponds to about 550 tonnes of carbon dioxide emissions, or roughly the annual operation of 123 gasoline‑powered passenger cars.
Differences at data centers are also striking: a conventional cloud provider rack typically consumes about 20 kilowatts (kW), while an AI workload running continuously can demand up to ten times that power — roughly 200 kW. Nagy‑Gál used an image that highlights the cooling and density problem: it is like having 200 hairdryers roaring inside a two‑door cabinet. Such loads require not only supply of electricity but also efficient cooling and distribution.
Cooling, distribution and infrastructure become strategic
As AI loads grow, cooling, power distribution and infrastructure resilience become strategic priorities for data centers. Producing more energy is not enough: it must be delivered to the consumers, the increased heat load must be managed, and continuous availability ensured. If AI adoption is exponential, the energy system must be upgraded to keep pace.
AI as a tool to improve energy efficiency
AI is not only a large new consumer; it can also process vast amounts of data to reduce waste and optimize consumption. In buildings, for example, heating, ventilation and air conditioning, environmental conditions, occupancy data, weather impacts and individual equipment consumption all interact. Humans cannot continuously integrate all these signals at scale, but predictive AI systems can optimize control.
Smart buildings use building management systems as the ‘‘brain’’ to control ventilation, heating, cooling or blinds, and when occupancy data are incorporated the building behaves differently on an empty day versus a busy workday. This goes beyond simple automation toward data‑driven energy optimization.
Grid upgrades are unavoidable
During the coming years, upgrading and smarter control of the power grid will be essential. Adding renewable or nuclear capacity alone is insufficient without a suitable ‘‘road network’’ to deliver electricity to end users. Data centers, industrial electrification, household microgeneration and distributed renewables form an increasingly complex system that requires continuous monitoring and intelligent control.
Timely action turns efficiency into strategy
Nagy‑Gál stressed that energy efficiency can appear to be a convenience until problems arise; once they do, it rapidly becomes a strategic issue. Companies and data‑center operators face a dual challenge: they must handle rising direct power demand from AI while also deploying AI‑based solutions to reduce overall consumption.
Topics covered in the podcast (with timestamps)
- What is the energy demand of the AI transition? (01:09)
- What does energy mean from Schneider Electric’s perspective? (03:16)
- How can predictive AI help optimize building energy use? (07:05)
- Why does AI create new challenges for data center power and cooling? (09:42)
- What does it mean for a building to become “smart”? (14:42)
- Why did the energy crisis accelerate corporate thinking about efficiency? (18:59)
- Why could developing and controlling the Hungarian power grid be critical in the coming years? (29:58)
The episode underlines that AI technologies simultaneously create new electricity demand and provide tools to reduce waste—managing both requires comprehensive planning, infrastructure investment and intelligent control.
