Demand Response

In an electricity grid, exactly as much power must be generated every second as is currently being consumed. Normally, generation is adjusted to match this: when everyone is cooking in the evening, additional power plants start up. Demand response turns this idea around. Here, consumption adapts to generation. Large electricity consumers therefore switch off briefly or throttle their output when there is scarcity on the grid. In return, they are paid by the grid operator or by an electricity trader.

Why data centers and factories have suddenly become electricity traders

Solar and wind installations do not generate electricity on command. At midday, when the sun is shining, there is a great deal of it on the grid; on a windless winter evening, there is almost none. In the past, this was balanced out with gas and coal power plants that were ramped up and down as needed. The larger the share of renewable energy becomes, the more expensive and rarer these balancing power plants become. Flexible consumers are the cheaper alternative.

For companies, this is a business. An aluminum smelter, a cold-storage facility, or a cement mill can often postpone its operations by a few hours without suffering any harm. Whoever offers this flexibility earns money without producing anything. In Germany, this runs via so-called balancing power markets, where grid operators purchase such commitments.

The topic has become especially exciting because of AI. Data centers for AI training consume enormous amounts of electricity and are often located in regions where the grid is already at its limit. However, a training run can be postponed by two hours more easily than an intensive care unit. Large operators are therefore negotiating connections under which they give up power in an emergency – and in return get connected to the grid faster.

From price signal to switched-off cooling

Technically, demand response requires three things: a signal, a control system, and billing. The signal comes from the grid operator or directly from the exchange price for electricity. If the price rises sharply, electricity is scarce. If it falls, there is too much of it.

The control system sits with the consumer. Software continuously checks which systems can be throttled without problems and switches them automatically. A cold-storage facility, for example, is well insulated. It can go without a compressor for half an hour, because the temperature only rises by fractions of a degree. The cold stored in the warehouse is thus used like a battery.

Because individual installations are too small for the electricity market, service providers bundle many of them together. This aggregation is called a virtual power plant: thousands of heat pumps, charging stations, and machines behave, from the outside, like a single, controllable power plant. It is important to distinguish this from saving electricity. Demand response shifts consumption, it usually does not reduce it. The laundry still gets washed, just later.

Dynamic tariffs, wallboxes, and headlines about grid stability

Private customers encounter this through dynamic electricity tariffs. Here, one pays the current exchange price by the hour instead of a fixed rate in cents. Anyone who charges their electric car at three in the morning can pay significantly less than in the early evening. This requires a modern meter that records consumption in fifteen-minute intervals.

In the news, demand response is often mentioned in connection with the expansion of heat pumps and charging stations. Since 2024, grid operators in Germany have been allowed to briefly limit the output of such devices in an emergency instead of refusing the connection. Critics call this forced shutdown; technically, it is a mild form of load control.

For investors, this field is interesting because software meets energy here. Providers of virtual power plants, manufacturers of smart meters, and operators of data centers with flexible contracts benefit from this. A common misconception is that demand response replaces storage and grid expansion. It makes both cheaper, but not unnecessary.