Energy-intensive AI data centers: ‘Connecting to high voltage’

AI helps society move forward, but it has a downside: energy consumption is growing rapidly. So there is work to be done. The power grid, which is significantly overloaded, needs to be addressed, and data centers themselves need to become more sustainable. Stijn Grove, Managing Director at the Dutch Data Center Association, the trade association for data centers in the Netherlands: “We are going to connect data centers to the high-voltage grid.” He is one of the speakers at the Watt Matters in AI conference.

In the Netherlands, data centers account for approximately 4% of total energy consumption, up from 1.5% in 2017. This trend shows no signs of slowing down in the coming years. According to a report by the International Energy Agency (IEA), global demand for electricity from data centers will more than double by 2030 to approximately 945 terawatt hours (TWh), which is more than Japan's total electricity consumption at present.

A major reason for this is that we are all increasingly using digitization and AI. And AI consumes a lot of power because it has to process enormous amounts of data, perform heavy calculations, and run powerful servers for long periods of time. Grove: “There is no doubt that this is an issue that needs to be addressed. The pressure on the energy grid is only increasing. It is crucial to look at how we can make changes to the power grid and in data centers themselves.”

Relieving the strain on the power grid

First, there is the overloaded power grid. The problem is that we are dealing with fluctuating renewable energy. Solar panels and wind farms do not supply power constantly, which causes peaks and troughs in supply. At the same time, consumption is increasing. “It used to be simple: one power plant supplying electricity to homes. Clear and predictable. That is no longer the case. The current situation calls for a new, smart network that can respond flexibly, so that we can avoid congestion and use energy more efficiently,” explains Grove.

A promising approach to relieve the grid is to use the high-voltage grid. Most data centers are now connected to the medium-voltage grid, where congestion is common. “Connecting data centers directly to the high-voltage grid relieves the medium-voltage grid and frees up more space for homes and other businesses. The challenge, however, is that the data center must be large enough to make this feasible, because it is more expensive.”

An example of this can be found in Groningen. On behalf of grid operator TenneT, a 220 kV underground cable is being laid down that connects a data center directly to a high-voltage substation. This connection makes it possible to supply large amounts of electricity efficiently. The work is expected to be completed by the end of the year.

Another solution is to install larger batteries at data centers. “Instead of switching on the emergency power supply with large diesel generators at every peak on the power grid, as was always the case, batteries can offer a solution,” says Grove. With the falling prices of batteries, there are more and more opportunities to use them in a scalable way. This not only reduces CO₂ emissions, but also provides flexibility to run AI and cloud applications reliably.

Greening data centers: progress

So there are plenty of opportunities to tackle the power grid, but data centers themselves also have an important role to play: they can often significantly reduce their energy consumption. According to Grove, one of the measures is to make greater use of cloud solutions. He explains: "In the past, every company had its own servers, which were often only partially utilized. By moving these to shared cloud servers, multiple companies can efficiently use a single set of servers. One company calculates at night, another during the day. This means that ten companies can often be served by two servers instead of ten, resulting in enormous energy savings." Economies of scale and smart management significantly reduce the total load on the network.

A second way in which data centers save energy is by optimizing cooling. This is necessary because cooling consumes a lot of energy. “Small adjustments to racks and airflow can already yield significant gains,” says the director.

A third approach is to make better use of residual heat. Data centers produce a lot of heat that is often lost. However, this heat can be put to good use, for example in nearby greenhouses. This means that these greenhouses do not need to use extra power, which reduces the load on the electricity grid. This is happening, for example, at Agriport A7 in North Holland, where greenhouses are heated with residual heat from a Microsoft data center.

Modern, powerful chips also offer opportunities for residual heat projects, Grove knows. Because these chips consume much more power, they also produce more heat. These higher temperatures are ideal for heating greenhouses or buildings. “This turns a challenge – the higher heat production of powerful chips – into an opportunity.”

No simple, straightforward solution

There is no simple, straightforward solution for power-hungry AI. Every little bit helps. Ultimately, making data centers more sustainable and creating an efficient, smart electricity grid will ensure that the large-scale deployment of AI becomes feasible for society. Grove concludes: "Let's not forget that AI also brings us enormous benefits. It's not for nothing that we use it every day, from search queries to video calls. Those benefits are just as important to mention."

Data centers' energy usage soars as efficiency improves

Forecasts see data centers electricity consumption doubling by 2026. We dig into the numbers in this new episode of Behind the Figures.

Read More