Granlund designs specialised AI data centers for Nordic climate

“Companies need data centers to train their machine learning models. These data centers are different from a data center designed for colocation services,” explains Ilari Kauppi, Senior Lead Mechanical Engineer at Granlund.

A colocation data center is designed to meet the variety of needs of companies renting space. An AI data center, however, is focused on one task and is defined by higher power density and larger size.

“AI needs enormous computing power and a similarly robust energy infrastructure,” Kauppi says. “A data center designed for AI requires more energy and hence more cooling. The equipment is also scaling up and getting more power dense: Nvidia’s new DGX rack system weighs 1.36 metric tonnes.”

An AI data center, however, is focused on one task and is defined by higher power density and larger size.

Ilari Kauppi, Senior Lead Mechanical Engineer at Granlund

Cheap cooling and clean energy

Handling the generated heat is one of the biggest challenges of an AI data center. Traditionally, data centers used air cooling, but for AI water cooling is becoming more common.

“The Nordic climate is one of the reasons companies want to put their data centers here,” says Timo Ranne, Director of International Business Development at Granlund.

“We have a lower ambient temperature which makes cooling cheap and efficient. It is even possible in the future to use only pumps and dry coolers for AI, not chillers. Also, towns will take the waste heat for their district heating networks.”

Another difference between traditional and AI data centers is where they are located. Typically, to reduce latency a data center is close to big cities where most international optical fibre connections terminate. This is less important for an AI data center, so they are placed closer to sources of power and the energy infrastructure connection points.

AI data centers are placed closer to sources of power and the energy infrastructure connection points.

Timo Ranne, Director of International Business Development at Granlund

“Companies want cheap and clean electricity for their data centers. For instance, the state-of-the-art supercomputer LUMI in Kajaani, Finland, uses hydro- and wind power,” Ranne says. “Sustainability is important for these power-hungry data centers, so companies look for locations with excess renewable energy capacity.”

While there is increasing interest in renovating old data centers for AI, most AI data centers are new builds because of their special requirements. These requirements are constantly changing as new servers come to market, but still need to be foreseen due to long construction times.

More information

More news