The AI revolution has led to an unprecedented surge in demand for data center capacity. For operators like Polar DC, who are creating a platform of AI-ready data centers, the kind of scale that was previously reserved only for occasional rare mega-projects has now become table stakes for many builds.

“If you think about traditional data centers, they might have two-or-three-megawatt halls, with tens of customers in there, maybe taking up to a megawatt of capacity. Typically, with traditional cloud, there were five-or-10-megawatt halls, and maybe a little bit bigger for the hyperscalers,” says Andy Hayes, CEO at Polar DC.

“But now with AI, we’re seeing one customer asking for 70 megawatts in one or two data halls. So, there are now single halls the same size as entire traditional data centers would have been.”

This shift is being driven predominantly by neocloud startups, who offer scalable GPU-based compute as a service for downstream customers renting infrastructure on which to build their own AI models. Thanks to the extreme demands of running AI training and inference at scale, Hayes says, these facilities need to be equipped for high-capacity workloads from day one.

He adds that Polar DC’s neocloud customers are also targeting “a speed to market that we’d never seen before”, with aggressive timelines that drastically compress typical multi-year build schedules.

The inside-out data center

Rather than trying to balance the needs of traditional data center buildouts with these new usage models, Hayes says Polar DC has chosen to go all-in on AI, designing and building their facilities specifically to cater to AI-first neocloud workloads.

However, the enhanced power and compute requirements of high-density AI applications place significantly more taxing demands on the mechanical, electrical and plumbing (MEP) infrastructure of a data center. Internal power grids must be able to cope with sudden, intense spikes in power consumption, while thermal management infrastructure needs to be able to handle extreme localized heat loads using direct liquid cooling.

This is one of the reasons why Polar aims to work closely with OEMS and manufacturers, like NVIDIA, AMD, HPE, Dell, and Supermicro, to gain a robust understanding of the full IT stack. This allows us to “take a much more inside-out approach” to constructing a new facility, working with clients to understand their foundational needs and challenges before the project breaks ground.

“You’ve got to really understand the technology first, rather than just constructing a building, putting some infrastructure around it, and then the customer will do what they want inside of it,” he says. “We very much start from an operational mindset, to try and bring people in earlier, because the customers – particularly neoclouds – are wanting more help. They want more help in managing their projects, managing the installation of the white space, and they want more help in the ongoing management of the space.”

Northern lights

As data centers become more and more critical to business strategy, operators across the globe are grappling with logistical challenges when it comes to launching new facilities. These include availability of labor, growing materials costs, and uncertainty around zoning permits for new facilities.

Polar’s solution is to adopt a modular, prefabricated approach. Through its partnership with data center infrastructure provider Vertiv, the company produces pre-built units complete with factory-tested power and cooling capabilities, which can be deployed much more quickly than building infrastructure on-site.

“We’re seeing supply lead times significantly increase for key equipment. We’re seeing factories being fully booked out, price points go higher, and if you haven’t got the right partnerships with the supply chain, it’s very difficult to move forward,” Hayes says. “We’re trying to move more work into a factory environment that can facilitate more customers at the same time. A lot of our builds are prefabricated and modular; we move that into the factory, train the people in the factory, and try to remove the amount of on-site work that may need skilled labor which perhaps isn’t available.”

One of the reasons labor and materials for new data center projects are in somewhat short supply is that competition is fierce – and this also applies to land. Operators are scrambling to secure development plots for new data centers, but since the company’s inception, Polar has prioritized the Nordic region as a key stronghold for its facilities.

The company currently operates four sites across Norway, with a pipeline of further developments across Norway, Finland, and Sweden, as well as Germany. The region has many benefits for data center operators, according to Hayes, including plenty of space and a large proportion of high-level technical and engineering talent from industries such as heavy manufacturing or oil and gas.

The region also has extremely reliable energy grids, with heavy investments in renewable energy. Around 90 percent of Norway’s energy grid, for example, is powered by hydroelectric energy from the country’s many rivers and fjords, alongside favorable environmental conditions that allow for extensive ambient cooling. This puts Polar DC in a strong position in terms of sustainability metrics, which are becoming increasingly critical for customers who want to get ahead of Net Zero regulations.

The company is leveraging other opportunities to make its facilities more sustainable, such as working with local industrial parks that can use the facility’s waste heat as part of their processes.

Hayes is also keen to point out that, despite widespread misconceptions around the water usage of AI data centers, the technology used in Polar’s facilities does not use a particularly high volume of water. While data centers that rely heavily on evaporative cooling systems can consume significant volumes of water, Polar uses closed-loop water cooling systems, where systems are filled once and continuously recirculated without ongoing evaporative consumption.

“We fill our systems up with water once, and then they use a minimal amount of water during a year,” Hayes explains. “We’re not constantly using or consuming lots of water.”

“It was a conscious decision for us to go down that route. That type of design is more prevalent in Europe; I think in the US they’re actually playing catch up to that a bit.”

Beyond ‘braggawatts’

Hayes’ primary focus is on building long-term stability for Polar’s clients. He notes that Europe is seeing “a lot of power applications for projects that may never happen,” with providers focused on headline-grabbing “braggawatt” capacities, but Polar’s model will still be built around 15 to 20-year leases.

According to Hayes, the wave of AI adoption is just getting started, and although there’s still some skepticism around AI, the level of uptake is only set to increase as more people experience the benefits it can bring.

“I think what we’ll see is a much wider scale adoption and positivity about AI and people will truly understand the use cases and how it complements their everyday lifestyle, and also how businesses will benefit from the use of AI.”

For more information, please visit polardc.com.