The Borealis Data Center is a nondescript building next to a horse farm on the outskirts of Blönduós, a town of 800 inhabitants in the north of Iceland. Entering the building, after you make it past the multiple levels of security, and into the inner sanctum of the server room. The sound in that room is so loud — like a jet engine — that you need professional ear protection, and you are hit by a wall of warm air from cooling fans radiating the heat of hundreds of machines running 24/7 fielding artificial intelligence (AI) queries from around the world.
From bitcoin to AI
Iceland is on an upward trajectory as a destination of choice for AI server farms, promising economic growth. Currently the sector is five percent of Iceland’s GDP, and an ecosystem of innovation. Iceland is seen as a utopia of low-cost renewable energy in a cold and stable climate, minus of course the volcanoes and slowing Atlantic Meridional Overturning Circulation, or Gulf Stream, predicted to drastically change Iceland’s climate.
Indeed Iceland’s data centres have gone through a sea change over the past few years, from cryptocurrency speculators mining bitcoin to large learning models mining data for AI.
“The industry is now providing higher-value services to AI clients, compared to some years ago when the industry was serving more blockchain. Consequently we are increasing hires in our sites around the country and providing more high-value technical jobs which also positively impacts the Icelandic economy,” says Bergþóra Halldórsdóttir, chief of staff at Borealis Data Center & board member at The Federation of Icelandic Industries.
With AI, says Bergþóra, you are dealing with clients who are more concerned with quality than with volume. “This is very valuable equipment, and there is an increased need to provide 24/7 services.”
Last year, tens of millions of dollars worth of bitcoin servers were scrapped by a client at the Borealis Data Center, and replaced by more advanced and costly AI machines. And it’s just the beginning as Iceland’s AI data centre landscape is poised to grow substantially. Renewable energy has been at the core of the value proposition for Borealis, and other data centres like atNorth. Iceland’s climate is well-suited for data centres. It is not just that it is cold, but that the temperature does not fluctuate much. The stable cool temperatures mean data centres are able to use less energy overall, and therefore operate at a lower cost.
Plugging AI into the power grid
“The cooling of computer equipment alone is responsible for 40% of the total electrical costs of most data centres,” says Eyjólfur Magnús Kristinsson, CEO of atNorth. Additionally, he points out, there are a number of fibre-optic cables connecting Iceland to both North America and Europe, and Iceland ranked as the world’s safest location for data centres in the 2016 data centre risk index.
To further cut down on energy use, up to 90% by their estimates, AI Green Bytes, an Icelandic data-centre startup is offering immersion cooling where the computers are submerged in a non-corrosive liquid, eliminating the need for fans.
The limiting factor to growth in Iceland is the power grid. Currently, roughly six percent of electricity generated in Iceland is used for data centres, and the energy supply is not as large and stable as is needed. Iceland has a closed energy grid that is all renewable, mostly hydroelectric and geothermal. The Borealis Data Center draws power from a nearby hydroelectric dam, and the variation in rainfall in the past year has had an impact on energy output.
“We need to increase available power to attract the big tech giants which are looking to deploy to larger and larger sites,” says Bergþóra. Borealis has been encouraging Icelandic energy producers and transmission companies to make sure that they are optimising the potential as they see a huge opportunity to expand.
“We know Iceland has been looked at for several AI projects that are being planned and we believe it would positively impact the industry to see large scale sites being developed with leading companies in AI.” As of today, none of the major US tech giants — Meta, Google, Microsoft — house their servers in Iceland.
Covert AI data farming
“We don’t like that they are building these data centres and using more energy, and we don’t know what the data centres are used for,” says Björg Eva Erlendsdóttir from the Icelandic conservation organisation, Landvernd. Landvernd took a position against bitcoin mining, but they haven’t yet formulated a position on AI.
“We don’t have information, we have asked why we are not getting information from the government or the companies. We can’t just welcome all data centres to Iceland.”
Árni Finnsson from the Icelandic Nature Conservation Association concurs. “The risk is that energy is used for great expansion of data centres. The transparency is nil.” This summer Iceland broke ground on a new wind farm, which Bergþóra describes as an untapped resource for the energy grid.
“No one knows what all this wind energy is for,” says Árni.
Iceland’s AI policy doesn’t cover data centres
To keep up with this rapid evolution of AI, just this month the Icelandic government released an AI Action Plan to ensure “smart, sustainable and sovereign digital systems”, according to the Ministry of Culture, Innovation and Higher Education. Bergþóra would like to see a cohesive plan for data centres similar to other Nordic countries, but that was not in the scope of the AI Action Plan, which focused on government services and local society.
As everyone is on summer holiday, a spokesperson for the Ministry replied by email when asked about the plan.
“One of the main themes of the action plan is to ensure responsible development, implementation and use of AI systems. The policy highlights respect for human rights and democratic values, liberty, equality and transparency. The framework around management and processing of data will be in line with the international commitments including European guidelines and regulations.”
Data centres in Iceland are currently used primarily for training AI models. Such models are often trained on voluminous amounts of data in violation of copyright, though whether that is the case in Iceland remains unknown and either way, the government’s policy won’t apply to Icelandic data centres.
Iceland is a small data nation
Kristinn R. Þórisson, research professor at Reykjavík University and director of the Icelandic Institute for Intelligent Machines, takes a more philosophical approach to the overall question of AI, saying that we have to rethink the value of large learning models for society in general and Iceland specifically.
“Iceland was a leader in doing AI research in the Nordics,” Kristinn says. “We are celebrating the 20th anniversary of the institute.” Neural networks is just one of the things they study at the Center for Analysis and Design of Intelligent Agents, think the curious octopus with mini brains spread throughout their body. Kristinn admits this approach of intelligent agents has gone out of fashion with the rise of large learning models, but he is skeptical about both the promise and fears of artificial general intelligence.
“AI will be a hot technology for at least another 30 to 50 years. Intelligence is a complex phenomenon, and we can not forget about the theoretical part of it. In spite of what leaders of the US tech giants try to convince us of, we are very far from general machine intelligence.”
Kristinn questions the value of large learning models for Icelandic society. “These machines are dependent on big data, and Iceland doesn’t have big data, Iceland is a small data nation.” The data being stored in Iceland is not about Iceland, “It is data for English, data for Spanish, data for Chinese, it is for other purposes, it doesn’t serve the Icelandic community, Icelandic lawyers can’t use large learning models trained on English laws.” The Icelandic that comes out of these models is terrible, he says, because there simply aren’t enough Icelandic texts.
Kristinn has been urging the Icelandic government to think about the next type of AI, small data AI. “It’s not just the amount of data that is doing a disservice to Iceland, but also the trustworthiness, the ability to use AI for creating cybersecurity, for example, for a small country automating that would be a boon. But that needs to be local.”
Going to new models of AI other than large models, he adds, is akin to when Africa decided not to lay phone cables, but went directly to wireless technology.
Asked whether he worries that a sentient all-knowing AI will emerge from an Icelandic data centre to take over the world, Kristinn says, “It’s possible.” He adds, however, “Nature still outperforms any machine you can make today.”
When asked if the Borealis Data Center will give rise to an AI overlord, Bergþóra laughs, “Our staff is quite comfortable around the computers we serve, so we are not concerned about such risk. This is incredible technology, and we are already seeing the positive contributions to our own internal work, so we only see AI as providing opportunity for the time being.”
Buy subscriptions, t-shirts and more from our shop right here!
