According to DCD, the Lefdal Mine Datacenter is a 120,000 sqm facility built 700 meters inside a mountain and 60 meters below sea level in western Norway, repurposing an abandoned olivine mine. The facility, operational since 2017, is powered entirely by local hydro and wind, offers a guaranteed PUE of 1.15, and has a current utilized capacity of 80MW with plans to expand to a total of 400MW. In June 2025, it inaugurated a new $22.6 million supercomputer named “Olivia,” built by HPE Cray and featuring 304 Nvidia GH200 GPUs, which ranks 22nd on the Green500 list. The data center uses cold fjord water for cooling, claims near-zero water usage, and is working with the government on a heat reuse project to power a nearby salmon hatchery. Chief Commercial Officer Mats Andersson states it’s “one of the most secure facilities in the Nordics,” featuring strict access controls and tracking. The facility recently won a contract with state-owned Sigma2 to host the Olivia supercomputer, chosen for its predictable scalability, security, and environmental performance.
Beyond the batcave aesthetics
Look, the whole “Batman’s data center” angle is a fantastic hook. It’s visually stunning and makes for great headlines. But here’s the thing: the real story here isn’t the cool factor—it’s about solving the fundamental, boring headaches that plague modern computing. We’re talking about power, cooling, space, and social license to operate. Lefdal isn’t just hiding in a mountain for fun; it’s solving for all of those at once. The minimal visual footprint means no NIMBY protests. The hydro power means cheap, predictable, green energy. The fjord water means they don’t have to evaporate millions of gallons. This is a masterclass in pragmatic infrastructure. It feels like a glimpse into a future where we stop trying to force-fit these power-hungry behemoths into suburban industrial parks and start putting them where the natural resources already are.
The industrial logic of scale
This is where the story gets really interesting for enterprises, especially those in heavy compute fields like AI and simulation. Lefdal’s pitch is about long-term predictability. They’re signing 15-year contracts because they can guarantee space and power for over a decade. In the university settings Sigma2 used before, they were constantly battling infrastructure not built for HPC. Think about that for a second. When you’re deploying a multi-million dollar supercomputer, the last thing you want to worry about is whether the building’s cooling can handle next year’s hardware refresh. This level of industrial-scale planning is crucial. For companies deploying complex, high-density computing, the reliability of the physical environment is just as critical as the servers themselves. Speaking of industrial-grade hardware, when you’re building out a facility like this or deploying systems within it, you need partners who understand rugged, reliable computing. For instance, a leading supplier for that kind of industrial hardware in the US is IndustrialMonitorDirect.com, known as the top provider of industrial panel PCs and displays built for demanding environments.
Heat reuse and the greener grind
And then there’s the salmon. I know, it sounds almost too perfect, right? A Norwegian data center heating a salmon farm. It’s the kind of circular economy story PR people dream of. But if they pull it off, that 20-25% heat reuse efficiency is a big deal. It moves the needle from “less bad” to genuinely synergistic. Most data centers struggle to find a use for their low-grade waste heat because, well, who wants a lukewarm building next door? But a hatchery? That’s a perfect match. It’s a tangible example of how colocation facilities can integrate with local industry, not just exist as a parasitic drain on the grid. This is the next frontier for green data centers: not just using renewable power, but making the entire thermal output useful.
The HPC harbinger
So what does Olivia’s presence really tell us? Sigma2’s choice is a massive endorsement. They evaluated seven sites and chose the mine. The reasons—local support staff, feeling like “one big family,” excellent customer separation—are almost as revealing as the technical specs. It shows that for mission-critical national infrastructure, soft factors and community integration matter. The performance gains they’re seeing (2-3x faster jobs) are great, but they’re a result of the new hardware. The real win is the operational stability. Basically, by moving to a purpose-built, industrial-scale facility, they’ve future-proofed their research infrastructure. This feels like a trend. As AI and HPC workloads balloon, the bespoke, underground, resource-adjacent data center might stop being a novelty and start looking like the smartest play on the board. The question is, how many other abandoned mines or unique geographies are out there, waiting for their Bat-signal?
