HIVE taps Sweden’s Boden for up to 10,000 NVIDIA GPUs under LOI

HIVE eyes large-scale GPU colocation in Sweden’s Boden

HIVE Digital Technologies, a Nasdaq-listed digital infrastructure provider, has signed a non-binding letter of intent (LOI) for a long-term lease at its Boden data center in northern Sweden, according to information provided in an investor update circulated via a PR agency email.

The proposed arrangement is structured around high-performance computing (HPC) colocation. If progressed to a definitive agreement, the tenant would use a portion of the site’s capacity, and HIVE would retrofit the facility to support large-scale GPU deployment.

What the LOI says

The LOI contemplates an up to 10-year term. The customer would take approximately 25 MW of critical IT load at HIVE’s 32 MW Boden facility, with the power allocated for HPC use cases.

HIVE’s plans for the site include upgrades aimed at supporting up to 10,000 NVIDIA GB300 GPUs. The company also outlined target rack-level power densities of up to 150 kW, using a mix of hybrid cooling approaches, including liquid cooling designs intended for direct-to-chip use alongside air cooling.

Because the LOI is non-binding, it should be treated as an initial step rather than a guaranteed contract. In data center deals, LOIs often precede final commercial terms, engineering scope confirmations, permitting, and the completion of technical and financial due diligence.

From crypto infrastructure to HPC and AI compute

In recent years, public digital infrastructure companies have faced a shift in how markets evaluate their business models. While many operators built their early capacity around crypto mining, demand for compute, particularly for AI training and inference workloads, has pushed the industry toward GPU-focused data center services.

HIVE’s announcement positions the Boden expansion as part of that broader transition. The company says the project marks a milestone in its evolution beyond cryptocurrency-oriented infrastructure toward next-generation AI and HPC data center services.

For investors and industry observers, the practical question is whether large GPU colocation commitments can translate into durable revenue streams. Long-term lease structures are designed to provide capacity utilization stability, which can be especially valuable in a sector where capex intensity and power availability are central constraints.

Why Sweden and why “sovereign” compute

The LOI also reflects a growing European theme: domestic compute capacity for regulated or government-aligned use cases. The email characterizes the customer as an “investment-grade sovereign” Swedish technology company, and links the site selection to technical evaluation, including site visits and due diligence.

HIVE attributed the selection to factors such as the facility’s established infrastructure, access to renewable energy, and its operational presence in Sweden. The broader framing aligns with the concept of “sovereign AI infrastructure,” where nations and strategic partners aim to source compute domestically to support policy goals, data handling requirements, and supply chain resilience.

Facility timeline and operational context

HIVE has been operating the Boden site since 2018. In its description of the LOI, the company referenced its experience managing large-scale GPU deployments in the region. It also stated that it previously operated approximately 130,000 GPUs across Swedish operations, according to the email’s background information.

The project is further described as building on a municipal approval process. The email says the Boden Municipal Council approved HIVE’s acquisition of the facility, reinforcing what HIVE describes as a long-term commitment to the location.

From an execution standpoint, retrofitting a data center to support high-density GPU racks typically requires careful planning across power delivery, cooling architecture, and facility uptime strategy. Hybrid direct-to-chip liquid cooling designs, combined with air cooling, can help meet thermal and space constraints, but they also add complexity for installation, maintenance, and operational readiness.

Potential industry implications

If the parties move from LOI to a binding agreement, the deal would underline several trends already visible across Europe’s data center market.

• GPU density is becoming a procurement requirement. Rack-level targets up to 150 kW suggest that customers are increasingly specifying performance envelope details, not just availability.
• Cooling design is shifting from conventional models. The plan to use direct-to-chip liquid cooling indicates the operational maturity needed for dense accelerators.
• Long-term contracts remain central to financing. For operators, multi-year lease structures can support the economics of large capex programs.
• Sovereign demand may widen the geography of AI infrastructure. A Swedish site focused on domestic compute capacity reflects how non-US regions are competing to host HPC capability.

What to watch next

With the LOI described as non-binding, the next steps for market participants are likely to include confirmation of the final lease terms, the scope and schedule of facility upgrades, and the customer’s commissioning timeline for the targeted GPU capacity.

HIVE’s broader strategy, including how it pairs AI workloads with existing operations and capital plans, will also matter for how this project fits into its longer-term financial profile. The email notes interest in interviews around sovereign AI infrastructure and HIVE’s multi-country infrastructure footprint, but it does not provide additional deal specifics beyond the LOI parameters and facility retrofit outline.

As GPU colocation becomes more mainstream, outcomes like this will help determine whether “sovereign AI” demand translates into repeatable data center economics across Europe, not only one-off custom builds.