HIVE Eyes Up to 10,000 GB300 GPUs at Boden HPC Site

HIVE Digital Technologies has signed a non-binding letter of intent for long-term high performance computing colocation at its Boden, Sweden facility, a move that underscores how data center operators are positioning existing sites to serve the growing market for AI and GPU-intensive workloads.

A long-term colocation proposal in Boden

The company said the letter of intent, or LOI, is with an investment-grade, sovereign Swedish technology company. Under the proposed arrangement, the client would lease the Boden site for up to 10 years, subject to the negotiation and execution of a definitive agreement.

HIVE noted that the LOI is tied to its 32 megawatt (MW) Boden facility and would target an estimated critical IT load of roughly 25 MW for HPC colocation. The company has operated at the site since 2018 and previously referenced running approximately 130,000 GPUs in the region.

Retrofit plan targets up to 10,000 GB300 GPUs

According to the release, the contemplated colocation buildout includes a retrofit designed to support up to 10,000 NVIDIA GB300 GPUs. The company also specified potential single-rack densities up to 150 kilowatts (kW) and said the cooling approach would be hybrid, combining direct-to-chip liquid cooling with air cooling.

From an infrastructure standpoint, the combination of higher rack power density and liquid cooling is consistent with industry trends in GPU clusters where heat removal becomes a central engineering constraint. While the company provided the upper-end GPU and density targets, it did not give a detailed timeline for the upgrade in the release, and the agreement remains non-binding.

Municipal approval and site ownership context

HIVE said the deal follows earlier steps related to the Boden facility. The company referenced a prior announcement that Boden’s municipal council approved HIVE’s acquisition of the site from Bodens Utvecklings AB. HIVE has operated there for years, and the LOI appears to build on that operating footprint by moving toward a longer-duration HPC hosting contract focused on high-density GPU compute.

Why the critical-load framing matters

In data center planning, gross utility capacity and critical IT capacity are not the same. HIVE indicated that while the site has 32 MW of gross utility load, the critical IT load for HPC colocation would be approximately 25 MW. That distinction generally reflects power devoted to cooling, infrastructure overheads, and reliability requirements, which become more pronounced as power density and utilization increase.

Implications for AI infrastructure demand

The LOI is notable for several reasons. First, it reflects the shift in demand from traditional compute hosting toward GPU-accelerated AI infrastructure, where customers increasingly seek predictable long-term capacity and reliability guarantees. Second, HIVE’s framing suggests the Boden site is being repositioned from prior GPU deployments into an upgrade cycle aligned with newer GPU architectures.

For investors and industry watchers, long-term colocation commitments can be a signal of demand durability, particularly when they are paired with clear retrofit plans. However, because the LOI is non-binding and subject to a definitive agreement, the next step is not only contract finalization, but also engineering delivery and commercialization details that were not provided in the announcement.

What happens next

HIVE did not provide final commercial terms in the release. The company said the LOI contemplates a definitive lease agreement, but the structure, pricing, and final capacity deployment would be established during the negotiation process.

In parallel, any retrofit program will likely depend on execution factors that are common across high-power GPU projects, including equipment lead times, electrical and cooling capacity design, commissioning schedules, and operational readiness. Until a definitive agreement is signed and the retrofit scope is fully documented, the 10,000 GB300 GPU figure should be viewed as a target under the proposed plan rather than an assured deployment.

Bottom line

HIVE’s letter of intent in Boden points to a broader industry pattern, where data center operators seek to convert existing capacity into AI-ready infrastructure with higher density, liquid-assisted cooling, and long-term offtake. While the announcement stops short of finalizing a lease, it adds a concrete example of how compute infrastructure is being shaped for institutional-grade, long-duration HPC and AI workloads.