Hong Kong-based technology firm 3 E Network Technology Group Limited has released the detailed engineering blueprint for its planned AI data center in Mikkeli, Finland, with physical infrastructure designed from the ground up to support NVIDIA's next-generation Vera Rubin DSX system-level architecture.

The company, which trades on Nasdaq under the ticker MASK, announced the plans on September 11, 2026, describing the facility as a core component of its ambition to become a next-generation AI infrastructure solutions provider.

The project reflects a deliberate attempt to address what the company characterizes as a critical structural problem in the data center industry: the mismatch between the long service life of physical facilities, which typically span ten to fifteen years, and the rapid iteration cycles of AI chips, which turn over roughly every one to two years.

By designing the physical layer of the facility to accommodate future generations of compute hardware before that hardware exists commercially, 3 E Network says it aims to reduce the cost and disruption of future upgrades while preserving the long-term value of hardware investments.

The Core Engineering Philosophy

3 E Network describes its approach as "forward-looking infrastructure readiness combined with pragmatic computing cluster deployment." In practice, this means separating decisions about facility construction from decisions about compute hardware procurement.

The civil, mechanical, electrical, and environmental control systems at the Mikkeli site are all being built to standards aligned with the Vera Rubin architecture, while the initial compute hardware deployed in the facility will consist of existing HGX and MGX-based systems.

The company frames this separation as a strategy for managing long-term total cost of ownership. By ensuring that the physical infrastructure does not need to be extensively modified when new generations of AI chips are introduced, the facility is intended to remain commercially viable across multiple hardware cycles without requiring major civil or mechanical engineering work each time compute hardware is refreshed.

Thermal and Cooling Infrastructure

One of the most significant engineering challenges the blueprint addresses is thermal load. The company anticipates that as the Vera CPU and Rubin GPU Superchip, alongside new HBM4 high-bandwidth memory, are deployed in future hardware generations, single-rack thermal loads will reach levels of tens to over a hundred kilowatts per rack.

To accommodate this, the Mikkeli facility is being designed with a direct-to-chip liquid cooling system and a Coolant Distribution Unit circulation architecture compatible with blind-mate technology.

The facility's location in Finland is cited as an advantage in this context. The blueprint describes plans to leverage Finland's natural cooling resources to optimize Power Usage Effectiveness under high thermal loads, while also building in adequate cooling redundancy to maintain performance continuity.

Networking and Interconnect Readiness

The facility's networking infrastructure is being built around the demands of NVIDIA's sixth-generation NVLink and the ConnectX-9 architecture. The announcement states that the data center's backbone fiber cable trays and routing pathways are fully aligned with 1.6T scale-out non-blocking network topologies. Ample rack-level routing space has been reserved to accommodate the high-density cabling requirements of intra-rack interconnects at this scale.

The design also reserves wiring space for next-generation Spectrum-X Ethernet and advanced InfiniBand architectures, with the stated goal of providing the physical infrastructure needed for rack-scale interconnects as those technologies become commercially available.

Structural and Power Systems

The physical weight of next-generation liquid-cooled computing racks presents its own set of engineering requirements. The company states that it anticipates rack weights of nearly two tons and has implemented reinforced flooring designed for high load capacities during the civil engineering phase of construction.

On the power side, the facility adopts what the company calls a flexible dual-track design. It supports the deployment of high-voltage smart power distribution units to accommodate standard MGX and HGX nodes, while simultaneously providing forward compatibility for 48-volt centralized DC power shelves and rack-level Battery Backup Unit specifications.

This design is intended to handle the high power density and transient power spikes expected with the Vera Rubin architecture while enhancing stability at the grid level.

The blueprint also incorporates out-of-band environmental monitoring and micro-leak detection systems, along with standardized interfaces between facility infrastructure and IT equipment.

The company says this design approach is intended to reduce the need for facility modifications during future compute hardware upgrades and minimize operational disruption when those transitions occur.

Commercialization Strategy During Initial Phase

While the physical infrastructure is being built to future-generation standards, 3 E Network's commercialization plan for the facility's first phase centers on a mixed deployment of currently available hardware.

The company plans to operate what it describes as an elastic compute pool combining HGX-based and MGX-based computing clusters, with each category targeting a distinct segment of the AI compute market.

The HGX core zone is designed to accommodate high-density deployments of high-performance HGX baseboards supported by centralized power and liquid cooling. According to the announcement, this zone is intended to serve AI enterprises and foundational model developers working on complex, trillion-parameter pre-training tasks.

The company states that the lossless interconnect architecture in this zone is designed to mitigate what it calls the "data wall" bottleneck, providing efficient compute throughput for large-scale training workloads.

The MGX elastic zone is positioned to address commercial inference demand, representing the second track in the company's dual-track monetization approach.

The company describes the overall mixed-cluster architecture as a mechanism for optimizing hardware asset utilization and return on investment by matching compute resources to the two dominant commercial demands currently present in the AI market.

Track Every Megawatt of Finland's Data Center Pipeline in One Place

As hyperscale operators and colocation providers race to meet surging AI and cloud demand, the pace of data center development across Finland has never been more intense, and the intelligence gap between market leaders and late movers has never been wider.

Staying ahead means knowing not just what is being built, but what is being planned, bid, and awarded before your competitors do. The Global Project Tracking (GPT) platform by Blackridge Research gives data center developers, contractors, and investors a structured, continuously updated view of the full Finland project landscape.

From power-hungry campus developments to edge deployments across Finland, the GPT platform ensures your team is working from the most current market intelligence available.

• Upcoming Projects

• Tender Notices

• Contract Awards

• Projects Under Construction

• Completed Projects

See how the Global Project Tracking (GPT) platform by Blackridge Research can sharpen your Finland data center strategy. Book a Free Demo with our team today.