High-density training and mining infrastructure — from Nordic hydro and geothermal grids to continental heat-reuse networks.
Illustrative design scenarios, not verified Global Data House projects. Capacities, delivery years and regional statuses are assumptions. Dojo-class is a conceptual D1-based design, not a Tesla deployment or affiliation.
4 campuses · All scenario stages
High-density AI
Luleå Aurora
Luleå, Sweden
Nordics · SE1 · Target 2027
Under construction
Site exterior
Illustrative rendering · not a photograph of a built site
Facility
60 MW
IT load
50 MW
Target PUE
1.20
32,768 H200 GPUs
4,096 eight-GPU servers; approximately 41 MW including hosts. Remaining IT allowance covers fabric and storage.
80–120 kW / AI rack
Power sourcing
Swedish grid connection with a proposed hydro-backed PPA; firm capacity remains grid-dependent.
Cooling
Warm-water liquid loop with dry coolers; winter economization.
Heat reuse
Proposed 10 MWth export to Luleå district heating via heat pumps. Seasonal demand and utility acceptance govern delivery.
Full EPC · 130 kV substation · direct-to-chip cooling
Engineering
Procurement
Construction
Commissioning
Operational
Scenario: phased build; SE1 grid capacity and heat offtake subject to agreement.
Mining + AI
Reykjanes Basalt
Reykjanes, Iceland
Nordics · Iceland · Target 2028
Engineering
Site exterior
Illustrative rendering · not a photograph of a built site
Facility
30 MW
IT load
25 MW
Target PUE
1.20
4,000 SHA-256 ASICs
S21-class ASICs: 200 TH/s at 3.5 kW each; 0.8 EH/s at 14 MW. AI wing: 4,096 H200 GPUs in approximately 5.1 MW of servers.
80–100 kW / AI rack · immersion ASIC pods
Power sourcing
Icelandic hydro/geothermal grid; proposed firm supply agreement, not a dedicated geothermal plant.
Cooling
Separate immersion-mining and direct-to-chip AI loops; closed-loop dry heat rejection.
Heat reuse
Up to 3 MWth proposed for local greenhouse/process use. No district-heating export assumed where geothermal heat already serves demand.
Modular EPC · isolated immersion halls · AI training wing
Engineering
Procurement
Construction
Commissioning
Operational
Scenario: feasibility stage; power allocation, seismic risk and volcanic hazard assessment pending.
High-density AI
Frankfurt Main Compute
Frankfurt, Germany
Central Europe · Hesse · Target 2026
Commissioning
Site exterior
Illustrative rendering · not a photograph of a built site
Facility
48 MW
IT load
40 MW
Target PUE
1.20
24,576 H200 GPUs
3,072 eight-GPU servers; approximately 30.7 MW including hosts, with capacity reserved for network and storage.
80–120 kW / AI rack
Power sourcing
German grid with proposed wind/solar PPAs; annual renewable matching is not continuous carbon-free supply.
Cooling
Direct-to-chip cooling with redundant CDUs and heat-pump-ready return-water loop.
Heat reuse
Proposed 8 MWth district-heating tie-in via heat pumps, subject to network build-out, temperature and offtake contracts.
Urban fit-out · 110 kV intake · 800G training fabric
Engineering
Procurement
Construction
Commissioning
Operational
Scenario: staged commissioning; final grid energization and heat-network acceptance outstanding.
AI + Dojo-class concept
Aragon Ebro Training
Aragon, Spain
Southern Europe · Aragón · Target 2028
Engineering
Site exterior
Illustrative rendering · not a photograph of a built site
Facility
72 MW
IT load
60 MW
Target PUE
1.20
32,768 GPUs + Dojo concept
32,768 H200 GPUs in approximately 41 MW of servers. Separate conceptual 120-tile D1 hall: 25 dies/tile, 3,000 dies; approximately 1.8 MW tile power, excluding hosts and fabric.
80–120 kW / AI rack · dedicated tile cabinets
Power sourcing
Spanish grid with proposed regional wind/solar PPAs; storage smooths peaks but does not guarantee 24/7 renewable operation.
Cooling
Liquid-cooled training halls with dry coolers sized for hot summers; no routine evaporative cooling assumed.
Heat reuse
Proposed 5 MWth industrial/greenhouse heat offtake; district-heating connection only if a nearby network becomes viable.
