Operations

Systems & Capacity

The following capacity figures reflect commissioned and provisioned infrastructure. Figures are subject to revision as operational thresholds are validated and integration milestones are confirmed.

Capacity status

Capacity is reported as a combination of commissioned (in operation) and provisioned (built and ready for activation). Values represent sustained operating envelopes rather than peak or theoretical maxima.

Where ranges are provided, they reflect variation driven by cooling mode, redundancy configuration, and integration constraints. Operational envelopes may be revised as thresholds are confirmed under live load.

Table 1 — Stage One: Commissioned capacity (First Run)

Stage One refers to the initial commissioned run of the facility, including supporting energy and thermal systems required for stable continuous operation.

Category Metric Stage One Notes
Built form Continuous building length ~1.6km Single run (continuous envelope)
Built form Enclosed floor area ~900,000 Total area under roof
Compute Active compute floor (white space) ~320,000 Commissioned zones only
Power Installed IT load (sustained) ~1.55GW Continuous envelope
Power Total facility draw (sustained) ~1.8GW Includes cooling and conversion losses
Energy Daily energy consumption (at sustained draw) ~43.2GWh/day 1.8GW × 24h
Energy Annualised energy (at sustained draw) ~15.8TWh/year Indicative at continuous operation
Thermal Heat rejected (continuous) ~1.5GW thermal IT load dominance assumed
Cooling Cooling water circulation (peak internal loop) ~90–110m³/s Mode dependent; closed-loop circulation
Efficiency Target PUE (cold-season average) ~1.15–1.17 Envelope target
Operations Mean time to isolate (containment) < 90seconds Fault-domain isolation target
Operations Human staffing (typical 24h coverage) < 300 On-site staffing; excludes off-site support

Definitions: “commissioned” indicates operating under sustained load; “white space” refers to compute hall allocation under live service.

Table 2 — Provisioned expansion capacity (built and ready)

Provisioned capacity refers to infrastructure that has been constructed and integrated to the point where activation is a sequencing decision rather than a civil programme. Values reflect design envelopes reserved within installed power and thermal spines.

Category Metric Provisioned total Notes
Built form Total continuous length (design) ~4.2km End-state geometry reserved
Built form Total enclosed area (design) ~3.4million m² Under-roof envelope
Power Maximum sustained IT load (envelope) ~10GW Conservative planning envelope
Power Maximum sustained facility draw (envelope) ~8–12GW Cooling mode and redundancy dependent
Generation Hydro generation capacity (staged) ~6–8GW Installed and planned units
Storage Pumped storage buffer ~18–22GWh Commissioning and ramp support
Electrical Substation spines (linear) ~5km Distributed conversion and switching
Cooling Cooling galleries (linear) > 18km Terminated and blanked for activation
Networks Conduit banks (power + data) > 1,200km Includes dark fibre allocation
Energy Projected annual energy (full envelope) ~90–100TWh/year Indicative at sustained operation
Lifecycle Design operating horizon 40–60years Upgrade-in-place assumed

‘Continuous’ refers to a structurally and operationally unified envelope composed of repeated, compartmentalised segments rather than a single uninterrupted interior volume.

Provisioned capacity may be present as capped foundations, terminated galleries, energised spines, and reserved envelopes within installed generation and conversion assets.

Table 3 — Failure modes & containment (internal)

The following scenarios describe operational containment patterns and residual impacts. This section is provided for completeness and reflects system-level planning assumptions.

Failure scenario Trigger condition Immediate effect Containment response Residual impact
Hydro unit trip (single) Turbine instability; grid transient ~400MW loss Pumped storage discharge; staged shedding No external visibility
Hydro unit trip (multiple) Common-mode fault ~800–1,200MW loss Automated throttling; hall isolation Training paused; inference preserved
Pumped storage unavailable Maintenance overlap; lockout Loss of ramp buffer Commissioning suspended Political, not technical
Cooling spine degradation Intake blockage; fouling Rising delta-T Thermal derate; workload migration Localised performance loss
Cooling spine failure Pump cascade; structural fault Thermal runaway risk Hard isolation (< 60seconds) Hardware sacrifice expected
HV substation fault Insulation failure; switching error 200–500MW islanding Automated re-route across spine Audible event; no outage
Transmission corridor loss Physical damage Segment isolation Local generation absorbs shock Grid islanding tolerated
Compute fabric partition Control plane instability Workload fragmentation Automatic rebind; failover Training time penalty
Control system desynchronisation Clock drift; firmware mismatch Cross-domain misalignment Forced synchronisation halt Human oversight required
Black start required Total site collapse (modelled) Complete power loss Hydro-led restart External attention unavoidable
Compound failure (grid + cooling) Extreme event; cascading faults Unrecoverable thermal risk Controlled sacrificial shutdown Equipment loss accepted
Public disclosure event Leak; imagery circulation Narrative instability Meaning management Trust erosion

Note: Recovery time is measured operationally; reputational impact is measured institutionally.

Compute architecture (summary)

The facility is provisioned for heterogeneous accelerator architectures with varying power density, memory topology, and interconnect requirements. Capacity figures reported above reflect aggregate operating envelopes rather than interchangeable compute units.

Certain workloads are bound to specific accelerator characteristics and cannot be migrated without material efficiency or performance impact. Provisioning accounts for long-lead semiconductor supply cycles, constrained fabrication capacity, and components subject to export control regimes.

Compute fabric
Heterogeneous accelerators; non-fungible allocation
Rack power density
120–180kW per rack (liquid-assisted)
Integration constraint
Interconnect topology and memory bandwidth bound
Supply horizon
Multi-quarter lead times; staged intake and validation

Note: Published capacity envelopes do not imply equivalence across accelerator classes or workload categories.

Recordkeeping

Historical versions of this page are retained in accordance with recordkeeping obligations.

Relationship to other operational areas

Background on generation and water systems is described under Hydropower and Energy Systems.

Principles guiding site development and integration are outlined under Infrastructure Principles and Site Development.