Token computeInterstice Advisory
Constraint Builder
Labor × Chips Collision — Phoenix (WECC)
stableRegime II
Reference scenario — Clone to save changes
Why it bindsThe machine

Local workloads absorb 3.9% of physical capacity (0.50 TTokens/mo demand vs. 12.89 TTokens/mo supply). No pricing premium; revenue tracks directly with defined workload SLAs. Significant idle capacity under this thesis — site economics depend on full contract coverage.

Marginal
Labor × Chips Collision — Phoenix (WECC)US-AZ
Network

Marginal: network capacity is binding — addressable demand partially unservable.

Supply exceeds demand by 25.8×50% premium on binding network

Binding constraint

Bandwidth or path diversity restricts workload dispatch

A portion of addressable demand cannot be served, reducing revenue and increasing unmet-demand risk.

The DC industry built for 20 years on “cheap land, cheap power.” AI breaks that model for inference workloads.

Near-binding risks

Capital

15 percentage points below the binding threshold

Hardware

28 percentage points below the binding threshold

Timeline

50 percentage points below the binding threshold

Regime 2 — Marginal: Mild supply-demand imbalance. Utilisation pressure is building but the facility remains cash-generative; one or two constraint releases would restore balance.

Regime 2 — Marginal. Binding: Network. Near-binding: Capital.
0.0
PowerElectrical power availability and grid capacity.
0.1
ThermalHeat rejection capacity of the cooling system.
1.0
NetworkExternal bandwidth and path-diversity constraint.
0.7
HardwareGPU and accelerator supply constraint.
0.5
TimelineConstruction and commissioning schedule constraint.
0.0
LaborSkilled workforce availability for operations.
0.2
PermitsRegulatory and permitting approval constraint.
0.8
CapitalCommitted funding availability for the next expansion.
derivedgap metric & fill per node — engine-derived from registry assumptions