A single Arizona mega-fab, viewed through four progressively quantitative lenses.
The Dust Cloud Challenge asks a deceptively simple question: can a $20-billion EUV-class semiconductor mega-fabrication facility — call it Aether Chips — operate sustainably in Phoenix, Arizona over a thirty-year horizon? The honest answer requires more than a balance sheet or an environmental impact statement. It requires a systems view.
This showcase walks through the Arizona case study using the fellowship’s four-layer MBSE stack. The analog is real: TSMC Fab 21, currently the largest foreign direct investment in U.S. history. The constraints are real: a Tier-1-shortage Colorado River, structural drought entering its second decade, daily ultra-pure-water demand approaching 8.9 million gallons per fab, and a CHIPS-Act subsidy stack that ties federal disbursement to permit and disclosure milestones.
What unfolds across the four diagrams is a silicon–water nexus: a tightly-coupled feedback architecture linking GRI 303 (Water), GRI 413 (Local Communities), and GRI 201 (Economic Performance). Each layer translates the same system into a more rigorous form — qualitative loops become discrete states, states become parametric equations, and equations become inputs for autonomous decision agents. The result is a single defensible chain of reasoning from narrative to simulation.
From narrative reasoning to simulated governance
Each layer below corresponds to one section of this page. They are read top-to-bottom: the CLD surfaces the feedback dynamics, the state machine sequences regime transitions, the parametric diagram quantifies the equations, and the multi-agent simulator demonstrates how those equations could be governed in real time.
The Silicon–Water Nexus
The qualitative entry point. Ten system variables and four feedback loops — two reinforcing spirals (R1 Water Scarcity, R2 Social License Erosion) and two balancing loops (B1 Technology Pivot, B2 Federal Subsidy Buffer) — mapped across GRI 300s, 400s, and 200s with an Iceberg Model overlay surfacing events, patterns, structures, and mental models.
Five regimes, one absorbing endpoint
The CLD tells us what feeds back into what; the state machine tells us in what order. Five operational regimes — Stable, Stressed, Non-Compliant, In Recovery, Terminal — connected by guarded transitions tied directly to the parametric model’s constraint thresholds. The Terminal state has no exit arrow: once CHIPS disbursement falls below the 5% floor, the model treats it as absorbing.
The mathematical core
Five interdependent subsystems — Water, Energy, Production, Community, Economic — linked by constraint equations. WaterBalance and WaterStressIndex map to GRI 303; CommunityResistance and PermitRisk map to GRI 413; CHIPSFunding maps to GRI 201. Drag the sliders to see how a water-withdrawal increase cascades through community resistance into permit risk and CHIPS disbursement.
Externalizing the reasoning architecture
Five specialized AI agents — Water Sentinel, Grid Steward, Community Liaison, Compliance Watch, and Capital Guardian — an orchestrator, and a human review board jointly steer the fab through real-time pressure from the parametric model. This is the fellowship’s “Hybrid Analyst” posture made executable: the analytical chain from CLD to constraint equation gets handed to autonomous decision agents, with humans in the loop at the highest-leverage decisions.