An interactive system dynamics analysis of critical material dependencies, geopolitical feedback loops, and substitution timelines in the global semiconductor and high-performance computing supply chain.
Static analaysis. Runs directly in browser.
| # | Module | Type | Description |
|---|---|---|---|
| 00 | index.html |
Overview | Project landing page, methodology, five-loop architecture reference |
| 01 | 01-driver-tree.html |
Static | Five-level dependency map (L1 infrastructure → L5 HPC output). Click any node for substitution timeline assessment. |
| 02 | 02-cld.html |
Diagram | Full causal loop diagram. 5 feedback loops (R1, R2, R3, B1, B2, B3). Endogenous geopolitics. Clickable nodes and loop cards. |
| 03 | 03-materials-sim.html |
Simulation | R1 + B1 + B2 stock-and-flow. Adjust export control severity, refining build delay (B1), substitution horizon (B2), reshoring strength. Four scenario presets. |
| 04 | 04-escalation-sim.html |
Simulation | R2 escalation spiral. China capability stock vs. US control intensity. Perception lag, diplomatic friction, allied alignment. Four scenario presets. |
| 05 | 05-pain-decomposed.html |
Decomposition | Why "Western semiconductor pain" as a single scalar is a modelling error. Four inflow channels, five response functions, interactive policy threshold simulator. |
China material leverage → export controls → Western pain → reshoring → closing window → tighter controls. No natural dampener. Checked only by B1 and B2, both delayed by years. R1 compounds for 3–7 cycles before any correction fires.
US tech export controls → China domestic semiconductor investment → capability growth → US perceives threat (with 2yr lag) → tighter controls. No natural ceiling. Each iteration raises both stocks' floors. The only deflationary path requires a diplomatic variable not in the physical system.
Material scarcity → price spike → refining investment → Western capacity. Fires after R1 has already compounded. Produces structural overshoot.
Sustained price pain → substitute R&D → viable substitute → demand relief. The only loop that permanently breaks a choke-point. Longest delay; most underinvested.
Throughput pain → capex → new fab → output relief. US grid interconnection queue (5+ years) extends the effective delay to 6–9 years.
The CLD's most common modelling error is treating "Western semiconductor pain" as a single scalar. The correct structure has four distinct inflow channels:
| Channel | Visibility | Response triggered | Structural impact |
|---|---|---|---|
| Cost inflation | Quarterly earnings | Procurement hedging | Low — market noise |
| Throughput constraint | Guidance misses | Capex reshoring (B1) | High |
| Latent vulnerability | Invisible in financials | Strategic stockpiling | Very high — trigger for acute crises |
| Competitive erosion | Geopolitical premium | Export controls (R2) | High — irreversible |
| Position | Choke-point | Window | Entry signal | Exit trigger |
|---|---|---|---|---|
| Long (max conviction) | Gallium, Germanium, REE | 0–2 yr Critical | Export control announcement | Western refinery capacity announcement |
| Long (max conviction) | Helium | 0–2 yr Critical | Hormuz/GCC incident | Fab recycling investment scale |
| Long (high conviction) | Tungsten APT | 0–2 yr | China export controls activation | Almonty Korea ramp + scrap supply inflection |
| Long (medium) | HBM memory capacity | 2–5 yr | Throughput pain in guidance | Micron HBM3E volume ramp |
| Long (structural) | Grid infrastructure | Structural | Now — queue already 2,300 GW | NEPA reform + SMR construction starts |
| Short-watch | Legacy DRAM | Transitioning | HBM cannibalises margin | Demand shift complete |
- Pure HTML5 / CSS3 / vanilla JavaScript
- Chart.js 4.4.1 via CDN for simulation charts
- IBM Plex Mono + IBM Plex Sans via Google Fonts
- No build step, no npm, no frameworks
- All simulations run client-side in the browser
- S-rank (≥90%): ASML annual reports, USGS, IEA, BIS filings
- A-rank (75–85%): CSIS, Carnegie Endowment, BCG, JPMorgan, Goldman Sachs
- B-rank (60–75%): TrendForce, Silicon Analysts, Fusion Worldwide, BMO Capital Markets
- C-rank (<60%): Flagged explicitly where used
The stock-and-flow simulations are qualitative dynamic models, not econometric forecasts. Parameters are calibrated to order-of-magnitude real-world timelines but are not regression-fitted to historical data. The purpose is to reveal structural dynamics (overshoot, ratchet behaviour, delay-driven oscillation), not to generate point forecasts.
China's export control decisions and US policy responses are modelled as feedback variables inside the system — not external shocks. This reveals how the system generates its own crises through R1 and R2. The trade-off: the model cannot capture exogenous diplomatic breakthroughs, which would need to be represented as parameter changes (reduce diplomatic friction in Module 04).
- ASML Holding N.V. Annual Report 2024
- U.S. Geological Survey (USGS). Mineral Commodity Summaries 2024–2025
- International Energy Agency (IEA). Critical Minerals Market Review 2024
- Lawrence Berkeley National Laboratory. Grid Interconnection Queue 2024
- CSIS. Mineral Demands for Resilient Semiconductor Supply Chains. 2025
- CSET Georgetown. Tracing the Emergence of Extreme Ultraviolet Lithography. 2023
- Carnegie Endowment for International Peace. South Korea Semiconductor Vulnerability Assessment. 2026
- JPMorgan Asset Management. Mid-Year Investment Outlook 2026
- Goldman Sachs. Data Centre Power Demand: The New Infrastructure Constraint. 2026
- Fusion Worldwide. Inside the AI Bottleneck: CoWoS, HBM, and 2–3nm Capacity Constraints. 2026
- BMO Capital Markets. Tungsten Market Report Q1 2026
- Barclays Research. Asia Semiconductor Supply Chain: Helium Dependency Analysis. 2024
- TNFD. Water Dependency of the Tech Sector — Case Study. February 2026
Proprietary research synthesis · May 2026 · For institutional use only · Not financial advice