Semiconductor Choke-Points — System Dynamics Analysis

Key system dynamics findings

R1 Dominance reinforcer

Balancing loops always lose the race

R1 compounds exponentially while B1 and B2 fire linearly after 3–7 and 5–12 year delays. Structural overshoot is not policy failure — it is feedback architecture.

R2 Escalation spiral

R2 is a ratchet, not a cycle

The US–China tech war has no natural dampener. Each control tightening raises China's investment floor. The only deflationary path requires a political variable not in the physical system.

Pain node

Pain is a vector, not a scalar

Four distinct inflow channels (cost, throughput, latent vulnerability, competitive erosion) trigger four different response functions with different delays. Aggregating them loses all timing signal.

Methodology & framework

Driver tree vs. system dynamics

A driver tree is a static dependency map — acyclic, directional, excellent for identifying single points of failure. System dynamics adds feedback loops, stocks and flows, and delay structures. The key gain: it shows why choke-points don't self-correct quickly.

Driver tree asks	System dynamics asks
What depends on what?	What reinforces or counteracts itself?
Where is the vulnerability?	When does the system fail?
Single point of failure	Feedback loop structure
Static snapshot	Dynamic trajectory

Geopolitics as endogenous variable

China's export control decisions and US policy responses are treated as feedback variables inside the model — not external shocks. This reveals how the system generates its own crises through R1 and R2.

Substitution timeline ladder

Each choke-point is assessed across three windows. The investment thesis: hold long positions in materials in the 0–2 yr Critical window with no credible substitution path; monitor triggers that shift nodes to Constrained.

0–2 yr CriticalNo substitute. Enter long.

2–5 yr ConstrainedWatch triggers. Hold.

5–10 yr TransitioningWindow closing. Plan exit.

Five-loop feedback architecture

Loop	Type	Name	Mechanism	Natural dampener?	Investor window
R1	Reinforcing	Dominance reinforcer	China leverage → export controls → Western pain → reshoring → window closing → tighter controls	No — checked only by B1/B2 (delayed)	Now. Before B1 fires.
R2	Reinforcing	Escalation spiral	US controls → China domestic investment → capability growth → US perceives threat → tighter controls	None in physical system	Ratchet. No exit signal.
B1	Balancing	Price correction	Scarcity → price spike → refining investment → supply relief → price falls	Yes — but 3–7 yr delay	Peak scarcity = exit signal
B2	Balancing	Substitution escape	Price pain → R&D → substitute commercialisation → demand relief → price normalises	Yes — 5–12 yr delay	Longest thesis horizon
B3	Balancing	Fab capacity correction	Throughput pain → fab capex → new capacity → output relief	Yes — 3–5 yr delay	Grid queue now the gate

Analysis modules

Module 01 Static baseline

Driver Tree

Five-level dependency map. Level 1 infrastructure → Level 5 HPC output. Substitution timeline ladder per node. Click any node for detailed assessment.

Module 02 Core CLD

Causal Loop Diagram

Full feedback architecture. Five loops, polarity notation, delay markers. Geopolitics treated as endogenous. Clickable nodes with loop explanations.

Module 03 Interactive simulation

Materials Leverage Loop (R1 + B1 + B2)

Stock-and-flow simulation. Adjust export control severity, refining build delay, substitution horizon. Four presets: baseline, fast policy, full escalation, de-escalation.

Module 04 Interactive simulation

Technology War Escalation Spiral (R2)

Ratchet dynamics with perception gap. Adjust China investment multiplier, US perception lag, diplomatic friction. Phase labels: Dominated → Contested → Near-parity.

Module 05 Structural decomposition

Western Pain Node — Decomposed

Why treating "pain" as a scalar loses all timing signal. Four inflow channels, five response functions, one policy threshold. The correct causal structure for modelling investment response timing.

Source confidence hierarchy

Tier	Type	Confidence	Examples
S	Primary data, official filings, peer-reviewed	≥ 90%	ASML annual reports, USGS, IEA, BIS filings
A	Major institutional research with named methodology	75–85%	CSIS, Carnegie Endowment, BCG, JPMorgan, Goldman Sachs
B	Industry analysis, trade press with sourced claims	60–75%	TrendForce, Silicon Analysts, Fusion Worldwide, BMO
C	Market commentary, unverified forecasts	< 60%	Blog analysis, unsourced claims — flagged explicitly

Semiconductor EcosystemChoke-Point System Dynamics