0 cumulative citations
View corpus contextGlobal short-run supply responsiveness of commodity-grade silicon collapsed after early 2024 as AI infrastructure demand surged, implying price rises alone are unlikely to quickly unlock more output; producers appear to have exhausted spare capacity, raising near-term material bottleneck risks for AI rollout.
AI Infrastructure and Critical Minerals : Short-Run Silicon Supply Elasticity and Implications for AI-Relevant Mineral Markets
openalex
quasi_experimental
medium evidence
7/10 relevance
Summary only
summary available; pdf_status=error
Source
PDF
Using monthly global data and a structural-break IV approach, the thesis finds that the short-run price elasticity of supply for commodity-grade silicon was positive before early 2024 but fell sharply thereafter, consistent with producers moving from spare capacity to operating near limits as AI infrastructure expanded.
Citation observations
Cumulative provider counts captured on specific dates; providers are never combined.
This thesis estimates the short-run price elasticity of supply for commodity-grade silicon and examines whether it changed during the expansion of artificial intelligence (AI) infrastructure that began in late 2022. Using monthly global data from January 2015 to February 2026, a production proxy is constructed from international trade statistics and estimated using ordinary least squares with Newey-West standard errors, instrumental variables to address endogeneity, and a structural-break specification. The estimated supply elasticity is positive prior to the break but weakens sharply from early 2024, approximately a year and a half after the AI boom began. The structural break, a continuous interaction term for AI intensity, and a robustness check of the break date all point in the same direction. The results suggest that silicon producers had spare capacity prior to the AI boom but operated near their limits afterwards, implying that price signals alone are unlikely to generate sufficient additional supply in the short run.
Summary
Main Finding
The thesis estimates that the short-run price elasticity of supply for commodity-grade silicon was positive before the AI infrastructure expansion but fell sharply starting in early 2024 (about 18 months after the AI boom began in late 2022). This suggests producers moved from having spare capacity to operating near limits, so short-run price signals alone are unlikely to induce meaningful additional silicon supply.
Key Points
- Data span: monthly global observations, January 2015–February 2026.
- Production proxy: constructed from international trade statistics to approximate global silicon output.
- Econometric approaches: OLS with Newey–West standard errors, instrumental variables to address endogeneity, and a structural-break specification.
- Structural evidence: a clear weakening of the estimated short-run supply elasticity beginning in early 2024.
- Complementary specifications: a continuous interaction term for measured AI intensity and robustness checks that vary the break date all point to the same conclusion.
- Interpretation: prior to the AI-driven demand surge producers possessed spare capacity; after the surge they operated near capacity limits, constraining short-run responsiveness to price increases.
Data & Methods
- Data: Monthly, global-level trade-based production proxy for commodity-grade silicon (Jan 2015–Feb 2026).
- Primary estimation: OLS with Newey–West standard errors to correct for serial correlation and heteroskedasticity in monthly time series.
- Endogeneity: Instrumental-variables estimation used to address potential reverse causality or omitted-variable bias (thesis contains details on instruments).
- Structural-break analysis: explicit break specification to test for a change in the supply elasticity timed around the AI expansion; also estimated a continuous interaction between AI intensity and price to capture gradual changes.
- Robustness: alternative break dates and specifications tested; all produced qualitatively consistent results indicating a drop in short-run elasticity after the AI demand increase.
Implications for AI Economics
- Short-run supply inelasticity: With producers near capacity, short-run price increases will likely not translate into large additional silicon output; this raises the risk of price spikes and supply bottlenecks for AI hardware components that depend on commodity-grade silicon.
- Cost forecasting and deployment timelines: AI hardware cost projections should incorporate reduced short-run supply responsiveness and likely lead times for capacity expansion, which can slow deployment or increase effective compute costs.
- Policy and market responses: Because price signals alone may be insufficient in the short run, non-price interventions (e.g., capacity investment incentives, long-term offtake contracts, strategic reserves, expedited permitting for new plants) may be needed to stabilize supply and prices.
- Investment incentives: The observed shift increases the value of forward-looking investments in silicon production capacity and alternative supply-chain resilience measures (recycling, substitution, geographic diversification).
- Research and modeling: Macroeconomic and sectoral models of AI growth should treat silicon supply as potentially binding in the short run and model dynamic adjustment with realistic lags to capture the effects documented here.
Limitations to keep in mind: results concern short-run responsiveness (not long-run supply), rely on a trade-based production proxy (measurement error risk), and the IV strategy and break timing are described at length in the thesis for readers wanting technical detail.
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — The study uses long monthly series, multiple econometric approaches (Newey–West, IV, and structural-break tests) and robustness checks that consistently indicate a weakening supply elasticity after early 2024; however, causal attribution to the AI expansion is uncertain because the instrument is not described here, the production proxy (trade flows) may be noisy or endogenous, and other concurrent global shocks (energy prices, policy changes, supply-chain disruptions) could explain the break.
Methods Rigormedium — Methodologically the paper applies appropriate time-series corrections, uses IV to tackle endogeneity, and explicitly tests for structural breaks and alternative break dates, which is good practice; but rigor depends critically on the validity and strength of the instrument(s), the credibility of the production proxy, and whether the break detection was pre-specified rather than post-hoc — information not provided in the summary.
SampleMonthly global series for commodity-grade silicon from January 2015 to February 2026 (~133 months), including global monthly prices and a production proxy constructed from international trade statistics (aggregated imports/exports); analysis is performed at the aggregate (global) level rather than firm- or plant-level.
Themesadoption innovation
IdentificationTime-series estimation of supply using monthly global prices and a production proxy (constructed from international trade statistics) from Jan 2015–Feb 2026; baseline OLS with Newey–West standard errors, instrumental variables to address endogeneity of price/production, and a structural-break specification (binary/continuous interaction for AI intensity) with robustness checks on break date.
GeneralizabilityAggregate global analysis masks regional/country heterogeneity in production capacity and policy environments, Commodity-grade silicon is not the same as semiconductor-grade silicon used in many AI chips, so results may not generalize to semiconductor supply chains, Short-run elasticity estimates may not apply to long-run supply responses once capacity investment occurs, Production proxy based on trade flows may mismeasure true output (stocks, unrecorded flows, or changes in inventory can bias estimates), Concurrent global shocks (energy, geopolitics, trade policy, COVID aftereffects) during 2022–2026 may confound attribution to AI infrastructure expansion
Claims (8)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| The thesis estimates the short-run price elasticity of supply for commodity-grade silicon. Market Structure | null_result | price elasticity of supply for commodity-grade silicon |
Reading fidelity
high
Study strength
medium
|
n=133
|
| The analysis uses monthly global data from January 2015 to February 2026. Other | null_result | data sample / time coverage |
Reading fidelity
high
Study strength
high
|
n=133
|
| A production proxy is constructed from international trade statistics and the supply elasticity is estimated using OLS with Newey-West standard errors, instrumental variables to address endogeneity, and a structural-break specification. Other | null_result | production proxy estimation / econometric approach |
Reading fidelity
high
Study strength
medium
|
n=133
|
| The estimated supply elasticity is positive prior to the structural break but weakens sharply from early 2024 (about a year and a half after the AI boom began). Market Structure | negative | short-run price elasticity of supply for commodity-grade silicon (pre- vs post-break) |
Reading fidelity
high
Study strength
medium
|
n=133
|
| The structural break, a continuous interaction term for AI intensity, and a robustness check of the break date all point in the same direction (weakened supply responsiveness). Market Structure | negative | consistency of estimated change in supply elasticity across model specifications |
Reading fidelity
high
Study strength
medium
|
n=133
|
| The results suggest silicon producers had spare capacity prior to the AI boom but operated near their limits afterwards. Market Structure | negative | industry capacity utilization / ability to expand supply in short run |
Reading fidelity
medium
Study strength
medium
|
n=133
|
| Price signals alone are unlikely to generate sufficient additional supply of commodity-grade silicon in the short run after the AI-driven demand expansion. Market Structure | negative | short-run supply responsiveness to price signals |
Reading fidelity
medium
Study strength
speculative
|
n=133
|
| The AI infrastructure expansion that potentially affected silicon demand began in late 2022. Adoption Rate | null_result | timing of AI infrastructure expansion |
Reading fidelity
high
Study strength
low
|
not reported
|