IESET.
Hypotheses·trade·chips_act_2022_semiconductor_capacity_2024_2027

The 2022 US CHIPS and Science Act (~$52bn manufacturing subsidy + 25% ITC) produced a measurable acceleration in announced and realised US semiconductor fab construction starts and capacity additions over 2023- 2027, narrowing the US share-of-global advanced-logic capacity gap relative to Taiwan + South Korea, but the realised capacity addition by 2027 falls materially short of both the headline announcements (Intel Ohio, TSMC Arizona, Samsung Texas, Micron New York) and the pre-CHIPS US share-recovery rhetoric.

Asian competitor capex (TSMC Taiwan, Samsung Korea, SMIC China) continued at or above pre-CHIPS rates over the same window, so the US share of global advanced-node (≤7nm) wafer capacity rises modestly rather than dramatically.

INCONCLUSIVEengine/runs/chips_act_2022_semiconductor_capacity_2024_2027

INCONCLUSIVE_DATA_PENDING — insufficient obs after listwise deletion (9)

confidence cueResult card produced; verdict unclassified.

policy briefCoverage too thin

In ordinary language

When countries open more of the economy to trade and competition, do people end up with better long-run income or productivity outcomes?

plain answer

This test cannot make a firm call yet. insufficient obs after listwise deletion (9)

why it matters

This matters because trade claims should change belief only when they survive a pre-declared empirical test.

how the test works

It compares 8 country or place units from 2018 to 2027, using a did callaway santanna design, with fixed effects for country and year.

what was measured
What changed
  • Us post chips dummy
  • Us chips eligible advanced node interaction
What we checked
  • Log us semiconductor fab construction value added
  • Log us advanced node wafer capacity kwspm
  • Us share global logic capacity
what this does not prove

A single test is not the whole truth. It narrows the claim under a specific sample, time period, and method. Strong policy conclusions need the pattern to survive nearby tests, alternative data, and serious objections.

verification

No evidence packet has been generated yet.

Results

engine/runs/chips_act_2022_semiconductor_capacity_2024_2027
1007550250201820232027USATWNKORJPNCHNDEUIRL
illustrative sketch · run pending
No coefficients yet. When the model fires, this chart will show log_us_semiconductor_fab_construction_value_added across 8 sampled countries over 2018–2027.
The shapes above are stylised — none of the lines are real data.
Placeholder for chips_act_2022_semiconductor_capacity_2024_2027. Published chart will be generated from engine/runs/chips_act_2022_semiconductor_capacity_2024_2027/chart_data.json.

Pre-registration

pre-registered
first-spec commit 098ce96 · 2026-04-30T12:57:33Z
run generated · 2026-05-04T12:34:13Z

The 2022 US CHIPS and Science Act (~$52bn manufacturing subsidy + 25% ITC) produced a measurable acceleration in announced and realised US semiconductor fab construction starts and capacity additions over 2023- 2027, narrowing the US share-of-global advanced-logic capacity gap relative to Taiwan + South Korea, but the realised capacity addition by 2027 falls materially short of both the headline announcements (Intel Ohio, TSMC Arizona, Samsung Texas, Micron New York) and the pre-CHIPS US share-recovery rhetoric. Asian competitor capex (TSMC Taiwan, Samsung Korea, SMIC China) continued at or above pre-CHIPS rates over the same window, so the US share of global advanced-node (≤7nm) wafer capacity rises modestly rather than dramatically.

Falsification criterion — what would disprove this

set before the run · honoured after

This hypothesis is considered falsified if:

Not supported if (a) the US × post-2022 effect on semiconductor construction is not statistically positive at p<0.10 after country + year FE, OR (b) US share of global advanced-node logic capacity does not rise by at least 5 percentage points by end-2027, OR (c) realised capacity addition by 2027 is less than 30% of announced 2022-2024 capacity-addition headlines, OR (d) Asian competitor capex fell materially below pre-CHIPS trend (refuting the "narrowing the gap" framing — it would imply the US gain is a relocation of capex rather than expansion of the policy-target). If only construction-start activity rises but realised capacity by 2027 is well short of announcements, classify as "announcement bubble."

formal test & threshold
test:      chips_act_capacity_did_panel
threshold: US × post-2022 β on log construction > 0.20 at p<0.10 AND US advanced-node share rises ≥ 5pp by 2027 AND realised capacity ≥ 50% of announcement headline AND TSMC + Samsung capex remained flat-to-positive (no withdrawal). ≥3 of 4 hit = supported; 2 = mixed.

Method

Template
did_callaway_santanna
Fixed effects
country, year
Clustering
country
Sample
8 countries · 2018 – 2027
Evidence type
causal

Primary specification: differences-in-differences with treatment = USA post-2022, control = TWN, KOR, JPN, CHN, DEU. Outcome: log semiconductor-construction value, log advanced-node capacity, US share of global logic capacity. Test whether the US × post-2022 coefficient is significantly positive after country + year FE. Secondary specification: synthetic control for USA on semiconductor construction spending using donor pool of non-treated peers; the counterfactual USA carries forward pre-CHIPS construction trend. Decomposition asks: of (a) construction starts, (b) realised capacity, (c) global share, (d) outlay disbursement, which exceeded their pre- CHIPS trend by ≥1 sigma by 2027? Realised capacity is the binding measure; announcements + obligated funds are leading. Known limitations: (1) 2024-2027 outcomes still partially forecast; v1 reports through latest available year, v1.1 updates as fab construction completes (Intel Ohio delayed to 2027-2028; TSMC AZ Phase 1 ramping 2025). (2) Asian competitor capex is endogenous — TSMC + Samsung accelerated domestic capex partly in response to US export-control pressure rather than independent of it. (3) Advanced-node (≤7nm) is the policy-target segment; trailing-edge capacity (28nm+) trends differently and a wide-aperture "all semiconductor" outcome would dilute the policy signal. (4) BIS October 2022 / 2023 export controls are a confounder bundled with the subsidy — separating subsidy-induced reshoring from export-control-induced reshoring is hard.

Data

VariableSourceTransform
log_us_semiconductor_fab_construction_value_added
outcome
fred:TLPRVCONStier 1
log
log_us_advanced_node_wafer_capacity_kwspm
outcome
constructed:SEMI World Fab Forecast / TechInsights advanced-node (≤7nm logic) capacity by location (kWspm = thousand wafers per monttier 5
log
us_share_global_logic_capacity
outcome
constructed:SEMI / Boston Consulting Group + SIA (Semiconductor Industry Association) annual capacity-share report. Manual-drop pendtier 5
level
log_taiwan_korea_china_semiconductor_capex
outcome
constructed:TSMC + Samsung + SK Hynix + SMIC reported annual capex (company filings). Cross-check against Korea industrial-productiotier 5
log
us_chips_act_obligated_outlay_usd
outcome
constructed:Department of Commerce / CHIPS Program Office obligation tracker. Manual-drop pending.tier 5
log
us_post_chips_dummy
treatment
constructed:indicator = 1 for USA from 2022-08 (CHIPS Act enactment) onwards; 0 otherwise.tier 5
indicator
us_chips_eligible_advanced_node_interaction
treatment
constructed:interaction of post-CHIPS dummy with advanced-node (≤7nm logic) eligibility. Identifies whether the subsidy is binding otier 5
indicator
log_global_semiconductor_demand
control
constructed:WSTS (World Semiconductor Trade Statistics) global billings. Manual-drop pending.tier 5
log
log_real_gdp
control
world_bank_wdi:NY.GDP.MKTP.KDtier 2
log
real_interest_rate
control
fred:DFII10tier 1
level
us_export_control_intensity_china
control
constructed:BIS export-control rule count (October 2022, October 2023, December 2024 expansions) — count of HS6 / ECCN line items retier 5
level

● ready  · ● pending  · ● reconstruct-needed

Detailed result card

Result card — chips_act_2022_semiconductor_capacity_2024_2027

Verdict: INCONCLUSIVE_DATA_PENDING — insufficient obs after listwise deletion (9)

Pre-registration

  • Claim: The 2022 US CHIPS and Science Act (~$52bn manufacturing subsidy + 25% ITC) produced a measurable acceleration in announced and realised US semiconductor fab construction starts and capacity additions over 2023- 2027, narrowing the US share-of-global advanced-logic capacity gap relative to Taiwan + South Korea, but the realised capacity addition by 2027 falls materially short of both the headline announcements (Intel Ohio, TSMC Arizona, Samsung Texas, Micron New York) and the pre-CHIPS US share-recovery rhetoric. Asian competitor capex (TSMC Taiwan, Samsung Korea, SMIC China) continued at or above pre-CHIPS rates over the same window, so the US share of global advanced-node (≤7nm) wafer capacity rises modestly rather than dramatically.
  • Falsification rule: Not supported if (a) the US × post-2022 effect on semiconductor construction is not statistically positive at p<0.10 after country + year FE, OR (b) US share of global advanced-node logic capacity does not rise by at least 5 percentage points by end-2027, OR (c) realised capacity addition by 2027 is less than 30% of announced 2022-2024 capacity-addition headlines, OR (d) Asian competitor capex fell materially below pre-CHIPS trend (refuting the "narrowing the gap" framing — it would imply the US gain is a relocation of capex rather than expansion of the policy-target). If only construction-start activity rises but realised capacity by 2027 is well short of announcements, classify as "announcement bubble."

Estimate (Callaway-Sant'Anna staggered DiD, TWFE approximation)

  • Error: insufficient obs after listwise deletion (9)

Variables resolved

  • fred:TLPRVCONS → log_us_semiconductor_fab_construction_value_added (outcome, n=34)
  • constructed: TSMC + Samsung + SK Hynix + SMIC reported annual capex (company filings). Cross-check against Korea industrial-production semiconductor index (oecd or kosis source). → log_taiwan_korea_china_semiconductor_capex (outcome, n=80)
  • constructed: indicator = 1 for USA from 2022-08 (CHIPS Act enactment) onwards; 0 otherwise. → us_post_chips_dummy (treatment, n=80)
  • world_bank_wdi:NY.GDP.MKTP.KD → log_real_gdp (controls, n=14066)
  • fred:DFII10 → real_interest_rate (controls, n=192)

Missing data

  • constructed: SEMI World Fab Forecast / TechInsights advanced-node (≤7nm logic) capacity by location (kWspm = thousand wafers per month). Manual-drop pending under data/manual/derived/. (outcome)
  • constructed: SEMI / Boston Consulting Group + SIA (Semiconductor Industry Association) annual capacity-share report. Manual-drop pending. (outcome)
  • constructed: Department of Commerce / CHIPS Program Office obligation tracker. Manual-drop pending. (outcome)
  • constructed: interaction of post-CHIPS dummy with advanced-node (≤7nm logic) eligibility. Identifies whether the subsidy is binding on the targeted leading-edge segment vs trailing-edge. (treatment)
  • constructed: WSTS (World Semiconductor Trade Statistics) global billings. Manual-drop pending. (controls)
  • constructed: BIS export-control rule count (October 2022, October 2023, December 2024 expansions) — count of HS6 / ECCN line items restricted. Cross-check against fred:IRPRTS003M086NEST commodity export indices. Manual-drop pending. (controls)

Generated by scripts/run_did_callaway_santanna.py at 2026-05-04T12:34:13+00:00

Strongest opposing argument

Every hypothesis ships with its charitable opposing argument. The framework earns credibility by handling objections at their strongest, not weakest.

Notes

Data readiness: - FRED construction series: ready - WDI GDP + FRED interest: ready - SEMI World Fab Forecast / TechInsights / SIA capacity panels: manual-drop pending under data/manual/derived/ - WSTS billings: manual-drop pending - CHIPS Program Office obligation tracker: manual-drop pending - Asian competitor capex (TSMC, Samsung, SK Hynix, SMIC filings): manual-drop pending Adjacent existing hypothesis (industrial_policy_semiconductor_chips_act_effectiveness) takes a different angle (broader effectiveness framing); this spec narrows to the realised-capacity vs announced-capacity decomposition with a fixed 2027 endpoint.

Authored framework. Read the transparency note.