---
id: "lib-2026-10-05-capital-skill-complementarity-beyond-firm"
type: "library"
title: "Capital-Skill Complementarity Beyond Firm Boundaries (w35857)"
lang: "EN"
org: "NBER"
source_type: "Paper"
date: "2026-10-05"
week: "2026-W41"
url: "https://nosov.ai/library/lib-2026-10-05-capital-skill-complementarity-beyond-firm/"
original_url: "https://www.nber.org/papers/w35857"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "NBER working paper by Anders Humlum (Chicago Booth) with National Bank of Belgium and KU Leuven co-authors, linking firm investment, firm-to-firm transactions and worker skills for 3,863 investing firms over 2003-2020."
summary_basis: "full-text"
original_minutes: "120"
tags: ["future of work", "skill demand", "supply chains"]
---

# Capital-Skill Complementarity Beyond Firm Boundaries (w35857)

Tier A (Strong evidence): NBER working paper by Anders Humlum (Chicago Booth) with National Bank of Belgium and KU Leuven co-authors, linking firm investment, firm-to-firm transactions and worker skills for 3,863 investing firms over 2003-2020.

Summary based on the full text.

## In short

NBER Working Paper 35857 (October 2026) by Gert Bijnens (National Bank of Belgium), Anders Humlum (University of Chicago Booth) and Stijn Vanormelingen (KU Leuven). Firm-level studies have found that companies buying new machinery do not change their own skill mix, which contradicts the long-held idea that machines raise demand for skilled labour. Using Belgian data, the authors show the skill demand does rise, but at external suppliers: integrators, machine vendors and later software and service firms. Measured on the investing firm's payroll alone, the effect is invisible or even reversed.

## Key numbers

- 45%: rise in investing firms' total demand for high-STEM workers via supplier purchases after a machinery spike
- <5%: rise in in-house hiring after the same spike, spread evenly across skill groups
- 13.9% vs 5.0%: high-STEM share of machine suppliers' workforce vs investing firms' in-house workforce
- 3.1 log points: rise in the high-STEM wage premium from machinery growth 2003-2021 in the estimated model

## What it reports

- After a machinery investment spike, purchases from suppliers raise total demand for high-STEM workers by 45%, three times the rise for other skill groups; in-house hiring raises it by less than 5%, evenly across groups.
- Counting in-house plus external labour, high-STEM employment rises 50% in the spike year (11-20% for other groups) and stays 33% above baseline four years later.
- A typical spike more than triples machinery stock per worker, from about 19 to 61 thousand euros; total fixed investment peaks near 20% of sales in that year.
- The average spiking firm employs 39.7 FTE in-house and buys the equivalent of 22.9 FTE externally, 3.6 of them from machine suppliers.
- Machine suppliers' workforces are 13.9% high-STEM against 5.0% for investing firms, and they pay about 3.6% higher wages at a fixed skill mix.
- External demand comes first from integrators (about +15% high-STEM demand in the spike year), then shifts to complementary services such as software.
- The effect is stronger for smaller firms, services firms, lumpier investments and machinery from suppliers new to the firm; it is absent for intangible investment.
- In the equilibrium model, machinery growth from 2003 to 2021 raises relative demand for high-STEM workers by 8.8 log points and their wage premium by 3.1. Integration accounts for 36% of the rise in relative demand while employing 4.3% of high-STEM workers.

## How they got it

The authors link Belgian firm-level data on capital investment, buyer-supplier transactions and worker education records for 2003-2020. External employment is measured by proportionality: if a firm buys 10% of a supplier's sales, it is assigned 10% of that supplier's workforce by skill group. A stacked difference-in-differences design compares 3,863 firms at their first machinery spike with not-yet-spiking firms in the same 3-digit industry, and the estimates feed a calibrated general-equilibrium model.

## Limits

- The proportionality assumption is an approximation; the authors validate it for specialised suppliers, but it may fit general suppliers less well.
- Belgium imports its machines, so labour spent building them abroad is not counted; results come from one small open economy.
- The data end in 2020-2021, before generative AI; applying the result to AI adoption is an inference.

## Why it matters for companies

When a firm automates, the skilled work often lands at vendors and integrators, not on its own payroll. Internal headcount data will therefore understate how much specialised expertise a technology rollout consumes, and where that expertise sits. For AI programmes, the pattern suggests tracking which capabilities a company buys through partners each year and whether it can afford to keep them external as the stack grows. Smaller firms rely on outside skills most, which increases their dependence on a few suppliers.

Original: https://www.nber.org/papers/w35857