Where AI Value Lands

Sectoral AI Diffusion Pattern

4 min read

Sectoral AI Diffusion Pattern

AI use is concentrated where AI is made. In 2024, ~45% of firms in the ICT sector and ~26% in Professional, scientific & technical activities used AI across OECD countries — three to six times the adoption rate of Construction (7.2%), Accommodation & food services (7.8%), or Transportation & storage (9.2%). The ICT sector also scores top-quartile on every dimension of AI intensity (human capital, innovation, exposure, use). Source: oecd-sme-ai-adoption-2025.

This is the sectoral compounding factor behind the sme-ai-adoption-gap. Even comparing firms of similar size and age, sectoral composition shifts adoption rates by a multiple.

The pattern

Three concentric layers:

  1. AI-intensive sectors. ICT (Media, IT services, Telecommunications), Scientific R&D, Legal & accounting, Finance & insurance. Top-quartile across all four AI-intensity dimensions in the Calvino et al. (2024) sectoral taxonomy.
  2. Mid-intensity sectors. Manufacturing of transport equipment, machinery, computer & electronics; Wholesale & retail. Second or third quartile. AI-relevant but not AI-native.
  3. Low-intensity sectors. Construction, Hotels & food services, Textiles, Wood & paper, Transportation & storage. Bottom-quartile across most dimensions. These are the bulk of the SME population in OECD economies.

The same hierarchy holds in every G7 country: ICT adoption rates are typically more than 3× manufacturing’s.

What “AI intensity” packages together

The Calvino et al. multidimensional indicator distinguishes:

  • AI human capital — share of workers in AI-relevant occupations.
  • AI innovation — patents and publications in the sector.
  • AI exposure (barrier-adjusted) — share of tasks technically exposed to AI, weighted by adoption barriers.
  • AI use — share of firms adopting.

The four scores correlate but are not identical. ICT and IT services score top-quartile on all four — they make the technology, adopt it earliest, and have the workers to deploy it. Manufacturing sub-sectors are split: Computer & electronics is top-quartile, transport equipment / machinery is middle, food / textile / wood is bottom. The decomposition lets us see that “manufacturing” as a single category hides a 3–4× internal range.

The good news

The increase is an all-sector phenomenon. Between 2021 and 2024, OECD-wide adoption roughly doubled in most sectors:

  • ICT: +20 percentage points (from ~25% to ~45%)
  • Professional services: +13 pp
  • Wholesale trade: +8 pp
  • Manufacturing: ~+5 pp (a 70% relative increase from a low base)
  • Transportation & storage: ~+3.5 pp (a 60% relative increase)

The largest absolute increases are in already-high-adoption sectors — i.e., the sectoral gap is widening in absolute terms even as it narrows in relative terms. ICT pulled away from manufacturing by ~15 pp in three years.

Why this matters for the U-curve

  • The “long-tail application layer” is concentrated in low-intensity sectors. If the bottom-right of H2_u-curve-of-value is “SMEs in construction, hospitality, transport, retail” — the literal majority of European economic activity by employment — those sectors are at single-digit AI adoption rates. The U’s bottom-right is thin where the population is thickest.
  • The “shaping” sectors are also the producing sectors. Calvino et al. note that ICT and professional services are simultaneously the largest adopters and the largest suppliers of AI. This is a recursion: the part of the economy that captures most AI surplus today is the part that builds the technology. This is consistent with the capital-labor-divergence story — value flows from the rest of the economy into the AI-producing sector via vendor relationships.
  • Diffusion S-curve. Per diffusion-vs-innovation, the GPT winner is whoever absorbs the technology fastest across all productive sectors. The sectoral pattern shows the OECD is still early on that curve — manufacturing, construction, and transport are at the bottom of the S.

Tensions

  • Sector-mix vs sector-effect. Larger firms tend to be in different sectors than smaller firms; some of the sme-ai-adoption-gap is a sector compositional effect. But OECD work (Calvino & Fontanelli 2023) shows the size gap persists within sector — sector is a compounding factor, not the full explanation.
  • AI intensity ≠ AI value-capture. The Computer & electronics sector scores top-quartile on use and intensity, but its margin profile is not obviously richer than Media or IT services. The sectoral taxonomy measures adoption, not capture.
  • Definitions drift. “AI use” in 2021 surveys covers a narrower technology set than “AI use” in 2024 surveys (generative AI is the most striking addition). Trend comparisons should be read with this in mind.

Graph View

Backlinks