G7 SME AI policy pluralism
The OECD G7 discussion paper’s country profiles reveal systematic variation in how the seven advanced economies are trying to push AI into their SME bases. The variation is interesting because it doesn’t map cleanly onto the existing digital-empires-tripartite frame (US market / China state / EU rights). It cuts orthogonally in some places and confirms the frame in others.
The four observable models
- Market-led + thin federal SME programmes (US). Federal role limited to enabling infrastructure (NAIRR), standards (NIST), seed funds (SBIR/STTR), and an SBA AI Resource Hub. Most capital flows through private VC. The 2025 AI Action Plan calls for reducing regulatory burden. Workforce policy: extend tax code to qualify AI literacy as educational assistance.
- State-coordinated industrial policy + sovereignty (France, Italy, partially Germany). Heavy use of national strategies tied to industrial policy. Big-ticket sovereign-compute investments (France EUR 109bn AI pledge, Italian IT4LIA AI Factory, German Jupiter AI Factory). Subsidy-driven SME support (IA Booster, Transition 5.0, KI4KMU). EuroHPC AI Factory network as the binding multilateral layer.
- State-coordinated + soft-law (Japan, UK). Strong central direction (Japan’s AI Act, UK’s AI Opportunities Action Plan + Modern Industrial Strategy) but with light regulation. Japan: large capex commitments (USD 65bn), Rapidus semiconductor consortium, soft-law sectoral guidance, catalogue subsidies. UK: GBP 2bn Compute Roadmap, AI Growth Zones, AIRR expansion, sector-specific BridgeAI.
- Regionally-delivered federal (Canada). Federal funding + regional delivery via RDAs. CAD 2.4bn over five years; AI Compute Access Fund (CAD 300m); CIFAR-coordinated Pan-Canadian AI Strategy. Strong emphasis on technological sovereignty and SME compute access.
Does this confirm or complicate the tripartite frame?
Partially confirms. The US really is the most market-driven; the EU members (France, Germany, Italy) really do gravitate toward state-coordinated, rights-respecting frameworks; the broad regulatory posture matches brussels-effect-and-mirage.
But it complicates in three ways:
- Japan as a fourth model. Japan combines state coordination with soft law and turnkey subsidies. It is neither market-led nor rights-led nor state-controlled in the China sense. The Hiroshima AI Process (HAIP) — a Japanese G7-presidency initiative that produced the International Code of Conduct for Organisations Developing Advanced AI Systems — positions Japan as a normative leader without EU-style hard regulation. This is a genuine fourth axis. See divergent-value-stack-optima.
- The UK is its own thing. Post-Brexit it has explicitly chosen “state-led + market-enabled” rather than Brussels-style horizontal regulation. The Modern Industrial Strategy + AI Action Plan combination looks more Japanese than European.
- Canada bridges US and EU. Federally funded, regionally delivered, sovereignty-tinged but private-sector-anchored. The “Canadian Sovereign AI Compute Strategy” explicitly uses sovereignty language — but executes through partnerships with private compute.
Common ground
Despite different rhetorics, G7 instruments converge on a narrow set of levers:
- Compute access (Canada AI Compute Access Fund; UK AIRR; US NAIRR; Germany Jupiter; France AI2F; Italy IT4LIA; Japan ¥500bn subsidy programmes)
- Sectoral SME subsidies (Italy Transition 5.0; France IA Booster; Japan Catalog-Type; UK BridgeAI; Germany KI4KMU)
- Skills/training programmes (UK Flexible AI Upskilling Fund; France Osez l’IA AI Academy; Germany AI Studios; Japan METI guidebooks; US SBA AI Resource Hub)
- Innovation hubs / regional delivery (Italy Competence Centres; Germany Mittelstand-Digital; France IA-Clusters; Canada RDAs; US Small Business Development Centers)
The convergence on instruments alongside divergence in framing is a finding in its own right: G7 governments quote different ideologies but reach for the same toolkit when actually pushing AI into SMEs.
The barrier-and-support survey
Table 4 in the OECD survey adds another layer of pluralism — different SMEs report different perceived barriers and preferred supports:
| Country | Top reported barrier | Top requested support |
|---|---|---|
| Canada | Generative AI not suited to our work | Training |
| Germany | Clients would not approve | Information campaigns |
| Japan | Employees lack skills | Financial assistance |
| UK | Generative AI not suited to our work | Training |
Germany’s response is the most striking: client/customer disapproval of AI is the leading barrier — a trust- and reputation-driven constraint, not a skills or money one. This is uniquely European in flavour and provides empirical anchor for the rights/trust posture documented in bradford-digital-empires.
Implications for the working hypotheses
- divergent-value-stack-optima sharpens: at the SME-policy layer, four optima exist (not three), with Japan and the UK as distinct.
- digital-empires-tripartite partially holds at the level of rhetoric and regulatory framework, less so at the level of operational instruments — these converge.
- diffusion-vs-innovation: every G7 country has now publicly committed to diffusion as a strategic objective, regardless of bloc. That is new relative to a 2020 baseline.
- H1_L0-L7-ladder: nation-states differ less on what layer they target than on how they finance access to it.