OECD Paper Flags Persistent AI Adoption Gap Between Small Firms and Large Enterprises

The productivity case for AI is well established at this point. What's less settled is who actually gets to capture it. A discussion paper released by the OECD on December 9, prepared at the request of Canada's 2025 G7 Presidency, puts a number on a gap many analysts have long suspected: small and medium-sized enterprises are adopting artificial intelligence at meaningfully lower rates than large firms, and lower rates than they adopted previous waves of digital technology.

That gap matters because SMEs are not a rounding error in most G7 economies. They represent the bulk of firms and a substantial share of employment across the bloc. If AI diffusion follows the pattern of prior technology cycles, where scale and capital determine who benefits first, the result could be a widening productivity split between large incumbents and the small businesses that make up most of the private sector.

Canada made SME AI adoption a named priority of its G7 Presidency this year, and the OECD paper is explicitly framed as groundwork for a proposed Blueprint for SME AI Adoption. In other words, this isn't an academic exercise. It's meant to feed directly into policy design discussions among G7 governments.

Why the Adoption Gap Persists

The OECD Secretariat's paper does something useful beyond restating the gap: it introduces a taxonomy of SME AI adopters based on three dimensions, digital maturity, complexity of use, and scope of application. That's a meaningful shift in framing. Rather than treating "SME AI adoption" as a single binary outcome, firms are sorted along a spectrum, which allows policymakers to target interventions at firms where they are, rather than designing one-size-fits-all programs.

This distinction matters in practice. A small manufacturer experimenting with a single AI-enabled scheduling tool is a very different case from a services firm that has layered machine learning into core operations. Blanket subsidies or generic training programs tend to underserve both ends of that spectrum. The OECD's case studies, drawn from across G7 countries, illustrate just how varied these adoption pathways can be, even within similar sectors and similar firm sizes.

The paper identifies four enablers as prerequisites for successful SME AI adoption: connectivity, data and algorithms and compute, skills, and finance. None of these are new themes in the digitalization literature, but their persistence is itself the finding. SMEs have historically struggled with the same four bottlenecks during earlier waves of digital transformation, from e-commerce to cloud computing. AI appears to be reproducing that pattern rather than disrupting it.

Connectivity remains a basic constraint in some G7 regions, particularly for firms outside major urban centers. Compute and data access are capital-intensive barriers that disproportionately burden smaller firms, which lack the balance sheets of large enterprises to invest in infrastructure or licensing. Skills gaps are compounded by the tight labor market for AI-literate talent, which large firms can outbid smaller competitors for. And finance, perhaps the most structural constraint, reflects the broader reality that SMEs often face higher borrowing costs and more limited access to capital markets than large firms, a disadvantage that predates AI but that AI investment needs may now be sharpening.

The paper's framing suggests the policy response needs to be differentiated rather than uniform. A connectivity subsidy helps little if the binding constraint for a given firm is finance. A generic AI literacy course helps little if the firm already has sophisticated digital capabilities but cannot access affordable compute. The taxonomy approach, in theory, lets governments match specific support instruments to specific adoption barriers, which is a more disciplined approach than past broad-based digitalization pushes.

It's also worth noting what the paper does not claim. It does not assert that AI adoption automatically translates into productivity gains for SMEs, nor does it offer a timeline for closing the gap with large firms. The tone throughout is diagnostic rather than promotional, consistent with the OECD's role as a secretariat feeding technical input into a G7 policy process rather than an advocacy body with a stake in the outcome.

The Bottom Line

For investors and business leaders tracking the broader AI diffusion story, this paper is a useful corrective to narratives built entirely around hyperscalers and frontier labs. The real economic impact of AI, if it materializes broadly, will depend heavily on whether the tens of millions of small and medium-sized firms across G7 economies can actually adopt the technology in ways that improve output. Right now, the evidence says they're not keeping pace with larger firms, and the gap echoes prior digital technology cycles where capital and scale determined who benefited first.

The practical takeaway is that AI adoption in the SME segment is not primarily a technology problem. It is a structural problem involving connectivity infrastructure, financing access, data and compute costs, and skills availability, the same bottlenecks that have constrained SME digitalization for years. Governments that want broad-based productivity gains from AI will need differentiated policy tools rather than blanket programs, matched to where individual firms sit on the adoption spectrum the OECD has outlined.

Whether the proposed G7 Blueprint for SME AI Adoption translates this diagnosis into effective policy remains an open question. But the paper itself is a reasonable starting point: it identifies the right variables, resists overstating the case, and gives policymakers a framework rather than a slogan. For a technology narrative that has been heavy on hype and light on structural analysis, that's a modest but genuine contribution.