The Next AI Advantage Will Be Organisational, Not Computational

Artificial intelligence is becoming easier to access and considerably harder to differentiate through access alone.

On 14 October, the European Commission’s Joint Research Centre published new evidence showing that AI adoption across ten strategic EU sectors increased from 7% of companies in 2021 to 18% in 2025. Across the wider EU economy, adoption increased from 8% to 20% over the same period. Joint Research Centre

The headline suggests rapid diffusion. The detail is more consequential.

Much of the increase has been driven by increasingly accessible general-purpose AI, particularly systems generating or analysing text and audio. In many businesses, applications remain concentrated in marketing, sales and other relatively accessible functions rather than core production processes. The JRC also finds a positive relationship between organisational change and the integration of AI into production: sectors in which companies have reorganised more extensively around AI are also more likely to use it in core operations. Joint Research Centre

A separate European Commission survey released the same day reinforces the distinction. Among approximately 12,800 heads of European SMEs and mid-cap companies surveyed, 42% reported using AI or algorithmic-management systems, but mostly on a limited or experimental basis. Another 21% expected to adopt them. High investment costs, legal uncertainty and lack of expertise were the three most frequently cited barriers. Employment, Social Affairs and Inclusion

These datasets measure different things and should not be compared directly. Together, however, they point towards an important strategic question.

If AI tools are becoming increasingly available to competitors at comparable cost, where will sustainable economic advantage actually come from?

The evidence above is sourced. The interpretation that follows is SMAM analysis.

Adoption is accelerating faster than transformation

Corporate discussions around AI often begin with an adoption metric.

How many employees use AI? How many licences have been deployed? How many use cases have been identified? How many departments are running pilots?

These measures can be useful for monitoring activity. They say considerably less about economic transformation.

The JRC data demonstrate why.

AI adoption in cultural and creative sectors reached 38% in 2025, compared with 34% in energy and 30% in electronic communications. Yet the report explicitly distinguishes between businesses using increasingly commoditised, ready-to-use applications and those integrating AI more deeply into production, workflows and infrastructure. Joint Research Centre

That distinction matters because deploying an AI assistant to support written communication is fundamentally different from redesigning an industrial maintenance process, a procurement workflow, an underwriting system or a production-planning function around AI.

The first can often be implemented quickly and generate incremental productivity.

The second changes how the company operates.

It may require access to proprietary datasets, integration with enterprise systems, changes in decision rights, validation procedures, employee training and potentially the removal of existing process steps.

The relevant management metric should therefore move beyond AI usage towards AI penetration into economically material processes.

A company can have thousands of active AI users while leaving its fundamental cost structure, revenue model and capital productivity largely unchanged.

Conversely, a relatively small number of well-designed applications embedded into a critical workflow can potentially have far greater economic significance.

For boards, this makes adoption statistics a starting point rather than an outcome.

Compute is becoming strategic — but access to compute is not the whole advantage

The supply side of AI is moving in the opposite direction: towards unprecedented strategic importance.

An OECD policy brief published on 13 October describes AI compute as a strategic economic asset. Among the 18 national or regional AI strategies adopted or updated during 2025–2026, 13 — approximately 72% — treat compute as a dedicated policy pillar, compared with eight of 26 strategies adopted between 2019 and 2024. OECD

Energy systems are being pulled into the same strategic equation.

The International Energy Agency’s new Powering Artificial Intelligence report, published on 14 October, examines policies and corporate strategies across 34 jurisdictions representing approximately 90% of global data-centre capacity. The IEA identifies 134 policies and regulations addressing the integration of data centres into electricity systems. IEA

The investment implications are substantial. Governments want sovereign or resilient compute capacity. Technology companies require increasingly specialised infrastructure. Electricity grids, generation assets, cooling systems and data-centre development are becoming interconnected components of the AI capital stack.

But what is strategically scarce at national level is not necessarily what differentiates an individual company.

Most mid-market businesses will not compete by owning GPUs, constructing data centres or developing foundation models.

They will compete through what they do with increasingly accessible intelligence.

As model access broadens, a significant part of AI’s competitive advantage is likely to migrate downstream: towards proprietary data, domain expertise, workflow integration and the ability to redesign the organisation around the technology.

AI access is becoming increasingly commoditised. AI absorption capacity is not.

That is an important distinction for capital allocation.

Proprietary context may matter more than the model

The economics of enterprise AI should increasingly be understood through context.

General-purpose models possess broad capabilities. What they do not automatically possess is the specific institutional knowledge required to operate effectively within an individual company.

Customer histories.

Technical documentation.

Pricing logic.

Supplier performance.

Production records.

Contractual precedents.

Engineering specifications.

Risk frameworks.

Internal policies.

Management decisions accumulated over years.

This information is often fragmented across ERP systems, CRM platforms, documents, emails, spreadsheets and the experience of individual employees.

The company that can structure, govern and connect that knowledge to AI systems has the potential to produce outputs that a competitor using the same underlying model cannot replicate as easily.

This changes how boards should think about data.

Historically, data architecture was frequently treated as an IT issue. In an AI-enabled operating model, it increasingly becomes part of the company’s productive infrastructure.

Poor data quality limits automation.

Fragmented systems increase integration costs.

Undefined ownership makes governance difficult.

Unstandardised processes make AI deployment harder to scale across business units.

This creates an important sequencing question.

A company may be tempted to acquire increasingly sophisticated AI capabilities before resolving weaknesses in the operating architecture into which those capabilities must be deployed.

That can produce technologically impressive pilots without corresponding enterprise value.

The more mature approach begins with the economic process being improved, then determines what data, technology and organisational changes are required to improve it.

The model should follow the use case.

Not the reverse.

AI is becoming an operating-model decision

Once AI moves into core processes, implementation ceases to be predominantly a technology project.

It becomes an operating-model decision.

Consider procurement.

An AI system might analyse supplier performance, identify pricing anomalies, forecast demand or recommend alternative sourcing strategies.

The technology is only one component.

Management must still determine who can act on the recommendation, which decisions require human approval, how exceptions are handled, how supplier risk is assessed and who remains accountable when the system produces an incorrect conclusion.

The same principle applies to credit, recruitment, pricing, production planning or capital expenditure.

This explains why organisational transformation is so important in the JRC findings.

The issue is not simply whether a company possesses an AI capability. It is whether processes and responsibilities have been redesigned sufficiently for that capability to influence economic outcomes. Joint Research Centre

The European Commission’s SME and mid-cap survey also illustrates how cautiously companies are approaching decision authority. Although 61% of surveyed companies using these systems said AI helps them make decisions, 90% reported that people still make the final decisions; only 2% described decision-making as fully automated. Employment, Social Affairs and Inclusion

Human oversight should not automatically be interpreted as technological immaturity.

In many functions, retaining human accountability is economically and legally rational.

The management question is more precise: where does human judgement add sufficient value to justify remaining inside the process?

That question can reveal a more sophisticated operating model than either complete automation or complete resistance to it.

AI expenditure needs to become capital allocation

As corporate AI budgets increase, boards will need stronger investment discipline.

The relevant question cannot remain: Do we have an AI strategy?

It should become: Which economic variables are we trying to change?

A credible AI investment case might seek to reduce equipment downtime, shorten engineering cycles, improve conversion rates, reduce fraud losses, increase sales capacity, lower procurement leakage or improve working-capital forecasting.

Each objective has a baseline.

Each can potentially be measured.

And each should be assessed against the full cost of achieving it.

That cost extends beyond software subscriptions.

It can include systems integration, data preparation, cybersecurity, governance, employee training, process redesign and ongoing model supervision.

Companies should therefore distinguish between experimentation expenditure and scaled transformation capital.

Experimentation is legitimate. Businesses need to test technologies before the economics are fully understood.

But once a pilot is proposed for material deployment, a higher standard should apply.

What measurable outcome has the pilot produced?

Can the result be replicated?

What does implementation cost across the organisation?

What operational dependencies emerge?

What happens to the legacy process?

Who owns the benefit in the P&L?

And what is the expected payback?

If management cannot explain which economic variable an AI investment is expected to improve, the project may still be useful experimentation.

It is not yet a capital-allocation case.

This discipline becomes especially important because AI can make experimentation exceptionally easy. The ability to create a prototype quickly can create the illusion that enterprise deployment will be equally straightforward.

Often, the expensive part begins after the prototype works.

Organisational absorption capacity may become the real moat

The strategic implication is not that compute, models or technical capabilities are unimportant.

They remain essential components of the AI stack.

The more subtle point is that their relative contribution to differentiation changes as they become more widely available.

A comparable dynamic has occurred with previous generations of enterprise technology.

Access to cloud computing became widespread. The competitive advantage did not reside simply in possessing cloud access. It depended on what businesses rebuilt around it.

AI may develop similarly, although potentially at greater speed.

Companies with higher absorption capacity should be better positioned to convert successive improvements in model capabilities into economic results.

That capacity consists of several reinforcing elements: usable proprietary data, sufficiently standardised processes, clear decision rights, management willingness to redesign workflows, employees capable of working with the technology and governance robust enough to deploy it responsibly.

This also changes the meaning of AI readiness.

It is not solely a measure of technical infrastructure.

It is a measure of organisational adaptability.

For investors, this may eventually become relevant to due diligence. Two businesses can purchase access to the same models while possessing very different abilities to convert them into margins, growth or capital efficiency.

For CEOs, it implies that responsibility for AI cannot remain isolated within the technology function.

For boards, it means oversight should connect AI investments with strategy, operating performance and capital allocation.

The companies that build these capabilities early may not simply implement today’s tools more effectively.

They may become structurally faster at absorbing whatever the next generation of AI makes possible.

SMAM Perspective

The AI investment cycle is currently highly visible at the infrastructure level.

Governments are securing compute. Capital is flowing into data centres. Electricity systems are being adapted to accommodate additional demand. Model capabilities continue to advance.

But enterprise value will ultimately be created further downstream.

The latest European evidence suggests that adoption is already accelerating while deep integration remains considerably less uniform.

That gap deserves attention.

For boards and management teams, the next stage should be less concerned with demonstrating that the company is “using AI” and more concerned with identifying where AI can materially alter the economics of the business.

That requires a progression from tools to processes, from pilots to operating models, and from technology budgets to capital-allocation discipline.

The strategic question is therefore evolving.

It is no longer simply:

Which AI should we adopt?

It is:

Which parts of our operating model should AI materially change — and what economic return should that change produce?

As technology becomes more accessible, the quality of that answer may become one of the more durable differences between companies that adopt AI and companies that convert it into enterprise value.