Enterprise AI has entered its second, harder phase. According to Deloitte’s 2026 State of AI in the Enterprise report, worker access to AI tools rose by 50% in 2025, and 66% of organizations now report measurable productivity and efficiency gains. Yet the same report reveals a sobering counterpoint: only 25% of companies have moved 40% or more of their AI pilots into production, and Deloitte’s European-facing data shows just 34% of firms describe their AI use as a deep business transformation rather than a peripheral experiment. For CFOs, General Counsel, and M&A Directors evaluating technology risk and enterprise value, the gap between AI enthusiasm and AI execution is now the defining strategic question of digital transformation.

From Pilot to Production: The Scaling Bottleneck

The headline finding from Deloitte is encouraging on its face: companies with at least 40% of AI projects already in production are expected to double within six months. This suggests an inflection point in enterprise AI adoption is near. But the underlying data tells a more nuanced story. The primary obstacle, cited consistently across the report, is the AI skills gap — not compute capacity, not model quality, but the organizational capability to operationalize AI responsibly and at scale.

This aligns with Dataiku’s June 2026 analysis, which reframes AI adoption not as a tooling problem but as an enterprise redesign challenge. Successful scaling, Dataiku argues, depends on three interlocking factors: people, orchestration, and governance. Readiness gaps in data infrastructure, governance frameworks, and workforce skills remain the most common blockers preventing pilots from reaching production. For boards overseeing digital strategy, this means the constraint on ROI is rarely the algorithm — it is the operating model surrounding it.

Physical AI adds a further layer of complexity. More than half of surveyed companies already use physical AI — robotics, autonomous systems, embedded sensors — at least in limited form, with broader adoption expected over the next two years. This expands the AI governance perimeter beyond software into operational technology, supply chains, and safety-critical environments, raising the stakes for compliance and risk functions.

Governance, Sovereignty, and the European Compliance Layer

European enterprises face a distinct set of pressures that global peers do not. Data residency requirements, the EU AI Act’s risk-tiered obligations, and sector-specific regulation in financial services and healthcare mean that AI scaling decisions cannot be separated from compliance architecture. Market commentary increasingly centers on sovereign AI economics — the ability to control where models run, where data resides, and who can audit AI decision-making.

This is driving demand for agent gateways: infrastructure layers that secure and manage AI traffic between enterprise systems and third-party or internal models. For firms in regulated industries, agent gateways are becoming a practical mechanism for reconciling innovation management with the accountability, transparency, and human-oversight obligations embedded in EU AI Act compliance. Enterprise software vendors are responding with private and hybrid AI deployment models, emphasizing tighter integration between AI systems and enterprise data rather than reliance on public cloud-only architectures — a trend directly relevant to cloud migration strategy for mid-market and regulated firms across Europe.

Implications for Business Leaders

For CFOs and M&A Directors, three implications follow directly from this data:

  • Due diligence must now include AI operating maturity. Target companies claiming “AI-enabled” capabilities should be assessed against Deloitte’s production-vs-pilot benchmark, not marketing language.
  • Skills investment is a capital allocation decision, not an HR line item. Given that the skills gap is the dominant barrier to scaling, workforce upskilling budgets should be evaluated with the same rigor as infrastructure capex.
  • Governance infrastructure — including agent gateways and data residency controls — should be treated as a prerequisite for AI scaling, not an afterthought. This is particularly critical for boards managing EU AI Act exposure and cross-border data flows.

CTOs, meanwhile, should prioritize orchestration layers and governance tooling over incremental model upgrades, since Dataiku’s findings suggest these are the true rate-limiters on enterprise value creation from AI.

Key Takeaway

The 2026 data confirms that enterprise AI adoption is no longer an experimentation question — it is an execution and governance question. Organizations that pair digital transformation ambition with disciplined operating-model redesign, workforce readiness, and compliance-grade governance will be the ones that convert Deloitte’s productivity gains into durable competitive advantage. Those that don’t risk being permanently stuck in pilot purgatory.