Enterprise AI Access Is Scaling Faster Than Value

The seventh edition of Deloitte’s State of AI in the Enterprise: The Untapped Edge — unveiled on 21 January 2026 at Davos — draws on a survey of 3,235 director-to-C-suite leaders across 24 countries and six industries, with fieldwork conducted between August and September 2025. Its central finding is blunt: access is scaling faster than value. Worker access to sanctioned AI tools rose 50% in 2025, from under 40% to roughly 60% of the workforce, yet only 25% of organisations have moved 40% or more of their AI pilots into production. The same report notes that 54% of organisations expect to reach that production threshold within three to six months.

Access is scaling faster than value.

That gap between experimentation and operational impact is the defining tension of enterprise AI entering 2026. For executives and founders, the Deloitte findings and a separate market analysis from Menlo Ventures converge on a clear message: the bottleneck is no longer tooling or ambition. It is operating-model redesign, governance and data readiness — and those constraints are now reshaping procurement, vendor choices and build-versus-buy decisions.

The pilot-to-production gap is still widening

Deloitte’s headline numbers show a familiar pattern: adoption is broad, but transformation remains narrow. While 66% of organisations report productivity and efficiency gains from AI, other realised benefits are less universal. Better insights and decision-making were cited by 53% of respondents, cost reduction by 40%, and stronger client relationships by 38%. Yet only 20% of organisations say they are already growing revenue from AI, even though 74% hope to do so.

The deeper issue is how AI is being used. Just 34% of organisations are using AI to “deeply transform” the business, while 30% are redesigning key processes and 37% remain at surface level. Deloitte notes that transformative impact did double year-on-year, reaching 25% of leaders, but the absolute level remains modest. The report also finds that 54% of organisations expect to reach the threshold of moving 40% or more of AI pilots into production within three to six months — an expectation that, if unmet, could further separate leaders from laggards.

Agentic, physical and sovereign AI are the new fronts

Beyond the familiar productivity narrative, Deloitte identifies three emerging fronts that will define the next phase of enterprise AI.

• Agentic AI: Roughly 75% of companies plan to deploy AI agents within two years, and 85% expect to customise those agents. But only 21% have mature agent governance in place — a gap that raises significant risk as autonomous systems move closer to core operations.
• Physical AI: Already 58% of organisations use physical AI, and adoption is projected to reach 80% within two years. Manufacturing, logistics and defence lead the way, with Asia-Pacific ahead of other regions.
• Sovereign AI: 83% of leaders call sovereign AI strategically important. Country of origin now factors into vendor selection for 77% of respondents, and nearly three in five organisations are building AI stacks primarily with local vendors.

Talent remains the top integration barrier. The AI skills gap is cited as the No. 1 obstacle to AI integration, and education — not role redesign — is the most common response, chosen by 53% of organisations. Strikingly, 84% of companies have yet to redesign jobs around AI, suggesting that many are layering new tools onto old workflows rather than rethinking how work should be done. This pattern helps explain why efficiency gains are widespread while revenue growth remains concentrated among a minority of organisations.

The market is consolidating fast — and spending is shifting

Menlo Ventures’ third annual enterprise AI report, based on a survey of 495 US decision-makers conducted between 7 and 25 November 2025, sizes enterprise generative AI at $37 billion in 2025 — up 3.2 times from $11.5 billion in 2024 and equivalent to 6% of the global SaaS market. The report also shows a decisive shift toward buying rather than building: 76% of AI solutions are now purchased, up from 53% a year earlier. AI deals convert at 47%, compared with 25% for traditional SaaS.

Vendor concentration is accelerating. Anthropic leads enterprise LLM spend at 40%, up from 24% a year earlier, while OpenAI’s share fell to 27% from 50% in 2023. Google holds 21%. Together, Anthropic, OpenAI and Google account for 88% of enterprise LLM use. Anthropic also dominates the coding market with a 54% share, according to Menlo.

Menlo explicitly frames the data as “boom vs. bubble,” pushing back against an MIT study that claimed 95% of generative AI initiatives fail. Deloitte is more cautious, flagging a persistent gap between experimentation and transformation, lagging agent governance, and revenue growth that remains largely aspirational. Named voices in the reports include Nitin Mittal, Deloitte Global AI leader; Jim Rowan, Deloitte US head of AI; and Deedy Das, partner at Menlo Ventures.

What this means for leaders

The two reports tell a consistent story. Enterprise AI is no longer constrained by model capability or initial adoption. The real constraints are organisational: how quickly companies can redesign jobs, establish governance for autonomous agents, improve data readiness, and convert pilots into production systems that generate measurable revenue or cost impact.

Procurement is also becoming more strategic. With 76% of AI solutions bought rather than built, and with sovereign-AI concerns influencing vendor selection for more than three-quarters of organisations, the enterprise AI stack is increasingly being assembled from a small set of trusted vendors — but with growing pressure to localise. Anthropic, OpenAI and Google now account for 88% of enterprise LLM use, leaving little room for smaller providers in core model procurement. That tension between global model providers and national or regional sovereignty requirements will shape deals, partnerships and infrastructure decisions throughout 2026.

The coming year will test whether the surge in access can be converted into durable value. Organisations that treat AI as an operating-model change — not just a technology rollout — are the ones most likely to close the gap between pilots and production. For everyone else, the risk is that 2026 becomes another year of impressive adoption metrics and underwhelming business results.