55% of enterprises favor on-premises AI infrastructure
Schneider Electric and Omdia found that 55% of enterprises prefer on-premises AI as 2026 budgets rise, but most report no more than one in five proofs of concept reaches production.

Why it matters
- 55% of respondents prefer company-owned, on-premises AI infrastructure.
- More than half put 2026 AI budgets above USD $500,000, and 31% put spending above USD $1 million.
- Among companies with implemented AI projects, 60% say ROI meets or exceeds expectations.
- More than half report no more than one in five proofs of concept reaches production.
- 78% expect AI budgets to rise in the following year; nearly 30% forecast 11% to 25% year-on-year growth.
Fifty-five percent of enterprises prefer company-owned, on-premises infrastructure for AI and plan to own at least part of their AI infrastructure, according to global research that Schneider Electric commissioned and Omdia conducted.
Schneider Electric and Omdia identified data privacy and security, reliability requirements and regulatory compliance as the main reasons for that preference.
Schneider Electric published the findings on November 20, 2026. Omdia gathered responses from enterprise leaders across Europe, North America and Asia-Pacific on AI strategy, adoption, spending, infrastructure and returns.
The respondent pool included executives from IT, product engineering, research and development, and general management. It also covered healthcare, financial services, life sciences, government, education and manufacturing.
Schneider Electric and Omdia say the findings show AI use moving beyond experimentation, while compliance, privacy and operational demands continue to shape deployment choices.
What are enterprises adopting?
Schneider Electric and Omdia connect enterprise AI spending to three operating priorities: higher productivity, lower costs and time savings. The survey also asks how widely companies have piloted AI, where they plan to run it and whether returns meet expectations.
- Increased productivity was the main reason for adopting more AI: 35%.
- Cost reduction was another leading reason: 25%.
- Time savings drew 22%.
Nearly three-quarters of respondents said they were running five or more proofs of concept. Another 63% plan to deploy generative AI applications, and 23% said AI models were already adding value within their organizations.
The contrast matters because broad experimentation has not established production value across every organization. Schneider Electric and Omdia treat pilot volume and realized returns as separate measures of enterprise AI maturity.
At the same time, the 23% reporting current value shows that the deployment phase has begun for some organizations. Their survey measures adoption, infrastructure and returns separately rather than treating one proof of concept as evidence of organization-wide deployment.
Are returns meeting expectations?
Among companies with implemented AI projects, 60% said return on investment met or exceeded expectations. Within that group, 29% reported returns of two to five times and 17% reported returns of five to 10 times.
That ROI measure applies only to companies with implemented projects. It should not be read as an organization-wide result for every firm still running proofs of concept.
More than half of respondents put their AI budgets for 2026 above USD $500,000, while 31% put spending above USD $1 million.
A majority also expects larger AI budgets in the following year. Seventy-eight percent said budgets would rise, and nearly 30% forecast year-on-year growth of 11% to 25%.
Schneider Electric and Omdia therefore pair rising spending with constrained production. Enterprises are committing substantial money while internal skills, integration work and infrastructure requirements still block many projects.
Where will AI workloads run?
The research indicates that enterprises are not relying solely on public cloud or outsourced environments as budgets grow. The findings show a mix of company-owned systems and outside support rather than a single route to production.
Schneider Electric and Omdia connected the on-premises preference to three requirements: control over data handling, dependable system performance and regulatory compliance. The study also names security as a principal concern.
The survey covers regulated sectors such as healthcare, financial services, life sciences, government, education and manufacturing. That range places the infrastructure question beyond model hosting. It also reaches the systems needed to run and scale AI models.
The result places owned compute, power and cooling directly inside enterprise AI planning. Schneider Electric and Omdia also show that company infrastructure decisions cannot be separated from software strategy, because reliability and compliance determine where workloads can run.
Why do so many pilots fail?
More than half of respondents said no more than one in five proofs of concept reach production because of human and infrastructure barriers. Omdia described this as a recurring delivery problem across organizations trying to move projects into production.
- Lack of in-house expertise was an obstacle to full-scale deployment for 33%.
- Integration problems with existing systems and architectural complexity blocked another 32%.
When deploying new systems for AI, respondents most often named scaling AI IT clusters, costs, and sourcing power and cooling infrastructure as their top challenges.
Those findings explain why pilot counts can rise without a matching increase in production deployment. Human capacity, integration work and physical infrastructure all constrain the path from testing to operation.
The survey also identifies where outside support enters: 59% of respondents rely on external partners to help shape AI strategy, while 67% depend on outside support for implementation.
Who is filling the infrastructure gap?
The partner mix spans cloud providers, infrastructure vendors, software platforms, systems integrators and specialist AI firms.
Schneider Electric and Omdia say this broad supplier network helps companies bridge skills shortages and reduce deployment complexity.
Paul Tyrer, Global Vice President, AI Ecosystem at Schneider Electric, said the results show a shift in enterprise priorities as AI projects mature.
“As AI moves from experimentation to real-world application, enterprises are focused on making these technologies deliver measurable business value at the right speed, accuracy and cost,” Tyrer said.
Tyrer also said the research highlights both opportunities and challenges as enterprises formulate AI strategies and supportive technologies. He called clear identification of where organizations make progress and where they need support critical to moving forward.
Maxine Holt, Vice President, Enterprise & Channel Research at Omdia, said the lack of internal expertise remains a significant barrier and a leading reason proofs of concept fail.
“By leveraging existing partners and resources such as reference designs and validated architectures, organisations can navigate the complexity of AI infrastructure while aligning investment with deployment needs and accelerating time to adoption and business value,” Holt said.
What changes next?
Omdia said the delivery gap is feeding demand for reference designs, validated architectures and implementation support. Those offerings target the specific failures the survey measures, from skills gaps to integration work and power and cooling constraints.
The commercial stakes are visible in two numbers: 78% of respondents expect budgets to rise, while 67% depend on outside implementation support. Companies can spend more without resolving the constraints that keep most proofs out of production.
The next test for suppliers and internal technology teams is whether investment can lift production beyond one in five proofs of concept. The survey gives rising budgets and heavy partner reliance, but skills gaps, architectural complexity and power and cooling constraints remain the practical constraints on scale.
Source: itbrief.com.au
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
802 articles
Related articles
- 51% of AI projects miss targets, yet 85% of firms raise budgets
- Productivity leads enterprise AI, but autonomy remains scarce
- Private AI Needs Infrastructure Built to Evolve, Not Just GPUs
- Only 10% of Enterprises Let AI Agents Decide Without Human Approval
- Enterprises Kill 46% of AI Pilots Before Production, S&P Global Finds