Summary
- Omdia's global survey found 55% of respondents preferred company-owned, on-premises AI infrastructure.
- Scaling AI clusters (27%), costs (25%), and sourcing power and cooling infrastructure (22%) were the leading deployment challenges reported.
- The findings cover Europe, North America and Asia-Pacific, but the published headline percentages are not specific to Europe.
More than half of enterprises surveyed by Omdia for Schneider Electric prefer company-owned infrastructure for AI deployments, although scaling computing clusters and securing the power and cooling systems required to support them remain significant obstacles.
The research, published on 18 November, found that 55% of respondents favoured on-premises, owned infrastructure rather than relying entirely on external computing environments. Data privacy and security, reliability requirements and regulatory compliance were among the principal reasons.
Omdia conducted the survey on Schneider Electric’s behalf, gathering responses from enterprise leaders across Europe, North America and Asia-Pacific. Participants represented several sectors, including financial services, healthcare, manufacturing, government, education and life sciences.
Data privacy and security were cited by 30% of respondents as factors influencing decisions to retain AI infrastructure in-house, while reliability requirements accounted for 22% and regulatory compliance for 21%.
The research also identified difficulties associated with deploying the necessary infrastructure. Some 27% of respondents cited scaling AI computing clusters, 25% identified costs, and 22% reported challenges sourcing power and cooling equipment.
These findings describe the views of survey participants rather than measured investment in completed data centres. The published figures represent a global sample and should not be interpreted as equivalent percentages for the UK or Europe.
They nevertheless identify a distinction between organisations’ preferred computing arrangements and their ability to implement the physical systems those arrangements require.
Company ownership can provide greater control over equipment specifications, deployment locations and operational arrangements. It also transfers responsibility for securing suitable facilities, electricity supply, thermal management and the technical personnel needed to maintain the installation.
Private AI infrastructure requires suitable facilities
Operating AI systems within company-owned infrastructure does not necessarily require constructing a new data centre. Organisations may install computing equipment within existing enterprise facilities, expand established infrastructure or locate their own hardware in a colocation environment.
Each approach creates different requirements for electricity distribution, cooling, physical security, connectivity and maintenance.
Existing facilities can face substantial engineering limitations when additional computing equipment requires more power per rack than the original installation was designed to accommodate.
The electricity available through a building’s connection must pass through transformers, switchgear, uninterruptible power systems, distribution equipment and associated protection before reaching computing hardware.
Adding processors without assessing those systems can create constraints even where the building’s overall electricity supply appears sufficient.
Higher electrical loads also increase the heat that must be removed. The appropriate cooling arrangement depends on the computing equipment, its operating requirements and the infrastructure already installed.
Some high-density systems require direct liquid cooling, introducing coolant distribution equipment, pipework, heat exchangers and controls that may not exist in conventional enterprise server rooms.
Retrofitting those systems can involve substantial work within an operating facility. The installation may need new mechanical plant, changes to electrical distribution and modifications to existing maintenance arrangements.
Electrical and cooling upgrades must also be coordinated. Installing additional cooling capacity does not establish that the electricity system can support it, while an electrical upgrade alone cannot ensure that computing equipment will remain within acceptable thermal conditions.
The survey’s finding that 22% of respondents encountered difficulties sourcing power and cooling infrastructure therefore identifies a physical component of enterprise AI deployment that cannot be resolved through software procurement alone.
External expertise remains central to deployment
Although respondents expressed a preference for greater ownership of computing infrastructure, many continued to depend on outside expertise when planning and implementing AI systems.
Omdia found that 59% of surveyed companies relied on external partners to help develop their AI strategy, while 67% used external assistance during implementation.
Those figures do not establish that every participating organisation requires specialist data centre engineering services. External support can also involve applications, systems integration, procurement and other technical work.
Infrastructure deployment nevertheless introduces engineering disciplines extending beyond the installation of processors and servers. Electrical design, cooling, network architecture, monitoring and operational resilience must be coordinated within the chosen facility.
Organisations seeking direct control over sensitive information must also determine where data is processed, how it is transferred and which parties can access the systems involved.
Physical ownership and regulatory sovereignty are related but separate considerations. Equipment located within a particular jurisdiction may still depend on external software, management services or contractual arrangements.
Likewise, operating a private connection does not independently establish that every relevant data protection or sector-specific obligation has been satisfied.
European organisations must consider those arrangements against the regulatory requirements applying to their activities, as well as the available locations and capabilities of their computing infrastructure.
Schneider Electric supplies electrical equipment, cooling systems and other data centre infrastructure, and has a commercial interest in the equipment categories covered by the research.
The findings are consequently best treated as supplier-commissioned evidence of enterprise preferences and reported deployment difficulties rather than an independent forecast of Schneider Electric sales or future data centre construction.
Nor does the survey quantify the electrical capacity, investment or additional floor space required to accommodate respondents’ preferred deployment arrangements.
Turning those preferences into operational infrastructure will depend on equipment procurement, facility suitability, electricity availability and the cost of maintaining reliable systems.
The research establishes substantial interest in company-owned AI computing, but the rate at which that demand becomes operating capacity remains dependent on decisions and investments beyond the choice of hosting model.

