Summary
- Opengear surveyed 514 CIOs and CISOs across five countries, including 100 respondents in the UK.
- Among UK respondents, 73% reported delaying, scaling back or pausing AI and edge projects over network resilience concerns.
- Organisations with more developed resilience capabilities reported estimated outage costs £3.2m lower than less-prepared UK counterparts.
Almost three-quarters of UK organisations surveyed by Opengear have delayed, scaled back or paused an AI or edge computing project over concerns about network resilience, according to research published in November.
The company’s Resilience Advantage study found that 73% of UK respondents had changed an infrastructure project during the previous 12 months because of concerns about their ability to maintain access to critical systems and recover from network disruption.
Opengear surveyed 514 chief information officers and chief information security officers in July and August 2026, covering the UK, United States, Germany, India and Australia. The UK sample comprised 100 respondents, making the findings a measure of the experiences reported by surveyed technology leaders rather than a census of British organisations.
Across the full international sample, 72% reported delaying, reducing or pausing an AI, edge or distributed infrastructure project. The research also found that only 23% of respondents planned for network resilience before deployment.
The findings come as organisations increasingly distribute computing workloads between central data centres, cloud environments and smaller edge installations. Maintaining those systems requires not only working production networks but also arrangements for diagnosing and recovering equipment when normal connectivity fails.
Network recovery needs an independent access route
Network equipment is commonly managed through connections that depend on the wider production environment. When routing equipment, a firewall or an upstream link fails, engineers may lose access to the same systems needed to investigate and resolve the fault.
An interruption can therefore affect both customer services and the tools used to restore them. The consequences depend on the architecture, including whether an alternative management connection remains available.
Out-of-band management provides a separate route for reaching network devices and other infrastructure when the primary network is unavailable. Such systems may use independent telecommunications connections or physically separated management equipment to provide access for authorised engineers.
The distinction between production traffic and management traffic is important. Restoring access to a failed switch can allow an engineer to inspect its configuration or attempt recovery, but it does not itself restore customer applications that depend on other affected infrastructure.
Recovery procedures must account for the actual fault, the configuration of redundant equipment and the state of connected services. A second communications path offers limited protection if it shares the same vulnerable equipment or physical route as the primary connection.
Remote management also introduces security obligations. Access must be authenticated, appropriately restricted and monitored because a management system capable of changing critical infrastructure can become a valuable target for attackers.
Opengear’s survey assessed organisations according to their ability to reach and recover infrastructure when the primary network was unavailable, their recovery testing practices and their reliance on sending engineers to affected locations.
UK respondents classified as having embedded resilience reported estimated average outage costs £3.2m lower than those with less developed capabilities over the preceding year.
The comparison is based on estimated costs derived from survey responses. It does not demonstrate that deploying a particular recovery product would produce a £3.2m saving, nor does it establish that differences in resilience practices alone caused the reported financial gap.
Distributed computing increases operating demands
AI infrastructure creates several distinct network requirements. Accelerators within a computing cluster may need to exchange large quantities of information, while storage systems must provide data and retain results generated by the workload.
Communication failures between components can reduce computing utilisation or interrupt distributed processing. Some systems can recover from an earlier checkpoint, but this may require repeating calculations and retrieving stored information.
Network capacity, latency and packet loss therefore influence the performance and reliability of computing installations. The significance of each depends on the workload and the architecture used to distribute it.
Edge computing adds geographical complexity. Smaller installations may operate in locations without permanent engineering staff, making physical attendance more difficult when equipment becomes inaccessible.
Remote recovery can shorten the time needed to investigate certain faults, provided the management infrastructure remains operational. Where recovery requires replacement equipment, repair of damaged connections or other physical intervention, engineers may still need to attend the site.
Organisations deploying AI infrastructure must also consider resilience across storage, computing and supporting systems. Independent access to network equipment cannot compensate for insufficient backup power, unavailable cooling or applications lacking suitable recovery arrangements.
These dependencies make operational testing important before production workloads are introduced. Recovery exercises can identify incorrect assumptions about redundant connections, equipment configuration and the ability of staff to restore systems under realistic conditions.
The survey’s 73% UK finding covers projects that were delayed or reduced as well as those paused. It does not establish that all affected developments were abandoned, or quantify the computing capacity or investment value associated with the reported decisions.
Opengear supplies network resilience equipment, so its commercial interest and the use of self-reported survey data should be considered when interpreting the findings. The research nevertheless identifies a practical issue: computing infrastructure may be technically capable of supporting workloads before the operational arrangements needed to recover it have been adequately established.
Network architecture, independent management access and tested recovery procedures form part of that readiness assessment. Their effectiveness ultimately depends on the behaviour of the installed equipment and the organisation’s ability to restore services when a fault occurs.

