EMA Report Finds Enterprise Networks Unprepared for Edge AI

EMA Report Reveals Edge AI Network Readiness Gaps EMA Report Reveals Edge AI Network Readiness Gaps

A new Enterprise Management Associates (EMA) study finds that many organizations are not fully prepared to support artificial intelligence inference at the network edge, despite growing interest in real-time AI applications. Based on a survey of 319 IT professionals, the report, Enterprise Strategies for AI Inference at the Network Edge, finds that only 40% are completely confident their network infrastructure can handle these workloads, while just 43% expect their edge inference strategies to succeed completely. The findings highlight a growing infrastructure challenge: bringing AI closer to where data is generated requires more than deploying models. Enterprises must also modernize network connectivity, identity management, security and IT operations.

Edge Inference Is Becoming a Network Engineering Challenge

Edge inference involves running trained AI models close to the locations where data is generated, rather than sending every request to a centralized cloud environment. Processing can take place in factories, warehouses, retail branches, remote industrial facilities or regional data centers.

The architecture can reduce the time required to move data to distant computing resources and return a result. It can also help organizations keep sensitive information within specific locations or jurisdictions. These characteristics make edge inference relevant to applications such as industrial computer vision, predictive maintenance, real-time retail analytics, robotics and autonomous systems.

But distributing inference also distributes infrastructure responsibilities. Models need access to compute resources, data pipelines and applications across locations that may have different connectivity, power, security and staffing conditions.

EMA Vice President of Research Shamus McGillicuddy argues that these demands will require upgrades across the network’s data, control and management planes. The network must do more than connect remote sites: it must support predictable performance, secure access and consistent operations across a distributed environment.

Hybrid Deployment Is the Enterprise Preference

One of the report’s most notable findings concerns where enterprises choose to run inference workloads.

Only 18% of respondents deploy all their edge inference workloads at the remote locations where data originates. By contrast, 64% use a hybrid approach, distributing workloads between edge sites and cloud environments, including public cloud platforms and data centers. Security policies, cost and data sovereignty are among the leading factors influencing placement.

This suggests that enterprises are not treating edge inference as a wholesale replacement for centralized cloud computing.

Instead, they are selecting locations based on workload requirements. A manufacturing application that needs an immediate response to a machine anomaly may benefit from local processing. Other workloads may be better suited to a regional data center or public cloud, where organizations can access more computing capacity and centralized management.

Hybrid deployment also introduces coordination challenges. IT teams must manage model distribution, data movement, network routing, security policies and operational visibility across different environments.

Network Readiness Remains a Major Weakness

EMA found that only 40% of surveyed IT professionals are completely confident their networks are ready for edge inference. Just 43% expect their overall edge inference strategies to achieve complete success.

The research also identifies a gap between executive expectations and technical implementation teams. IT executives were generally more optimistic about the prospects of edge inference than the professionals responsible for deploying and operating the infrastructure.

That disconnect can become a problem when strategic AI plans move into production. Executives may approve new use cases based on expected business benefits, while network teams face practical constraints involving bandwidth, congestion, addressing, identity and site reliability.

The infrastructure modernization figures are striking. According to the report, 97% of respondents are updating their software-defined wide-area networking (SD-WAN) environments, another 97% are updating DNS, DHCP and IP address management (DDI) platforms, and 96% are updating certificate and identity management technology.

These technologies form important parts of the infrastructure supporting distributed AI. SD-WAN helps manage connectivity between locations, while DDI provides essential network addressing and name-resolution services. Certificate and identity management helps establish trusted connections between users, devices, workloads and systems.

The findings indicate that edge inference is pushing network modernization beyond bandwidth upgrades alone.

Security and Data Sovereignty Influence Workload Placement

As inference moves closer to remote operations, organizations must account for the security of distributed infrastructure.

A remote facility may have fewer local IT resources than a central data center. That can complicate software updates, certificate management, identity verification and incident response. Inconsistent configurations across sites can also make it harder to enforce enterprise-wide security policies.

Data sovereignty adds another consideration. Organizations operating across jurisdictions may need to control where information is processed, stored and transferred. The location of an inference workload can therefore be as important as its latency or computing requirements.

The report identifies security policy, cost and data sovereignty as leading factors in workload placement. These considerations help explain why hybrid architectures are more common than running every workload locally.

Network teams must coordinate with security, cloud and AI engineering groups to establish which workloads can run at each location and what controls must apply.

Business Value Depends on Operational Outcomes

The research identifies three leading business outcomes for edge inference initiatives: improved operational efficiency, stronger safety and risk management, and better decision-making through real-time insights.

These priorities connect edge AI with operational technology rather than limiting it to office productivity applications. In manufacturing, local inference could help identify equipment anomalies. In retail, it could support faster analysis of customer-facing activity. In remote industrial environments, it could help systems respond to safety conditions without relying entirely on distant cloud connectivity.

However, the report’s findings suggest that technical readiness remains a prerequisite for achieving those benefits. Organizations need networks capable of supporting latency-sensitive applications, secure data access and consistent management across distributed locations.

The research was sponsored by BlueCat and Scale Computing. Its survey results offer a useful snapshot of enterprise priorities, but the sponsorship should be considered when interpreting the report’s technology recommendations.

For enterprise IT leaders, the central message is clear: edge inference is not simply an AI deployment decision. It is a network, security and operations modernization project. Organizations that address those foundations early will be better positioned to move beyond isolated pilots and scale real-time AI applications.

Market Landscape

The edge AI market is moving toward distributed inference architectures that combine local computing, regional data centers and centralized cloud infrastructure.

Several developments are shaping this transition:

  • Hybrid AI infrastructure: Enterprises are distributing inference workloads according to latency, cost, security and data sovereignty requirements.
  • Network modernization: SD-WAN, DDI, identity management and network observability are becoming important components of AI infrastructure.
  • Industrial AI: Manufacturing, logistics, energy and resource extraction are adopting real-time inference for operational intelligence and safety.
  • Physical AI: Robotics, autonomous systems and computer vision require reliable connectivity and low-latency processing.
  • Edge security: Distributed deployments increase the importance of identity controls, certificate management and consistent security policies.

The wider infrastructure market is responding to these requirements. IDC has projected that 45% of enterprises will deploy inference at the edge by 2028, reflecting the growing importance of distributed AI infrastructure. Simplifying secure, AI-ready infrastructure

EMA’s findings suggest that infrastructure readiness, rather than interest in AI applications, could become a defining factor in which enterprises successfully scale edge inference.

Top Insights

  • EMA’s survey of 319 IT professionals found that only 43% expect complete success from their enterprise edge inference strategies.
  • Just 18% deploy all inference workloads at remote sites, while 64% use hybrid edge, cloud and data center architectures.
  • Only 40% of respondents are completely confident their networks are ready to support demanding edge inference workloads.
  • Network modernization is widespread, with 97% updating SD-WAN, 97% DDI and 96% certificate and identity management technologies.
  • Operational efficiency, safety and risk management, and real-time decision-making are the leading business objectives for edge inference.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI

Grow Your
Brand Visibility

Looking to publish a press release, guest article, interview or podcast? Connect with us.

GET FEATURED
Subscribe

Sign up today for exclusive insights and updates.

Newsletter Signup