In life there are certain things that pair well together. There is chocolate and peanut butter and who doesn’t like a good ribeye paired with a Napa cabernet? In tech there is mobile and cloud and UC and CC but another interesting pairing is on the horizon and that’s edge and AI. While edge computing has been on the horizon for some time, it was a bit of solution looking for a problem, but AI is driving the need for more data in more places, including the edge.
A new report from VMware (a Broadcom company), released at its Explore Barcelona event, sheds light on the challenges of implementing edge computing and artificial intelligence (AI). Organizations adopting these technologies have reached a critical point where network infrastructure needs to improve to handle the demands of edge computing at a large scale.
The State of the Edge report draws on data collected from VMware’s VeloCloud platform customer deployments. This includes over 5 trillion data points analyzed daily from more than 300,000 edge locations and 3,700 gateways. The report also incorporates responses from 192 people surveyed worldwide in October 2024.
AI and Edge Go Hand in Hand
AI is driving edge computing because it enables real-time data processing and decision-making. Almost every enterprise – from retail to healthcare to sports and entertainment want to better understand the data they are collecting and how to capitalize on it. This can lead to new services or experiences. As an example, in retail, AI enabled cameras combined with sensors can be used for cashierless transactions. Having to go to the cloud and back would be far too slow so that creates the need for edge computing. In public safety, more and more vehicles are being wired and connected and law enforcement and other agencies can analyze information much faster.
The edge creates faster access, a more secure environment as data isn’t being sent over the network and back, which results in a better user experience. However, despite the benefits, edge and AI are having a significant impact on the network. In a recent survey, I asked 600 enterprises across the US and Europe as to what is biggest challenge in networking today and the top response was “supporting AI workloads,” outpacing even security.
Shifting Traffic Patterns with AI Applications
AI applications change how data moves through networks. With “legacy” apps, most data flows from the server to the user (downstream). With AI, it’s often the opposite: large data, like video streams, is sent from devices to the edge and then often to the cloud. This shift requires networks that can handle more upstream data while still being reliable and secure. To manage this demand and prevent bottlenecks, many companies are switching from central data centers to more distributed network setups.
The study found that about 20 percent of network traffic is either direct internet access or branch-to-branch. Meanwhile, 53 percent of traffic goes to local gateways, which handle cloud apps and enforce security policies. The remaining 33 percent still goes through data centers—6 percent is direct, while 27 percent is routed through a hub.
New Capabilities, New Challenges
While AI and edge offer many benefits, it also creates significant demands on network infrastructure, particularly around connectivity and bandwidth. Edge applications produce large volumes of data, requiring fast and secure wide area network (WAN) connections to transfer data for machine learning (ML), analytics, and other functions. According to the report’s findings, network connectivity is the primary challenge for 57 percent of organizations, which is consistent with my survey.
As a result, there’s a shift happening from traditional MPLS networks to broadband, which offers higher speeds at a lower cost. Most organizations rely on multiple broadband links, with 92 percent using broadband to address high bandwidth demands and 75 percent for redundancy. Only 8 percent are still using MPLS.
Despite being accessible and economical, broadband frequently experiences disruptions and slowdowns. On average, monthly outages and degradation affect wired connections for 13 hours, cellular for 408 hours, and satellite for 71 hours. Technologies like VeloCloud’s dynamic multipath optimization (DMPO) address these issues by minimizing packet loss and latency while enhancing wired connections. VeloCloud, for example, now supports 4G, 5G, and satellite connections, which are increasingly common in remote edge deployments.
Fixed wireless access (4G/5G) and satellite connectivity are growing rapidly to support remote edge locations. Cellular adoption has increased by 42 percent and satellite has skyrocketed by 420 percent over the past two years.
Optimizing Performance and Traffic Management
A network infrastructure that provides visibility into application activity is necessary to keep edge workloads running smoothly. The setup should include strong Wi-Fi coverage and access to essential network services like domain name system (DNS) and dynamic host configuration protocol (DHCP). Data collected from approximately 2,000 customer deployments over the past year shows high wireless and mobile use at branch offices, with Android and iOS devices making up 55 percent of all connected devices.
Wi-Fi quality is a major concern in certain sectors like retail, where 92 percent of devices rely on wireless connectivity and 44.5 percent of companies report coverage gaps. In manufacturing, where wired connectivity is more common, 9.2 percent of companies have issues with DHCP.
That’s where effective traffic management comes in. It’s important to identify high-bandwidth applications and set policies that prioritize essential apps over less critical ones. On average, 30 percent of bandwidth is used by non-essential apps when resources aren’t limited.
Securing the Edge
As AI driven edge deployments continue to grow, data security and compliance become even more critical. The report found that 53 percent of organizations consider security and compliance a top priority in their edge setups. Since edge sites often include a mix of industrial machines and Internet of Things (IoT) devices accessible to both internal and external users, the risk of security breaches is higher. Therefore, servers, operating systems, applications, and IoT devices must all be protected.
Edge applications remain vulnerable to both browser-based and non-browser-based attacks. Common points of vulnerability include AI apps that use application programming interfaces (APIs), edge apps connecting to web resources without a full browser, and IoT devices linked to cloud apps. To protect against these risks, a multi-layered security approach is necessary, especially as edge sites interact more with internet-based apps.
Looking Ahead
Edge technology, now advancing toward the intelligent edge, has the potential to reshape many industries. My recommendation is that organizations proactively reassess and upgrade their network infrastructure to handle the increasing demands of AI at the edge. At the same time, integrating AI in network management, especially through a software-defined wide area network (SD-WAN), will allow organizations to adapt to new threats as they arise.
Typically network upgrades are done reactively when users start complaining about problems. AI is a top initiative for most companies and a failure to address the network prior to deployment of AI applications can lead to poor performance and obviate any kind of business benefit that was hoped to be gained.
