"Customers are adopting "self-driving networks" to deploy networks much faster and to improve the user experience on the network. For these reasons, investment in self-driving networks is rising quickly."
Bob Friday, chief AI officer for networking at Hewlett Packard Enterprise Company (HPE), said this at a media briefing on the 2nd at the InterContinental Seoul COEX in Samseong-dong, Gangnam-gu, Seoul.
A self-driving network is an operations system in which AI detects anomalies such as network failures, performance degradation, security threats, and deteriorating user experience in real time, analyzes the causes, and then automatically recommends or executes necessary actions. Unlike the conventional approach in which people check device status one by one and respond after a failure occurs, the key is to have the network respond on its own before problems affect users.
The chief AI officer, Friday, said, "We are extending a self-driving network strategy that automates network operations using AI from the enterprise edge to AI data centers," adding, "As Generative AI and AI agents spread across corporations' operations, the network is emerging as a core domain that determines the performance, security, and user experience of AI infrastructure, beyond a simple consolidation infrastructure."
◇ From device management to user experience management
Friday explained HPE's network strategy along two pillars: "AI for the network" and "the network for AI." He said, "If enterprise customers once focused on managing network equipment, they are now focusing on the user experience between clients and the cloud," calling it "a paradigm shift."
He said, "Customers are making critical decisions in their infrastructure to better predict and optimize user experience," adding, "They are looking end-to-end across applications, switches, routers, and access points."
On the network for AI, he said, "You have to consider compute, storage, and networking together," adding, "HPE can provide powerful solutions not only to enterprise corporations building private clouds but also to hyperscalers running large-scale public clouds."
He also stressed that approaches to network automation are changing. Friday, the chief AI officer, said, "If past network automation was deterministic automation based on Python scripts, we are now shifting to non-deterministic, agent-based network operations." Agentic NetOps is an operating model in which AI agents interpret network conditions and recommend or execute actions appropriate to the situation.
◇ Strengthening AI networking by integrating Juniper and Aruba
HPE is expanding AI-based network operations capabilities centered on HPE Mist and HPE Aruba Central. The Mist platform is adding support for HPE Networking CX wired access switches, and Aruba Central is introducing AI insights and self-healing automation based on HPE Marvis.
Friday, the chief AI officer, also emphasized the benefits of integration after acquiring Juniper Networks. He said, "With the integration of Juniper Mist and HPE Aruba, two market-leading data science teams have merged into one," adding, "We are moving around a single Marvis self-driving roadmap." He continued, "We are bringing Mist's large experience model to the Aruba platform and applying Aruba's agentic mesh to Marvis," explaining it as "design once, deploy to both platforms."
In AI data centers, the importance of network performance is growing. HPE said traffic generated by graphics processing units (GPUs) comes out at far higher volumes than traditional servers, and east-west traffic between GPUs accounts for most of it. As a result, interface speeds beyond 400G and 800G up to 1.6T are being required. HPE said it is applying QFX switches in blade form to the AMD Helios rack-scale GPU cluster architecture and has also released the QFX5250 switch supporting 1.6 Tbps.
◇ "Security must be built in from the start"
Security was also presented as a core pillar of the self-driving network strategy. Carlos Gomez Gallego, chief technology officer (CTO) for HPE Aruba Networking in Asia-Pacific and Japan, said, "I want to emphasize the zero trust principle most when designing networks and security," adding, "Instead of adding security later after building the network, you must embed security from the initial design stage."
He noted that corporations are creating data silos by using multiple tools to manage networks and security. Gomez Gallego, the CTO, said, "If you apply networking and security from the design stage, the network team and the security team can work with the same tools and the same data," adding, "Applying zero trust from day one is the key point to reducing future complexity and protecting business and assets."
AI-based automation also helps narrow the gap between network operations and security operations. Gomez Gallego, the CTO, said, "The value of AI is in maximizing data utilization," adding, "The more high-quality data you obtain through network and security equipment, the more actions a self-driving network can take based on that data."
To Korean corporations, he advised starting small and expanding successful cases. He said, "Do not try to do all network and security policies at once—start small," adding, "The starting point is mapping the company's core assets, people, applications, and data."
Friday, the chief AI officer, also stressed the importance of validating in real operating environments. He said, "Like the saying that you should taste the wine before you buy it, you should also do a PoC (proof of concept) for self-driving networks in production environments," adding, "Corporations should run PoCs in production and first confirm the business value."