Hewlett Packard Enterprise’s Marvis artificial intelligence engine now resolves 70% to 80% of potential network disruptions autonomously, notifying IT personnel only after a fix is successfully deployed. Rami Rahim, EVP, president and GM of networking at HPE, detailed the autonomous capabilities during the company’s Networking Investor Day.
What Autonomous Networks Actually Fix Today
The system targets common infrastructural friction points without human intervention. Poor performance, slow connections, roaming failures and DNS problems that keep clients from connecting fall squarely into the autonomous domain. HPE calculates its 70-80% success metric by logging every autonomous remediation. The Marvis AI engine processes every customer service ticket first, executing a repair or learning the signature for future automated resolutions.
Adoption depends on trust. Customers activate these automated capabilities individually, opting in one feature at a time. Rahim compared the progression to advanced driver-assistance systems, moving incrementally from basic cruise control to lane assist and ultimately to full autonomy. Over a recent 90-day window across HPE cloud platforms, customers authorized approximately 10,000 self-driving actions. When Marvis encounters a trouble ticket matching an unapproved category, it presents diagnostic evidence and a recommended fix, asking the operator for authorization. Once approved, identical future occurrences resolve automatically.
Operational Boundaries and Remaining Human Tasks
Autonomous routines stop short of active security threats. Security incidents involving an active attacker remain outside the system’s current scope, requiring human operators to evaluate intrusion vectors and propagate defensive policies across switches and firewalls. HPE enforces strict boundaries called no-fly zones. Any configuration change impacting regulatory compliance demands explicit administrative approval, complete with automatic rollback capabilities and post-change validation.
Regulated sectors like healthcare adopt autonomous management much slower than general enterprise workplaces due to these strict compliance hurdles. Rahim argued that the risk of the AI engine introducing configuration faults is essentially non-existent when compared against human operators, whose manual input errors account for the vast majority of network outages. Rival Cisco pursues a parallel path, having expanded its AgenticOps model to include autonomous troubleshooting across campus, branch and industrial environments.
Enterprise Deployments and Non-Human Traffic Pressures
Early deployments demonstrate measurable operational relief. ServiceNow cut its network trouble ticket volume by over 90% while expanding business operations without increasing its IT headcount.
Network automation gains urgency as machine-generated traffic rapidly eclipses human usage. Cloudflare data indicates bot traffic surpassed human traffic, reaching nearly two-thirds of all internet traffic. Non-human workloads do not generate traditional user complaints when encountering latency or packet loss. Consequently, continuous telemetry measuring end-to-end user and device experience supplants simple up-or-down node monitoring.
Carrier Strategies and Infrastructure Scaling
Telecom operators occupy a critical chokepoint as the primary on-ramp and off-ramp for enterprise AI consumers. Carriers are pursuing multiple revenue streams, including internal cost reduction via agentic AI operations, high-bandwidth low-latency data center interconnect fabrics for AI clouds, and localized AI inference services. Initial inference offerings leverage GPU-as-a-service models deployed primarily outside the U.S., allowing enterprises to buy sovereign AI compute from trusted local carrier brands rather than competing directly with hyperscalers.
HPE’s carrier footprint shifted following the completion of its sale of the Telco Solutions business, including its subscriber management and OSS software, to HCLTech. HPE’s current telecommunications strategy focuses on core hardware and software products, such as the PTX12000 routers and agentic Routing Director features unveiled at MWC Barcelona.
Financial Tailwinds and Hardware Supply Dynamics
Rising enterprise demand is outpacing manufacturing capacity. Networking growth remains supply-constrained, though relief is anticipated in fiscal 2027 following a doubling of networking supply purchase commitments last quarter. Financial forecasts received a major boost from the AMD Helios architecture. HPE secured its first Helios order through a $1.2 billion deal with Vultr, representing an estimated networking opportunity exceeding $1 billion across a two-year horizon. Additionally, Oracle announced an agreement to deploy HPE Juniper Networking routers and switches across its AI data centers.
HPE maintains an integrated hardware and software stack spanning silicon, operating systems, security tooling, and professional services. Rahim noted this breadth preserves customer architectural choice. Enterprises can deploy fully air-gapped environments managed entirely on-premises, utilize HPE professional services, or leverage country-specific clouds designed to satisfy regional compliance frameworks like FedRAMP, Australia’s IRAP and U.K. public sector requirements. In instances where regional clouds faced disruptions from active war, HPE executed seamless failovers to infrastructure located in allied jurisdictions.
