- HPE's self-driving network detects and fixes 70-80% of potential issues on its own, said networking chief Rami Rahim
- But networks still depend on people when responding to active attackers and making changes involving regulatory compliance
- Telcos' earliest AI inference businesses are sovereign GPU-as-a-service offerings outside the U.S., Rahim said
HPE's self-driving network can now detect and resolve 70-80% of potential network issues without a human operator, telling IT staff about the fix only after it is made, Rami Rahim, EVP, president and GM of networking at HPE, told Fierce Network.
However, networks still depend on people when responding to active attackers and making changes that touch regulatory compliance, and customers decide action by action how much control to hand over, Rahim said during an interview at the company's recent Networking Investor Day.
What can a self-driving network decide on its own today?
The issues the network handles autonomously include poor performance, slow connections, roaming failures and DNS problems that keep clients from connecting, Rahim said.
HPE arrives at the 70-80% figure by logging each autonomous fix, Rahim said. The Marvis AI engine sees every customer service ticket first and either resolves the issue or learns it for future automatic repair.
Customers turn on autonomous actions individually, one capability at a time, as their trust grows. Trust is the biggest barrier to adoption, and a gradual approach is how HPE overcomes it, Rahim said.
"You have to go through all the phases," Rahim said, comparing the progression to cars moving from cruise control to lane assist to full self-driving.
Customers authorized about 10,000 self-driving actions across HPE's cloud platforms in the last 90 days, said Sujai Hajela, EVP and GM of HPE's campus and branch business. When Marvis sees a problem a customer has not yet approved for automatic repair, it presents evidence and a recommended fix and asks permission. Once approved, the same fix runs automatically from then on.
What still requires a human operator?
Security incidents with an active attacker are beyond the system today. Detecting an intruder and pushing policy across the whole network, not only to firewalls, is a capability Rahim said the AI will grow into over time.
HPE has also defined "no-fly zones" that automation cannot touch, Hajela said. Changes that affect regulatory compliance require explicit customer approval and the ability to roll the network back to its prior state, with the new configuration validated as better than the old one. Healthcare and other regulated industries move to self-driving operation more slowly than general workplaces, he said.
Rahim said the risk of the system making bad decisions is "essentially non-existent," arguing that the comparison point is human operators, whose configuration errors cause most network problems.
Rival Cisco is pursuing the same goal. In February, Cisco expanded its AgenticOps model with autonomous troubleshooting for campus, branch and industrial networks.
What results has HPE reported?
ServiceNow cut network trouble tickets by more than 90% and grew its business without growing its IT team, Rahim said.
A large retailer switched from its incumbent Wi-Fi and wired switching vendor for faster software refreshes, which took years before and now take weeks and run overnight, Rahim said
At a shipyard where a previous vendor's wireless network kept failing, the AI traced the failures to radio interference from a neighboring installation and retuned frequencies, Rahim said.
The approach extends the early fixes HPE targeted when it launched automated network actions across HPE Mist and HPE Aruba Central in May.
Why does machine traffic make autonomy more urgent?
Who — or what — generates internet traffic is changing fast. Bot traffic surpassed human traffic in May and has reached nearly two-thirds of internet traffic, according to Cloudflare. That's not surprising, given how fast AI agents are proliferating, Rahim said.
Bot traffic presents challenges for network troubleshooting. When people hit network problems, they complain. "Non-human traffic doesn't necessarily have the same ability to complain," Rahim said. That makes continuous measurement of user and device experience, rather than simple up-or-down monitoring, a requirement, along with self-driving operations, he said.
How are telcos trying to profit from AI traffic?
Telcos carry much of that traffic. "Telcos are the on-ramp and off-ramp for enterprises and for consumers to reach the AI application of choice," Rahim said.
Carriers are pursuing AI opportunities from multiple directions, Rahim said: cutting operating costs with agentic AI operations; selling high-bandwidth, low-latency data center interconnect (DCI) fabrics to the companies building AI clouds; and using central offices near customers to offer AI inference.
The earliest inference businesses are GPU-as-a-service offerings, sold mostly outside the U.S. to enterprises that want sovereign AI and see their local carrier as a trusted brand to buy it from, Rahim said.
"[Operators are] not trying to compete with a hyperscaler because they realize that the capex involved in that would be completely prohibitive for them," Rahim said.
Scaling those services depends on enterprise demand and on carriers willing to invest ahead of revenue, Rahim said. Telcos are "dealing with a lot of traffic growth, with not necessarily as fast a growth in their revenue ramps," he said.
HPE comes to the opportunity with a narrower carrier business than a year ago. HCLTech completed its acquisition of HPE's Telco Solutions business, which included OSS and subscriber management software, on Aug. 5, eight months after HPE agreed to offload the unit. HPE's telco work now centers on products such as the PTX12000 routers and agentic Routing Director features it introduced at MWC Barcelona in March.
What's behind HPE's raised networking forecast?
The largest new factor is AMD Helios. HPE landed its first Helios order in a $1.2 billion deal with Vultr, announced the morning of the investor day, Sept. 30. Rahim said Helios represents a networking opportunity of more than $1 billion over two years.
HPE also cited a potential multiyear agreement, announced Sept. 2, under which Oracle will deploy HPE Juniper Networking routers and switches across its AI data centers. HPE did not disclose the deal's value.
Demand is outrunning HPE's ability to ship. Networking growth is supply-limited this year but will be less so in fiscal 2027, Rahim said. HPE doubled its networking supply purchase commitments last quarter.
Does HPE's stack give customers sovereignty or create dependency?
HPE now builds most of the networking stack, from silicon to operating systems, hardware, security, software and services. Rahim said that breadth widens customer choice rather than narrowing it.
"If they want a fully air-gapped technology stack and they want to manage it completely on their own because their security posture is that stringent, no problem at all," Rahim said. Customers can also have HPE professional services or a partner run operations, he said.
Hajela said HPE operates country-specific clouds to keep data in-country, meets compliance regimes including FedRAMP, Australia's Information Security Registered Assessors Program (IRAP) and U.K. public sector requirements, and can staff support with in-country personnel. When some customers' country clouds were affected by recent war, HPE failed them over to a cloud in an allied country, he said.
Those options govern where data lives and who operates the system. They do not change who supplies it: a customer running an air-gapped HPE stack still depends on HPE for silicon, software updates and hardware. HPE's partners fill gaps mainly in optics, where it relies on outside suppliers for optical transport and sells HPE-branded coherent pluggables built with partners, and in analytics and security information and event management (SIEM) software, Rahim said.
Learn more about HPE on Fierce Network
HPE targets ‘low-hanging fruit’ in self-driving networks push
HPE lands first AMD Helios order in $1.2B Vultr deal
HPE sharpens its AI edge, hones networking chops
HPE offloads its Telco Solutions business in deal with HCLTech
MWC 2026: HPE presses ahead with AI-driven networking strategy following Juniper integration