HPE advances its self-driving networking strategy for the AI era

Ben Baker, HPE's director of data center networking, has a blunt diagnosis for why enterprise networks keep breaking: too much complexity, not enough insight. Speaking with theCUBE Research's Bob Laliberte, Baker described network operators drowning in telemetry data while still flying blind on what's actually happening across thousands of virtual and physical devices. One misconfigured switch, he noted, can take down an entire network and cost a company real money and reputation — and without solid automated rollback, fixing it can eat up days.

HPE's answer is what it calls the self-driving network, a system meant to spot trouble before it cascades rather than after. The technical backbone is Apstra Data Center Director, now paired with HPE Morpheus for hybrid cloud automation. Baker leans hard on the graph database underneath Apstra as the differentiator, since it maps relationships between network nodes instead of treating each data point in isolation. That, he argues, is what lets HPE correlate scattered signals into something an operator can actually act on — a claim he frames confidently: "we're starting on second base and everyone else is catching up."

Sitting above all this is GreenLake Intelligence, HPE's agentic AI layer meant to tie networking, compute, storage, security and cloud into one operational model rather than a pile of separate consoles. Some of the AIOps muscle here comes from Juniper's Mist platform, acquired alongside the rest of Juniper, which HPE says was built around a smart idea from early on: linking tech-support staff directly with data scientists so problems get diagnosed with actual pattern recognition, not guesswork.

Integration has been the harder story since the Juniper deal closed. Baker pushed back on skepticism that folding Juniper into HPE would slow product development, pointing to Marvis — HPE's AI networking assistant — now showing up inside Aruba Central, and Mist gaining support for HPE's CX switches. On the hardware side, HPE is also shipping the QFX5250, a fully liquid-cooled switch with 64 ports running up to 1.6 terabits per second, aimed squarely at AI workloads that keep demanding more bandwidth with no sign of slowing down.

What Baker keeps returning to, though, isn't the speeds and feeds. It's the idea that people should stop thinking about networks at all — treating them like electricity or water, something that just works without anyone noticing. That's a modest-sounding goal dressed up in a lot of AI branding, but it's also the honest one: nobody wins an award for uptime nobody had to think about.