HPE puts all its chips in the agentic AI pot

Another OEM has decided we're now in the agentic AI age

HPE Discover 2025 In another sign that AI agents have taken over the enterprise zeitgeist, the theme at HPE Discover this year is all about cramming the automated workflow bots anywhere they'll fit, whether or not agentic AI is mature for all use cases. 

While hardware was a focus as always, the standout item from HPE CEO Antonio Neri's Las Vegas keynote was GreenLake Intelligence, a "new agentic AI framework" that, per the company itself, means adding agentic AI pretty much everywhere it could feasibly find a niche.

As explained by HPE, the GreenLake Intelligence framework is designed to serve as a single point of agentic AI contact across HPE's various systems, all managed from the central HPE Greenlake hybrid cloud platform. 

A number of examples were given during the keynote, and in briefings surrounding the announcement. Those include things like networking, where HP's Aruba products will see the incorporation of "new agentic mesh technology" that will be nestled behind a "multi-modal, conversational networking copilot" able to do root-cause analysis of networking problems and use a variety of different reasoning agents to figure out how humans need to fix problems it finds. 

Agents will also be coming to OpsRamp, storage management, and other areas, where they'll be used to automate things like cost operations, sustainability and business services as well, according to HPE.

For those hoping that the addition of agentic AI across HPE's hybrid stack means agents are finally good enough to take action without human oversight, sorry - despite years of promising that AI would soon be able to take over all that legwork, we're still not quite there.

HPE describes GreenLake Intelligence agents as "autonomous" in all the literature from this week's event, but they also explained that humans will remain in the loop to make a final call. Autonomous, but deferential, in other words. 

When asked about actual autonomy for its AI agents, HPE SVP and GM for cloud and OpsRamp Varma Kunaparaju told The Register that these things take time, but that's no reason to postpone the agentic future. 

"A few use cases which are very well-defined are autonomous," Kunaparaju explained. Fixing a network bug, or addressing a storage snafu - basically anything that exists in a tightly controlled single domain is already there, Kunaparaju said. Beyond that things get fuzzier.

"But if you broadly apply a probable root cause to go in and self-remediate, I would say it is a crawl-walk-run journey," the OpsRamp chief said. "We're still crawling and doing a little bit of walking."

Getting there won't be instantaneous, Kunaparaju told us, but he thinks the next year will bring considerable AI agent autonomy advances - the world was barely talking about AI agents at all a year ago, after all.

"Making the entire enterprise run autonomously is a journey," Kunaparaju said. "Systems need to evolve, processes need to evolve and implementation sophistication needs to come where these systems can be integrated to fully achieve the efficiencies that AI can drive." 

The software will get there, Kunaparaju said, but hardware is already where it needs to be to enable future AI efficiencies - so build it, in other words, and the AI you're dreaming of will come sooner or later.  

New hardware is all about serving AI

HPE is a hardware company at its heart, and there were announcements about new physical systems at this week's event, too. New systems took a backseat to AI, though, with all the hardware announcements being made in the service of furthering artificial intelligence products. 

New HPE Compute Cray XD690 systems, able to support eight Nvidia Blackwell GPUs were announced, as were new Blackwell-fitted ProLiant systems, all as part of HPE's expanded Nvidia AI factory (all-in-one software and hardware stacks offered by Nvidia for training AI on a company's data) alliance. 

HPE and Nvidia have been working together since last year to offer Nvidia AI factories on HPE hardware, and this year it's just more of the same. Whereas 2024 saw Nvidia AI factories from HPE only available for enterprise customers using the HPE Private AI cloud, now there is a "composable" service tier for service providers and companies operating at scale, and a "sovereign" tier for governments and other customers with enhanced data security and sovereignty requirements. 

There were other small announcements as well, but they're even less deserving of a formal footnote than the new Nvidia-driven, AI-first hardware. If last year's theme at Discover was HPE's boarding of Nvidia's hype train, then this year was just a quick stop before going full steam ahead toward pushing the world to embrace agentic AI, ready or not. ®