Around this time last year, most of the AI talk that Trish Damkroger heard revolved around generative AI, the prompt-and-respond form of AI that took hold soon after OpenAI rolled out its initial ChatGPT chatbot in late 2022. But that changed in the last twelve months, with the focus shifting to sovereign AI, the ability of organizations to own its AI technology – from infrastructure to models – and to keep their data local, secure, and compliant with evolving government regulations.

Damkroger has been in the HPC field for more than three decades, with long stints at Sandia National Laboratory and Lawrence Livermore National Laboratory as well as more than five years with Intel. Since 2022, she has been HPE’s senior vice president and general manager of HPC and AI infrastructure solutions. With Cray and similar supercomputing systems, the HPC field has long been doing sovereign computing, she told The Next Platform.

It has included the same elements, she said. However, AI is adding a whole new layer of complexity.

“Sovereign is still about control, compliance, and security, but AI adds so many more layers of threat,” Damkroger said during an interview at the HPE’s Discovery 2026 show this month. “We are used to being air-gapped, so you can reduce that attack surface from external threats.”

That said, what’s worrying many in the field is the internal threats, and in particular, the threat that comes with having autonomous AI agents running jobs that require minimal if any human intervention.

“With AI, you could put together the infrastructure, maybe you air-gap it, maybe you do other things to control who has access to it, but what about the agents?” she asked. “Now we have agents running on and doing some of our work for us. We're talking about the classification of agents on these different machines. It's really about who has access.”

At the SC25 supercomputing show last fall, HPE unveiled a new supercomputing architecture that included an expansion of its Cray supercomputing portfolio that included three new blades and that can be deployed into the Cray Supercomputing GX5000 platform that was announced a month earlier to drive supercomputing work as AI and HPC converge. It included such capabilities as direct liquid cooling for the second generation of exascale systems and to drive research in such areas as medicine and climate science.

The focus for HPE at this week’s ISC High Performance 2026 supercomputing show in Germany is on software, at a time that quantum computing is being added to the HPC and AI convergence and as multi-tenancy, regulatory compliance, and security become larger considerations. As Damkroger said, “multi-tendency is what most of our customers have, and even if you're at a national lab, you have different users trying to use the system.”

At last week’s HPE Discover show in Las Vegas, executives emphasized the need for security capabilities in the vendor’s Private Cloud AI platform as agentic AI expands, with such enhancements as adding more confidential computing capabilities in its Morpheus management platform, air-gapped features in Private Cloud AI for security and data sovereignty, and AI-native networking and firewall features.

Similarly, security plays a central role for HPE at ISC. Damkroger noted that the architecture itself was addressed in November with the new Cray computing system and Cray Supercomputing Storage Systems K3000, as well as putting the Slingshot 400 interconnect into the Cray GX5000 platform.

At ISC, “we've really looked at in the software stack and said, ‘OK, these are the new requirements and multi-tenancy all the way down through our storage and networking,’” she said. “The security elements [are] going to be a continuing theme that we're going to really push because that is what we are hearing from our customers as they are more and more concerned because of all the cyber risk. Software is the big thing that we're announcing, in the whole multi-tenancy all the way through the system, so not just on the compute, but networking and storage. That’s all about security.”

Central to that is HPE’s introduction at ISC of its new Supercomputing Programming Software, which includes integrating vendor, open source, and HPE software tools into a single programming environment for easier setup and deployment. In addition, software tool updates are delivered as containerized environments, making them safer and more efficient and able to be delivered across systems without having to rebuild anything.

HPE is extending the reach of the software into the ProLiant server line, including the ProLiant DL and XD servers, which are optimized for AI workloads like training, tuning, and inference.

In addition, HPE is unveiling a new version of the Slingshot 400 software to include multi-tenancy functions that creates secure separation between users and restricts unauthorized routing from open source and third-party groups. The new capabilities can be applied to switches that have already been deployed.

---PIC: HPE Slingshot---

The Slingshot 400 switch ASIC, presumably called “Rosetta 2” internally at HPE, has 51.2 Tb/sec of aggregate bandwidth using 100 Gb/sec SerDes with PAM-4 modulation; it supports 64 ports running at 400 Gb/sec. With cable splitters, you can have 128 ports at 200 Gb/sec and 256 ports running at 100 Gb/sec. HPE says Slingshot 400 can support 260,000 endpoints with a three level network, which is a lot better than the scale of InfiniBand while still being an Ethernet switch. The Slingshot 400 can be used in the Cray EX4000, Cray GX5000, Cray XD, and ProLiant server platforms. There is a matching Slingshot dual network interface card implemented in a mezzanine form factor used in the Cray EX and GX machines with a pair of 400 Gb/sec ports and a single port, single chip network interface card for the Cray XD and ProLiant server platforms implemented in a PCI-Express 5.0 form factor.

On the storage side, HPE developed a new GUI and API to create multi-tenancy setup and management for its Cray Supercomputing Storage Systems E2000 file systems. The GUI eases the setting up and management of multi-tenancy environments by making them more intuitive, and the API drives efficiency via automation in large-scale multi-tenant scenarios.

--PIC: HPE E2000---

All of this is important to help organizations keep up with the rapidly accelerating rate of innovation in the AI space and the strong adoption of AI technology by threat actors.

“I feel for our customer base because there's so much change,” Damkroger said. “The pace of AI is so much faster than anything we've ever faced before and the new tools are, again, great, but they're also a threat because they're going to be used against you. I feel for the governments that are trying to do the compliance of this. Think about the whole thing with Anthropic [and its conflict with the Trump Administration over the security and use of its Mythos AI and Fable 5 frontier models]. It's a challenge just to even keep up with this pace of change. It's a journey we're all on. We work very closely with our vendors, Nvidia, AMD, Intel, because you have to have security down to the core. But then you also look at providing an external root of trust so that you can make sure that you are really looking at that source.”