Hewlett Packard Enterprise (HPE) has unveiled a revised version of its host software and interconnect stack for its Slingshot 400 switches.

Revealed at ISC 2026, the updates see HPE add functionalities allowing operators to more easily set up and securely manage multi-tenancy environments.

Billed as a means to enable “sovereign AI research,” the update can allow for a single supercomputer to be divided into multiple distinct tenants that can be used securely at the same time. It uses Media Access Control (MAC) learning, where a switch dynamically maps source MAC addresses to specific virtual network segments (like VXLANs or VLANs).

By employing hardware-level MAC learning, HPE’s software allows operators to completely isolate routing across distinct environments, while also restricting unauthorized routing from open source or third-party groups.

Users with Slingshot 400 switches already deployed can make use of the newly added features, the vendor touting it for use in national labs’ sovereign AI research.

Slingshot 400 features in the vendor's GX5000 supercomputing platform. It provides 400 gigabit per second (Gb/s) speeds, twice the line speed compared to previous generations, and packs automated congestion management capabilities to allow engineers to run workloads while the switch dynamically adjusts network traffic to maintain performance and reduce packet losses.

The switch line’s multi-tenancy update also extends to storage, with HPE adding multi-tenancy setup and management capabilities to its supercomputing file system.

The effectiveness of its security-optimized multi-tenancy networking would be severely hampered if distinct data pipelines continue to deposit information into a shared, un-partitioned storage pool.

HPE introduced fine-grained multi-tenancy features for its E2000 Cray Supercomputing Storage Systems, which, through a new graphical user interface (GUI) and automation-friendly APIs, allow operators to securely isolate shared storage pools alongside network traffic.

In a statement, the vendor said its updated Slingshot 400 platform “answers customers’ multi-tenant needs for secure separation of users with a unified AI and HPC architecture, supported by innovative direct liquid-cooling for improved energy efficiency.”

The updates coincided with HPE’s Slingshot fabric overtaking 100G Ethernet as the most common interconnect deployed among supercomputers on the biannual TOP500. However, El Capitan, the HPE-supported supercomputer, which had held the title of world’s fastest supercomputer since November 2024, was knocked off the top by the mysterious all-CPU, all-Chinese system known as LineShine (or Lingsheng).