Why future AI data centers need to be open and heterogeneous

AI is evolving to be more than just a workload.

As initiatives transition from training to inferencing, from generative AI to agentic AI, AI will increasingly operate as a service and run across distributed environments, like the edge, cloud, and sovereign systems.

Future AI systems will also be heterogenous. Instead of being dominated by one architecture or platform, different workloads with different needs will require a diverse mix of accelerators, processors, and infrastructure.

Yet with the next era of AI comes a new set of IT requirements that we have to prepare for now.

Reimagining a new operating model for AI

In order to work, AI will require a new operating model that features core elements:

• Scalable compute to reliably run workloads across distributed environments
• Optimized performance for faster storage and greater networking efficiency
• Heightened security to protect all data, models, and AI-generated insights
• Operational efficiency to manage infrastructure, energy, productivity and cost

Additionally, as AI scales and innovation broadens – just as we’re seeing with an emergence of neoclouds, colocation providers, and sovereign AI programs – AI will grow into a larger ecosystem. That ecosystem has to be interoperable to work seamlessly.

To ensure effective design and implementation of key capabilities, along with a resilient ecosystem, we have to take an open compute standards-based model approach to make the future of AI successful.

Making AI open and heterogeneous at every layer

At this year’s OCP Global Summit, HPE is demonstrating our ongoing commitment to contributing to open compute and ushering in a new era of AI.

We also believe that the future of AI will be more heterogenous and open compute will be critical to effectively integrate diverse silicon architectures to deliver purpose-built end-to-end solutions for AI.

As a leader in designing and delivering fully integrated systems, from the edge to exascale, we are bringing open, heterogenous technologies to every layer of the stack: compute, networking, and large-scale HPC and AI systems.

Open compute: Advancing modular, secure, AI-accelerated compute from the edge to cloud

To support a rapidly evolving AI landscape, including the transition to agentic AI and inference, we are providing enterprises with open standards-based modular compute that we continue to advance with our newly launched HPE ProLiant Gen13 servers.

Our new open compute, which includes the HPE ProLiant DL525, DL585a, XD245 and XD285 Gen13 servers, delivers a flexible design for agility to run various workloads, and scale as needed. HPE ProLiant servers are also deeply rooted with trust, and the latest generation builds on our leadership chip-to-cloud security to protect data and system integrity through automatic security from HPE Integrated Lights-Out 8 (iLO 8) with enhanced HPE Compute Ops Management.

Open networking: Ethernet-based scale-up, scale-out, and scale-across

Open Ethernet is fundamental to future AI data centers, and we’re an industry leader in open Ethernet-based scale-up, scale-out, and scale-across.

In addition to offering options in network operating systems, like SONiC, our switches have embraced open industry specifications and initiatives, such as Open Rack V3 (ORv3), Ultra Ethernet Consortium (UEC), Ethernet for Scale-Up Networking (ESUN) and Ultra Accelerator Link (UALink). These developments help build interoperable and standards-based networking for AI infrastructure.

We are the first OEM to extend open Ethernet in rack-scale level infrastructure with scale-up switches that we are delivering through the AMD Helios™ AI Rack by HPE systems. Enterprises are already adopting these rack-scale solutions to speed up their AI initiatives.

Open large-scale AI systems: Leadership-class supercomputing and AI infrastructure

Scientists and researchers are turning to solutions that effectively converge modeling and simulation with AI to accelerate scientific discoveries. These workloads require extreme scale with greater flexibility and density to maximize existing capacity and resources.

We’re building on our proven world-leading exascale-class supercomputing with next-generation open infrastructure, the HPE Cray Supercomputing GX5000, to deliver unprecedented performance with the industry’s highest density for large-scale HPC & AI workloads. The GX5000 offers various blades that feature the industry’s latest CPUs and accelerators to provide computing options that address specific workload needs.

The GX5000 also features HPE Slingshot networking based on Ultra Ethernet Consortium (UEC) specs, in addition to supporting third party components and switches.

Extending decades of direct liquid cooling leadership to AI

As a leader in direct liquid cooling, with more than 50 years of expertise and 300 technology patents, we are building on our innovation to support higher thermal capacity for high-powered CPUs and accelerators across our open compute and large-scale HPC & AI systems.

Latest innovation also includes the industry’s first direct liquid cooled ORV3 switch, a next-generation Open Rack capability, to handle extreme heat for modern AI data centers.

Join HPE at OCP Global Summit 2026 at booth D24

OCP will bring immense benefits to AI. It will pave the way for a new ecosystem that will help organizations reliably scale, achieve operational efficiency, and lower costs.

We are committed to making ongoing contributions to OCP and look forward to highlighting our vision and solutions for open technologies at OCP Global Summit.

I encourage you to join us at our booth D34 to learn more about our open, heterogenous solutions, attend our sessions, including our keynote at on Tuesday, October 13 at 10:15am PT.

Come speak with our team on how we can work together to contribute to OCP and power a new era of AI!