What happened
Hewlett Packard Enterprise (HPE) unveiled the first four systems of its 13th-generation ProLiant server line, specifically engineered for enterprise AI inferencing and agentic workloads. The new hardware, including the DL585a, DL525, XD245, and XD285, utilizes AMD’s 6th Gen Epyc 9006 'Venice' CPUs. A key differentiator is the integration of iLO 8 management software, which introduces multi-party authorization for high-impact AI actions and expanded post-quantum cryptography capabilities to address emerging security risks associated with large-scale AI deployments.
Hewlett Packard Enterprise announced the first four systems of its 13th-generation ProLiant server line, targeting enterprise AI inferencing and agentic workloads. The systems include the ProLiant DL585a, a 10U server supporting up to eight double-wide GPUs from Nvidia, AMD, or Intel, and two AMD Epyc CPUs. The DL585a is scheduled for availability in March 2027 and is designed to handle high power and thermal demands for workloads like retrieval augmented generation (RAG) and agentic AI.
The ProLiant DL525, a single-socket, air-cooled 1U system, is set to become available next month. It features a single 256-core AMD Epyc CPU and supports both DDR5 and MRDIMM memory. HPE positions this system for high data-throughput tasks such as AI inference, electronic design automation (EDA), and fraud detection, noting a shift in customer needs from GPU-heavy to CPU-dense configurations for agentic workflows.
Two new XD systems, the XD245 and XD285, are also part of this generation, both housed in 2OU chassis within a 21-inch ORV3 rack infrastructure. The XD245 is liquid-cooled with four half-width dual-socket nodes, while the XD285 is air-cooled with two such nodes. Both support up to 256-core AMD CPUs and are designed for customers with modernized datacenters capable of handling high-density power and cooling requirements.
All new systems utilize AMD’s 6th Gen Epyc 9006 'Venice' CPUs, released in July. HPE emphasized that these servers are built to accommodate the growing token throughput and PCIe needs of AI workloads. John Carter, HPE’s vice president of product management for Compute, stated that customers are increasingly running out of power before space, driving the design of larger, air-cooled systems for mid-tier service providers and enterprises.
Source details: nextplatform.com ↗
Why it matters
This launch addresses the specific infrastructure bottlenecks and security concerns arising from the shift toward agentic AI and high-density inference. By security controls like multi-party authorization directly into the hardware management layer, HPE is responding to the unique risks of AI agents executing critical tasks. The focus on air-cooled, high-density designs also reflects a practical industry move to manage power constraints without exclusively relying on liquid cooling, offering enterprises a scalable path to integrate AI into existing datacenter footprints.
The introduction of these servers highlights the evolving infrastructure requirements for AI, particularly the balance between compute density and power management. HPE’s focus on air cooling for the DL585a and DL525 suggests a strategic move to make high-performance AI hardware more accessible to enterprises that may not have immediate liquid cooling capabilities, addressing a common barrier to AI adoption.
Security is a central theme of this launch, with HPE integrating enhanced management and security software into the iLO 8 platform. This includes automatic encryption of storage drives from boot and multi-party authorization for high-impact actions performed by AI or agents, such as key provisioning and secure erase. This feature requires two authorized human approvers, directly addressing the risks associated with autonomous AI agents executing critical infrastructure tasks.
HPE also expanded post-quantum cryptography (PQC) capabilities in the management construct. This is a proactive measure against the 'harvest now, decrypt later' threat, where data stolen today could be decrypted once quantum computing becomes sufficiently powerful. By these capabilities at the silicon and management level, HPE aims to provide a foundational root-of-trust for AI infrastructure.
The launch occurs against a backdrop of massive capital expenditure in AI infrastructure, with Bain & Co predicting annual spending could reach $1.5 trillion by 2031. HPE’s new systems are part of the OEM response to this demand, offering enterprises specific tools to manage the complexity and security risks of deploying large language models and agentic AI at scale.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Monitor the actual availability dates, with the DL525 expected next month and the DL585a in March 2027. Watch for independent benchmarks comparing the performance of these AMD-based systems against Nvidia-centric competitors. Additionally, observe how other OEMs respond to the specific security features, such as multi-party authorization for AI agents, which may become a standard requirement for enterprise AI procurement.
The ProLiant DL525 is expected to be available next month, while the DL585a is scheduled for March 2027. The XD245 and XD285 are set for release next year. Monitoring actual shipping dates and early customer adoption will provide insight into the market reception of these specific configurations.
Independent performance benchmarks comparing the AMD Epyc-based ProLiant Gen 13 systems against competing Intel and Nvidia-centric platforms will be crucial for enterprises evaluating total cost of ownership and performance per watt for AI inference workloads.
The multi-party authorization feature for AI agent actions is a novel security control. Watch to see if this becomes a standard requirement in enterprise AI procurement policies or if other vendors adopt similar hardware-level controls to mitigate risks associated with autonomous AI systems.
HPE’s emphasis on post-quantum cryptography in server management may signal a broader industry shift toward quantum-resistant infrastructure. Observing how other OEMs and cloud providers integrate PQC into their hardware management stacks will indicate whether this is becoming a baseline security expectation for AI datacenters.