HPE 发布面向 AI 时代的首批新一代 ProLiant Gen 13 服务器

COMPUTE

It is impossible to miss the impact that the rapid adoption of AI continues to have on the datacenter infrastructure space and everything that goes with it – power consumption and costs, for example – both over the past few years and in projections into the next decade.

In a report released late last month, Bain & Co analysts predict that by 2031, annual spending on AI infrastructure could hit $1.5 trillion, which includes everything from new datacenter infrastructure and compute capacity to GPU upgrades, memory, and networking hardware. Power capacity, which in the most advanced AI datacenters is nearing 1 gigawatt, could double by next year and power past 9 GW by the end of the decade.

The ravenous appetite among hyperscalers for infrastructure continues to grow, with capital spending from the likes of Microsoft, Amazon, Google, Oracle, and Meta expected to hit $780 billion this year, or about five times the amount from just three years ago, they wrote.

OEMs understand this and are rolling out datacenter systems aimed at feeding the AI beast. The latest example comes this week, with Hewlett Packard Enterprise unveiling the first of its ProLiant Gen 13 systems with greater compute, memory, and networking capabilities for enterprise AI inferencing and agentic workloads and a focus on such areas as security, performance, and reliability, which always have been important in datacenter systems but can take on greater roles with AI workloads.

The first four 13th generation ProLiants are aimed at the highest performing dense computing and GPU workloads, John Carter, vice president of product management for HPE Compute, tells The Next Platform. Those include traditional enterprises trying to get the most performance out of their infrastructure or companies that are pushing what can be done with agentic AI and big model simulation.

“That kind of scale and performance brings new security risks,” Carter says. “It brings new complexity in terms of managing many, many systems. We're talking about companies taking their most core critical IP and data and now putting it into these kinds of LLMs [large language models] and simulations and this kind of stuff. That's critical to be secure. The management and security that we've continued to bring from multiple generations here becomes even more important because it really lays that foundational silicon level root-of-trust to allow a build into the other levels of the infrastructure.”

Given that, what HPE is doing this week is both unveiling the latest generation of ProLiant systems as well as improvements in the software that manages and protects them. Common to all of the new systems is the use of AMD’s 6th Gen Epyc 9006 SP7 “Venice” CPUs, which the chipmaker released in July.

Anchoring the new systems is the ProLiant DL585a, a 10U server that can hold up to eight double-wide GPUs from Nvidia, AMD, or Intel and two of the latest AMD Epyc CPUs, which offer up to 256 cores each. The system also includes sixth-generation PCI connectivity. The server, which will be available in March 2027, is built to accommodate current and future power and thermal demands and address the growing token throughput and PCIe needs of such AI workloads as inferencing, retrieval augmented generation (RAG), and agentic AI, according to Carter.

“Most of our customers who are using a PCIe-type GPU, what they're really telling us is, ‘We’re running out of power long before we run out of space, so make it bigger,’” he says. “We want to keep air cooling, so that's really where this is positioned. This is more for the enterprise, for the mid-tier service providers, those kinds of folks.”

The ProLiant DL525, which becomes available next month, is single-socket and air-cooled 1U system with a single 256-core AMD chip with high-bandwidth, 16-chnanel memory, and support for both DDR5 and MRDIMM memory. The system is designed for high data-throughput workloads like AI inference, EDA, and fraud detection. It fits into standard 19-inch rack.

“Now we're shifting from GPU-focus to CPU-focus,” Carter says. “This is where we're going to talk more about the agentic side of the story, where we're expecting the ratio to shift from two to eight CPUs to GPUs, now more like one to one. The expectation here is as agentic gets bigger, our customers are going to need these kinds of dense CPU platforms as part of that workflow to offload many of the actions that just perform with their own CPU.”

The two new XD systems, both set for release next year, are follow-ons to what HPE in the past has referred to its Apollo 2000 or XD 2000 series, according to Carter.

“These new XD servers are now put into an open 21-inch ORV3 rack infrastructure part,” he says. “They sit in a wider and specialized rack. There are shared power and shared cooling capabilities between those systems to bring more efficiency. What you really get here is the ultimate density in compute. I can drive many, many, many cores in the same data space. It's really specialized for customers who have built modernized datacenters. . . so high density power to the rack, liquid cooling available or high-efficiency air cooling.”

Both the XD245 (below) and the XD285 are housed in a 2OU chassis, with the XD245 including four half-width dual-socket nodes compared with the XD285 having two such nodes. Each can run up to 256-core AMD CPUs, 16 memory channels per CPU, and MRDIMM support. The XD245 is liquid-cooled; the XD285 is cooled with air.

Weaved throughout these systems is enhanced management and security software, including in iLO 8. The improvements put a particular focus on security, a top-of-mind issue for organizations when it come to AI, and agentic AI in particular. iLO 8 offers HDD or SSD storage drives that automatically encrypt and decrypt the data that’s on the hardware from the moment the system boots. There also is multi-party authorization, which ensures high-impact actions by AI or agents, such as key provisioning and secure erase, have two authorized human n approvers, with Chris Bradley, director of mainstream computer customer advocacy and technical enablement at HPE, adding that they embedded it into individual nodes to work as rules within iLO.

HPE also expanded post-quantum cryptography (PQC), a security capability aimed at protecting data in the upcoming quantum computing era. Current encryption methods like AES-256 work well now, data protected by the technology will become more vulnerable when cybercriminals are able to use quantum computing’s much more powerful capabilities. A key worry now is what’s called “harvest now, decrypt later,” where threat actors are holding stolen data with plans to decrypt it when quantum computing becomes available to them.

HPE is “ensuring that we've got the both quantum cryptography algorithms and capabilities already in the management construct,” Bradley tells The Next Platform.

HPE’s newest ProLiant systems add to the rapidly growing infrastructure choices enterprises and HPC organizations have for their AI datacenters that the Bain analysts wrote about in September. But as they noted, there are cautions that need to be heeded in the rush.

“The unprecedented speed and scale of the AI buildout, with billions flowing into chips, data centers, networks, and power systems, have focused attention on the challenge of building capacity,” they wrote. “But the more important question may be whether enough economic value can be created to justify it.”

SPONSORED FEATURE: HPE and NVIDIA help organizations apply accelerated computing, software, enterprise infrastructure, networking, control plane, and services to the workloads and business outcomes that matter most.

October 1, 2026

September 30, 2026

AMD Buys World Labs For $8.2 Billion To Get AI Model Cred