HPE flexes some Juniper muscle at MWC26

Hewlett Packard Enterprise Co. announced new networking, compute hardware, cloud operations software and financing updates for service providers at this past week’s MWC26 in Barcelona. The updates center on meeting the new demands being created by artificial intelligence reshaping every aspect of network design — from centralized data centers to distributed edge environments.

AI adoption is driving more traffic into AI data centers, accelerating investment by hyperscalers and neocloud providers. In those environments, ultra-low latency and high reliability are no longer optional. They are basic requirements for delivering AI as a service.

Traditional network traffic is still growing but it’s quickly being dwarfed by AI traffic. But that traffic behaves differently and it’s far more sensitive to delays. The existing networking oversubscription model cannot handle today’s AI requirements.

AI creates new networking demands

To address these changes, HPE is expanding its Juniper-based PTX routing portfolio. The updates include new PTX12000 modular routers, which support dense 800G connectivity initially and can scale to 1.6 terabits per second without major redesigns. Additionally, HPE introduced a new line of PTX10002 fixed-form routers — a smaller, more efficient option for building AI networks and connecting data centers. The PTX Series routers run on Juniper Express 5 silicon, with an emphasis on throughput, deep buffering and power efficiency.

Routing is what Juniper has always done best and at MWC I had a chance to meet with HPE Executive Vice President AE Natarajan about the new products. He told me customers are building out larger and larger graphics processing unit clusters, which are geographically distributed and need to be connected making the network central to the growth of AI. “The appetite for the PTX12000 platform is very strong right now,” he said. “Some telcos are making big leaps into building out AI networking fabrics, inference edges or sovereign clouds.”

Custom silicon creates HPE differentiation

Though there are many scale-up, -across and -out products available, these are powered by the Express 5 Silicon, which came to HPE via its acquisition of Juniper Networks. The new application-specific integrated circuit delivers roughly 49% more power efficiency than the previous generation, with PTX10002 systems achieving up to a 54% improvement over earlier platforms. The silicon was built with AI in mind an includes things like inline MACsec (Media Access Control Security) for integrated security.

Another aspect of Express 5 is the load balancing and quality of service algorithms. With AI networking, it’s not enough just to be fast;  the network needs to know how to handle congestion, and that’s something merchant silicon doesn’t not handle well. “We built load balancing capabilities with least amount of switch over drops making it ideal for AI,” Natarajan explained. “By not dropping packets, the GPUs are never having to wait for the network to catch up.”

HPE is also updating Juniper Routing Director to be agentic AI-ready. The software platform provides end-to-end transport and wide-area network automation for service providers and large enterprises. Many large operators are building their own AI copilots and customized models rather than relying on vendor-provided assistants. With this update, Juniper Routing Director can integrate with customer-built AI copilots, allowing operators to automate network operations and speed up troubleshooting.

ProLiant gets an upgrade

On the compute side, HPE is introducing new ProLiant platforms aimed at service providers, including the Compute EL9000 chassis and EL140 Gen12 servers. These systems are designed to handle higher network traffic density for AI and 5G workloads. HPE is also integrating Juniper’s cloud-native routing software directly into select ProLiant servers, combining routing and compute into a single system for radio access network deployments. This includes the 1U HPE ProLiant Compute DL110 and the new 2U HPE ProLiant EL140 Gen12 servers.

Telco RANs are now software-based and do not need to have a separate compute server and routing platform. These layers have collapsed, giving HPE the opportunity to bring its strength in compute and networking together.

Additionally, HPE is expanding its CloudOps Software as a unified control plane for managing virtualization, containers, observability, automation and operations across multicloud and multivendor environments. The idea is to make complex cloud environments easier and cheaper for service providers to run by managing everything through a single platform instead of multiple disconnected tools.

To support adoption, HPE Financial Services is launching a new 90/9 Advantage financing program, which offers deferred payments followed by low monthly lease options. The program covers HPE’s portfolio across networking, compute, storage and software.

Final thoughts

As an industry watcher, I was curious as to how quickly or slowly HPE and Juniper would come together. When HPE acquired Aruba Networks years ago, it left that business unit alone as to not disrupt it and I thought HPE might take a similar approach with Juniper. That does not seem to be the case as the joint company has built out an aggressive roadmap of products that brings the best of HPE and Juniper together with former Juniper CEO Rami Rahim running the entire networking business.

In my discussions with HPE management, though the company is moving fast, it is also acutely aware of the product loyalty that HPE and Juniper had, and that’s part of the design principals where any new products do not require a “rip and replace.” Both sets of customers can benefit from cross-engineering but should not ever have to do anything that disrupts their businesses.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE.

Photo: Zeus Kerravala