As artificial intelligence (AI) evolves from being centred around training to inference, HPE has announced a range of compute and network products powered by Nvidia graphics processing units (GPUs) that it says will redefine how AI is delivered by moving intelligence to where data and users are based, and making the network a dependable fabric to support real-time experiences.
One of the standouts in the range is HPE AI Grid, designed to transform distributed AI factories into an intelligent AI infrastructure powered by Nvidia. These are said to offer a “critical” application for AI services and use cases that rely on low-latency, real-time connectivity, including retail personalisation and predictive maintenance in industries such as manufacturing, localised edge inference in healthcare and carrier-grade AI services.
HPE believes that AI‑native applications require predictable, low‑latency, distributed infrastructure. The HPE AI Grid service, part of the Nvidia AI Computing by HPE portfolio, is claimed to deliver predictable, “ultra‑low” latency performance at scale for real‑time AI services, zero‑touch provisioning and automated security with integrated orchestration.
The end-to-end system is built on an Nvidia reference architecture, and is designed to securely connect AI factories and distributed inference clusters across regional and far‑edge sites. It can facilitate the deployment and operation of as many as thousands of distributed inference sites, turning AI installations into a single intelligent system.
Looking at the intended use cases, HPE envisioned implementation from retail personalisation and predictive maintenance to edge healthcare and carrier‑grade AI services, which all require predictable, ultra‑low latency connectivity. The AI Grid is also designed to allow operators to convert existing sites with power and connectivity into RAN‑ready AI grids, enabling distributed inference and new services at scale.
As well as aligning with the AI Grid reference architecture to provide a unified hardware and software stack for service providers, HPE AI Grid is said to be differentiated by the ability to offer full-stack AI servers and AI networks.
The HPE AI Grid includes HPE Juniper’s multicloud routing and coherent optics for predictable long-haul and metro connectivity; cloud-native and multi-tenant security; firewalls; WAN automation; and orchestration to deliver zero-touch deployment and lifecycle operations. It also contains ProLiant Compute edge and rack servers with Nvidia accelerated computing, including Nvidia RTX PRO 6000 Blackwell GPUs, as well as Nvidia BlueField DPUs, Spectrum-X Ethernet switches, Connect-X SuperNICs, and AI blueprints for rapid AI inference.
Neil McRae, chief technology information officer at CityFibre, said: “Our customers increasingly expect millisecond responsiveness, low-latency connectivity and comprehensive security to support their applications and services.
“We’re exploring how AI Grid from HPE, based on Nvidia’s reference architecture, could support distributed AI inferencing, and bring intelligence closer to users and data. By leveraging our fibre network assets, we see potential to combine high-performance connectivity with intelligent services for customers.”
Chris Penrose, global vice-president for telco at Nvidia, said: “An AI Grid unifies geographically distributed AI clusters to place AI workloads where they run best – balancing performance, cost and latency across AI factories, regional sites and the edge.
“Together with HPE, we’re bringing that vision to life by combining Nvidia’s accelerated computing and networking with HPE’s telco‑grade multi-cloud routing and edge infrastructure to create a single, intelligent fabric for distributed inference.”
As part of advancing its AI grid strategy, operator Comcast has begun AI field trials on its distributed network for real-time edge AI inferencing to unlock faster, more responsive experiences for the next wave of AI applications. The initial trials addressed several use cases, including leveraging HPE ProLiant servers running small language models from Personal AI, part of HPE’s Unleash AI partner programme, on Nvidia GPUs.
