As AI inference moves from centralized clusters to regional and far-edge locations, the WAN becomes the AI WAN fabric binding distributed inference, centralized training, security, and operations into a single, coherent platform. New AI-native applications must deliver predictable latency, handle high concurrency, and improve cost per token at scale, which is accelerating the shift to more distributed AI infrastructure. Many inference requests now depend on the WAN to deliver deterministic performance, enforce policy, and coordinate activity across domains in real time.

In the AI era, intelligence doesn’t live in one place, and neither can the network that supports it. A single AI interaction may traverse multiple edge clusters, cross metro and regional networks, and coordinate continuously with centralized AI environments. In that environment, routing decisions directly shape AI performance, reliability, and user experience. This is the role of AI grid: a WAN network fabric that connects AI factories, regional hubs, and edge sites into one orchestrated intelligence platform.

Instead of moving data to AI, the AI WAN fabric moves AI closer to the data.

HPE and NVIDIA are partnering to accelerate the AI Grid—an end-to-end architecture where WAN routing, security, and automation are integral to how AI is deployed, scaled, and operated. Together, we’re building a fabric that connects NVIDIA AI Factories with consistent policy, predictable performance, and day-two operability designed in from the start. The AI WAN fabric provides connections from Data Centers to AI Factories where generative, agentic, and physical AI is conducted, bringing intelligence closer to users, devices, and sensors. Instead of moving data to AI, the AI WAN fabric moves AI closer to the data. A critical capability of the AI WAN fabric is workload- and resource-aware orchestration that enables routing tasks to the right place on the grid.

HPE today announced a new collaboration with NVIDIA to advance the AI Grid as an end-to-end architecture to securely connect AI factories and distributed inference clusters across regional and far-edge sites. As a significant expansion, the Grid Solution is designed to enable thousands of distributed inference sites to be efficiently deployed and operated as a single AI system. Leading telco service providers TELUS and CityFibre are interested in exploring HPE’s AI Grid Solution to expand the services they provide to customers.

The opportunity: inference is moving outward—and telcos can monetize it without multiplying complexityFor telco service providers, this shift creates an opportunity to deliver AI-powered services such as network optimization, automated operations, intelligent customer experiences, and private enterprise edge AI.

This means that existing facilities with space, power, and connectivity can be repurposed as AI inference edge, creating distributed AI capacity closer to users and enterprise data.

And the demand signal is already loud. In our Omdia study, 84% of large enterprise AI adopters use distributed AI, and service providers are already building regional edge footprints and routing workloads to the closest sites to preserve low-latency experiences. Customers aren’t just asking for “fast”—they’re asking for predictable and ultra-reliable.

At that point, the network stops being “transport.” It becomes the AI WAN fabric—the connective tissue across edge sites, the backbone, and centralized AI environments.

Now add agentic AI to the mix. A single inference request may coordinate across multiple edge clusters, with constant back-and-forth between centralized AI data centers and distributed edges. At that point, the network stops being “transport.” It becomes the AI WAN fabric—the connective tissue across edge sites, the backbone, and centralized AI environments. And as access to intelligence becomes as important as creating it, telcos can evolve from connectivity providers into AI service providers, selling intelligence alongside connectivity.

Introducing the HPE AI Grid: An End-to-End Solution, Serving as the Unified AI WAN Fabric for NVIDIA AI Factories

The HPE AI Grid Solution

Inside AI factories, NVIDIA Spectrum-X Ethernet fabrics provide optimized networking for GPU clusters. The HPE AI Grid extends that environment outward, using HPE Networks Juniper PTX and MX platforms to connect AI factories, regional hubs, and distributed inference locations across the WAN.

HPE ProLiant Servers — edge inference compute built for distributed AI deployment. When AI inference moves to the edge, the challenge is not only GPU performance. It is deploying, securing, and operating inference services across sites with different footprints, constraints, and operational models—without adding complexity. HPE ProLiant Compute servers, featuring NVIDIA RTX PRO ™ 6000 Blackwell Server Edition and NVIDIA RTX PRO ™ 4500 Blackwell Server Edition GPUs, NVIDIA BlueField DPUs, and NVIDIA Spectrum-X Ethernet Fabrics and NVIDIA Connect-X NICs, are built for that reality.

As part of the HPE AI Grid Solution, HPE ProLiant supports NVIDIA RTX Servers across configuration sizes, enabling a consistent architecture from smaller edge nodes to larger distributed inference deployments. The result is edge inference infrastructure designed for faster deployment, stronger operational consistency, and secure scale-out across real-world environments.

HPE Networks Juniper PTX Family — long-haul throughput built for AI traffic.  As customers expand their NVIDIA AI Factories, leveraging every bit of data and every watt of energy, you don’t just see more bandwidth demand. You see elephant flows, bursty patterns, and workloads that stress networks built for yesterday’s WAN traffic patterns. The PTX Series are built for high-density interconnect and large-scale WAN transport, with deep buffers and a full, industry-leading, proven IP/MPLS routing stack needed to carry those flows cleanly across long distances.

HPE Juniper PTX platforms deliver the throughput needed to scale AI Grid connectivity across metros and support 800GE, including coherent 800G ZR+ optics, for dense interconnect expansion. Importantly, they enable coherent optics deployment with no trade-off in port density or performance, while helping reduce operational overhead and improve TCO through a simpler, more scalable interconnect model. They also provide line-rate MACsec up to 800GE, delivering built-in security for high-speed links.

HPE Networks Juniper MX Family — the proven telco edge and multi-cloud on-ramp built for AI scale. When AI adoption accelerates, the edge is not just “more connections.” It is more tenants, more policies, and more clouds—because most enterprises are running AI services across multiple cloud environments while extending inference closer to users. This is where HPE has a clear advantage: the HPE Juniper MX Family has a deep, proven history as the telco edge and as the cloud on-ramp, including VPC edge deployments, making it a natural foundation for the AI Grid customer edge.

MX delivers the logical scale for multi-tenancy, segmentation, tunneling, and policy driven connectivity—so operators can aggregate and isolate AI traffic cleanly across large customer populations and high tunnel counts. It also serves as the secure peering edge and multi-cloud gateway that ties enterprise sites to regional hubs and cloud AI services, with line rate MACsec and inline IPsec for built-in link and traffic protection. And with traffic engineering and flexible matching, operators can apply workload aware steering, so connectivity stays aligned as inference placement shifts across clouds, regions, and edge locations.

HPE Networks SRX Series — consistent security at the edge.  Distributed inference collapses if security becomes bespoke per site. The SRX4700 provides a consistent enforcement point for secure peering, segmentation, and policy-aligned connectivity—so operators can scale edge inference without creating a new security architecture every time they light up a location.

AI Grid Operations — unified lifecycle management built for distributed AI at scale. As AI expands from NVIDIA AI Factories to distributed edge environments, the challenge is operating the entire AI Grid seamlessly across sites, domains, and teams while maintaining security, consistency, and speed. The HPE AI Grid Solution is designed for that reality, combining HPE networking controllers with NVIDIA orchestration to deliver a seamless operations model across the full environment.

Together, they provide full lifecycle management, from onboarding sites and establishing secure connectivity to ongoing policy updates, software upgrades, and continuous optimization. The result is a more efficient operational experience that helps teams deploy faster, maintain consistency across distributed locations, and run AI Grid infrastructure with greater control and efficiency.

This is the power of 3-2-1: three industry leaders—HPE, NVIDIA, and telcos building the AI Grid—two AI-scale realities—training and distributed inference—and one unified AI Grid solution built for the way intelligence actually moves. As AI-native apps demand predictable latency, high concurrency, and better cost per token, HPE delivers the full-stack foundation—compute, storage, networking, and security—aligned with NVIDIA AI Factory GPUs, Bluefield DPUs, Spectrum-X Ethernet and Connect-X NICs, to connect AI factories, regional hubs, and edge sites into one orchestrated platform. From site-to-site interconnect to client-to-site access, the HPE AI Grid solution unifies connectivity, security, and operations so distributed AI behaves like one system—and the WAN operates as the operational fabric for AI at scale.