Customer Success Story
High-performance, 800GbE data centers for AI-as-a-service
AI pioneer ionstream.ai simplifies AI deployments for enterprises, offering cloud-hosted GPU-as-a-Service (GPUaaS) and scalable AI infrastructure for on-demand AI workloads. With demand outpacing its existing infrastructure, ionstream.ai launched an extensive competitive evaluation process, selecting Juniper products and technologies for its next-generation AI data centers.
TCO
Open ecosystem
Competitive and accessible networking
800GbE
High performance
First to market
AI
Scalable
On-demand capacity
ionstream.ai’s on-demand, managed AI services provide flexible scalability, immediate time-to-market, and affordable consumption models for enterprises. But the high and rapidly changing demand for AI infrastructure makes it difficult to accurately forecast infrastructure scale and performance needs.
ionstream.ai’s challenge to accommodate changing customer need for inferencing and training necessitated a flexible and reliable high speed architecture. With their offering of both NVIDIA and AMD GPUs, ionstream required an open, multivendor networking fabric with superior availability, high performance and scale, and seamless support for its technical operations team.
“In the rapidly evolving field of AI and ML, selecting the right GPU infrastructure and deciding whether to build, buy, or rent it can significantly impact a company’s success,” said Jeff Hinkle, CEO at ionstream. “We chose Juniper’s Ethernet solution to reduce flapping often seen in InfiniBand networks as well as the flexibility of Ethernet to reconfigure a dynamic server farm to changing customers in a bare metal environment. The speed difference is negligible and the ease of use of Juniper Junos OS operating system compared to other less mainstream operating systems was critical in our decision for our network engineers to support the network.”
Hinkle continued: “Juniper provides us with easy-to-deploy, scalable, and AI-optimized networking so that we can deliver accessible, first-class GPU-as-a-Service and AI transformation for our customers, however they choose to consume it.”
The Juniper AI data center solution is foundational to all ionstream.ai’s managed AI service offerings. As a key element of Mist™, Juniper’s AI-native networking platform, the AI data center solution is built on a spine-leaf architecture using AI-optimized 800GbE Juniper Networks QFX5240 switches. It enhances GPU performance with RDMA over Converged Ethernet version 2 (ROCEv2) and uses Ethernet VPN (EVPN) to secure multi-tenant environments, making it a reliable and efficient choice for AI-driven networks.
Juniper’s network for AI workloads solution
Juniper delivers AI cloud infrastructure solutions that accelerate time-to-market and maximize ROI for service providers, neoclouds, and cloud providers. Solutions include secure, multi-tenant AI cloud infrastructures that simplify GPU virtualization to eliminate idle time and maximize efficiency.
Key features include:
- AI-native operations: Accelarate the delivery of AI cloud services
- Zero Trust security: Safeguard customer models and data
- Automated optimization: Enhanced AI service performance
- Intent-based automation: Simplified design, deployment, and operations
- GPU-as-a-Service: Scalable and secure networks for AI services
The Juniper QFX5240 switch is designed for AI data centers, offering 800GbE ports and advanced networking features. It supports high-density, high-speed connectivity and automation through Juniper Apstra Data Center Director software, ensuring efficient AI workload management and reduced job completion times (JCT).
Leading the industry in networks for AI workloads
Choosing the right GPU infrastructure is critical for success in AI. After evaluating networking options, ionstream.ai selected Juniper’s AI-optimized Ethernet data center solution for its openness, performance, and scalability.
Open Ethernet ecosystem
An open AI infrastructure seamlessly integrated multivendor GPUs and storage for flexibility, availability, time-to-market, and a broad technical support ecosystem.
High-performance AI data center networking
The high-capacity, 800GbE AI data center networking delivered high-performance and low-latency training for the lowest job completion time, inferencing with the fastest time-to-first-token (TFFT), and advanced software features for congestion avoidance and control.
Scalable networking for managed AI workloads
The extensible and flexible Ethernet data center fabric with EVPN/VXLAN isolation separates and secures multi-tenant AI workloads for scalable AI managed services.
Published August 2025