DE-CIX has launched a specialized internet exchange for AI applications aimed at supporting high-performance inference across the global network.
The first phase of the AI Internet Exchange (AI-IX) program has been completed. AI-related entities are now integrated into the DE-CIX platform from over 50 AI-related entities, from AI inference providers as services, and GPUs as services to various cloud service companies. According to the company, these providers have joined over 160 cloud-on-ramp networks and benefit from their unique multi-AI routing technology.
The company claims that the infrastructure allows for resilient, low latency, and secure connectivity, specifically tailored to meet the demands of real-time AI-powered applications. Newly introduced products are available at all DE-CIX locations around the world, in addition to the Frankfurt Exchange.
Second Rollout Phase
DE-CIX will follow the second phase of development, which will feature AI replacements in Ultra-Ethernet Ready, designed to support AI model training in distributed environments other than traditional centralized data centers. The company expects to be the first operator to provide AI-IX after this phase to serve both the training and inference aspects of AI operations.
Phase 2 is intended to support geographically distributed AI activities. This has become increasingly important as AI model training shifts to a decomposed metropolitan computing environment.
Industry Use Cases
The rise and applications of multimodal AI agents, such as self-driving cars, robotics, and digital assistants, are driving the demand for reliable, secure, low latency connectivity.
“The various forecasts made for millions of AI agents and hundreds of millions more AI agents have significantly increased the need for interconnected services to support our businesses,” explains Ivo Ivanov, CEO of DE-CIX.
AI operations are generally divided into two different phases: training and inference. Ivanov describes AI inference as the stages that are deployed by agents to perform real-time analytics, interact with users, and manage processes. He elaborates on the requirements for reliable AI operations, highlighting what he calls the “digital triangle of AI inference interconnections.” Ivanov identifies three core elements: rapidly growing AI agents, a diverse array of AI-powered devices and applications, and the deployment of advanced transmission technologies such as fiber optics, 5G Advanced, and LEO satellites. He states that direct interconnections or “peering” between these entities are important.
“To help businesses take advantage of AI, a digital lifeline consists of three elements: what is called the digital triangle of AI inference interconnections.”
Direct linking of these components is described as essential to providing the reliability, security and performance you need.
“This is the central advantage of De-CIX AI-IX, which uses its proprietary De-CIX AI router to enable seamless multi-agent inference for today's complex use cases and innovation of tomorrow across all industry segments.”
Infrastructure for training
As AI workloads become more and more complex, the infrastructure of our training models is changing. The deployment of Ultra Ethernet, a protocol designed to manage AI data traffic, is expected to replace existing protocols such as Infiniband, providing a more cost-effective, scalable solution for enterprise customers.
“To date, we've had to quickly handle AI computational loads in parallel clusters, so we needed huge centralized data centers,” explains Dr. Thomas King, CTO at DE-CIX. “Ultra Ethernet drives the disassembled computing trend, allowing AI training to be carried out in a geographically distributed manner within the Metropolitan area. This will revolutionize the AI training infrastructure and provide businesses with new alternatives to design resilient and cost-effective AI infrastructure.”
DE-CIX says its hardware is already equipped to support evolving Ethernet standards and will provide access to customers as network vendors make relevant protocols available.
Global Operations
DE-CIX operates exchanges in 60 locations in Europe, Africa, America, the Middle East and Asia. The Frankfurt Exchange connects almost 1,100 networks and has an annual data volume of almost 45 exabytes. The company currently provides interconnect services to over 4,000 network operators, including internet service providers, carriers, content services and enterprise networks across more than 100 countries.
