DE-CIX unveils global AI exchange to boost high-speed inference

DE-CIX has launched a specialised internet exchange for AI applications aimed at supporting high performance inference across its global network.

The first phase of the AI Internet Exchange (AI-IX) programme has been completed, with more than 50 AI-related entities - ranging from providers of AI Inference as a Service and GPU as a Service to a range of cloud service companies - now integrated into the DE-CIX platform. According to the company, these providers join a network of over 160 cloud on-ramps and benefit from a proprietary multi-AI routing technology.

The firm asserts that this infrastructure enables resilient, low-latency, and secure connectivity specifically tailored to meet the demands of real-time AI-powered applications. The newly introduced product is available at all DE-CIX locations worldwide in addition to its Frankfurt exchange.

Second rollout phase

DE-CIX intends to follow with a second phase of development that will equip its AI exchanges to be Ultra-Ethernet ready, a capability designed to support AI model training in distributed environments outside of traditional centralised data centres. The company anticipates that, after this phase, it will be the first operator to deliver an AI-IX capable of servicing both the training and inference aspects of AI operation.

Phase two is intended to support geographically distributed AI activity, which is becoming increasingly important as AI model training shifts towards disaggregated, metropolitan computing environments.

Industry use-cases

The growing proliferation of multimodal AI agents and applications such as self-driving vehicles, robotics, and digital assistants is driving demand for highly reliable, secure, and low-latency connectivity.

"With various predictions made for millions - even hundreds of millions - of AI agents in the coming years, the need for interconnection services to support their operations is growing massively," explains Ivo Ivanov, CEO of DE-CIX.

AI operations are commonly divided into two distinct phases: training and inference. Ivanov describes AI inference as the stage where agents are deployed to perform real-time analysis, interact with users, or manage processes. He elaborates on the requirements for reliable AI operation, highlighting what he refers to as the "Digital Triangle of AI Inference Interconnection." Ivanov identifies three core elements: the rapidly multiplying number of AI agents, a diverse array of devices and applications powered by AI, and the deployment of advanced transmission technologies such as fibre optics, 5G Advanced, and LEO satellites. He states that direct interconnection or "peering" between these entities is critical.

"For enterprises to be able to exploit the advantages of AI, the digital lifeline consists of three elements, what I like to call the Digital Triangle of AI Inference Interconnection."

The direct linking of these components is described as essential for delivering the required reliability, security, and performance.

"This is the core benefit of the DE-CIX AI-IX, which uses the unique DE-CIX AI router to enable seamless multi-agent inference for today's complex use-cases and tomorrow's innovation in all industry segments."

Infrastructure for training

As AI workloads become increasingly complex, the infrastructure for training models is changing. The introduction of Ultra Ethernet, a protocol designed for managing AI data traffic, is expected to replace existing protocols like InfiniBand, offering a more cost-effective and scalable solution for enterprise customers.

"Until now, huge, centralized data centers have been needed to quickly process AI computing loads on parallel clusters," explains Dr. Thomas King, CTO of DE-CIX. "Ultra Ethernet is driving the trend towards disaggregated computing, enabling AI training to be carried out in a geographically distributed manner within a metropolitan area. This will revolutionize the infrastructure for AI training and offer companies new alternatives for designing resilient and more cost-effective private AI infrastructure."

DE-CIX states that its hardware is already equipped to support the evolving Ethernet standard and will provide access to customers as network vendors make the relevant protocol available.

Global operations

DE-CIX operates exchanges at 60 locations across Europe, Africa, the Americas, the Middle East, and Asia. Its Frankfurt exchange connects nearly 1,100 networks with an annual data volume of almost 45 exabytes. The company now provides interconnection services to more than 4,000 network operators, including Internet service providers, carriers, content services, and enterprise networks spanning over 100 countries.