The Traffic Exchange Point operator, DE-CIX, has just announced that its new global interconnection ecosystem is ready to be integrated with artificial intelligence. For this purpose, in the implementation of Phase 1 of its AI-IX, efforts have been made to enable high-performance inference for AI agents.
After completing this initial phase, DE-CIX now connects more than 50 key networks for the AI ecosystem, including providers of Inference-as-a-Service, GPU-as-a-Service, and a wide range of cloud service companies (CSP). Currently, the platform features over 160 distributed cloud access points worldwide and relies on a proprietary, highly scalable cognitive routing technology.
Thanks to this, DE-CIX offers secure, low-latency, and highly resilient interconnections, designed to meet the real-time needs of AI. This service enhancement is not limited to the leading traffic exchange point in Europe, located in Frankfurt, but is available at all DE-CIX locations worldwide.
Phase 2 of the deployment, which will be carried out soon, will prepare DE-CIX’s AI Exchange Points for Ultra-Ethernet, capable of supporting geographically distributed AI training as AI model training begins to move outside centralized facilities and points. The deployment of both phases is underway globally. In this way, DE-CIX will become the first IX operator to offer an AI-IX that can support both AI training and inference.
Uninterrupted multi-AI inference for complex current use cases and future innovation
Just like with the launch of DE-CIX AI-IX, AI operations in general are divided into two phases: AI training and AI inference. Both phases benefit from AI peering, offering advantages such as cost reduction, enhanced security, increased performance, and decreased complexity. “With various predictions for millions, even hundreds of millions, of AI agents in the coming years, the need for interconnection services to support their operations is growing tremendously,” explains Ivo Ivanov, CEO of DE-CIX.
AI inference, where the AI agent’s use is applied to real-world situations to provide information, interaction, and assistance in real time, depends on highly resilient, secure, low-latency connectivity. “For companies to leverage AI benefits, the digital lifeline consists of three elements, which I call the digital inference interconnection triangle.” For Ivanov, these elements are, first, the millions of agents currently emerging, many of which are multimodal; second, AI-based devices/applications (cars, robots, processes, etc.); and third, modern transmission technologies (such as fiber optics, 5G Advanced, and LEO satellite networks, etc.). These three elements must be reliably, securely, and with extremely high performance interconnected through direct peering.
“This is the main advantage of the DE-CIX AI-IX, which uses the exclusive DE-CIX AI router to enable seamless multi-agent inference for current complex use cases and future innovation across all industry segments.”
Revolutionizing AI training with Ultra Ethernet
In the second phase of the DE-CIX AI-IX deployment, focus is on AI model training. With the advent of Ultra Ethernet, a new protocol for routing relevant data traffic for AI, the design of infrastructure that meets the high requirements to facilitate AI training is evolving. “Until now, enormous centralized data centers were required to quickly process AI computational loads in parallel clusters,” explains Dr. Thomas King, CTO of DE-CIX. “Ultra Ethernet is driving the trend toward disaggregated computing, allowing AI training to be conducted across geographically distributed areas within the same metropolitan region. This will revolutionize AI training infrastructure and offer companies new options to design resilient, cost-effective private AI infrastructure.”
The next version of the Ultra Ethernet standard will be implemented as network hardware providers enable this functionality, offering a significantly more cost-effective solution compared to the current AI network protocol, InfiniBand. DE-CIX hardware is already compatible with this new Ultra Ethernet standard and will be deployed as soon as it becomes available.