Next-Gen AI Infrastructure Unveiled at MWC

Hollow-core fiber (HCF) guides light through an air-filled core, allowing data to travel up to 47% faster than in traditional solid glass fibers. This near-light-speed transmission significantly reduces latency, a critical bottleneck for distributed AI models and complex, multi-tool creative workflows. The technology also promises lower signal loss and the ability to handle higher power levels. YOFC's MWC presentation is part of its broader "AI-2030" strategy, positioning itself as a key infrastructure provider for AI's voracious data demands. The company has already been involved in deploying real-world test networks, including the world's first 800G hollow-core fiber transmission test with China Mobile, demonstrating its viability for large-scale, high-throughput applications. This move signals a push to mass manufacturing and commercialization. On the computing side, Huawei's Atlas 950 SuperPoD is a massive computing cluster, not a single chip. It integrates up to 8,192 of Huawei's Ascend 950DT processors, delivering up to 8 EFLOPS of AI computing power. This architecture is designed for training large-scale models and is part of a larger product family that will include even more powerful SuperClusters. The scale of the Atlas 950 SuperPoD is substantial, occupying 160 cabinets and requiring 1,000 square meters of space. It boasts 1,152 TB of memory and an interconnect bandwidth of 16.3 PB/s, which Huawei claims is significantly higher than competitors' planned systems. This massive bandwidth is crucial for keeping thousands of processors working in concert on a single task without bottlenecks. These hardware advancements directly address the performance ceilings encountered by creative professionals chaining together multiple AI tools. The ultra-low latency of HCF can make real-time collaboration between different AI services—like an image generator feeding a video synthesis tool—more fluid and responsive. For developers building these tools, massive computing clusters like the SuperPoD accelerate the training and refinement of the increasingly complex models that underpin creative AI applications. The broader trend at MWC 2026 points toward AI becoming a foundational component of the tech architecture, moving from a niche feature to the core of devices and networks. This shift emphasizes the need for robust underlying hardware, as the performance of on-device AI and edge computing hinges on the speed and efficiency of the data center and network infrastructure.