Huawei debuts its Atlas 950 AI SuperPoD at MWC 2026, taking the AI data center fight to Nvidia and AMD (www.techradar.com)

At MWC 2026, Huawei launched its Atlas 950 and TaiShan 950 SuperPoDs, marking a strategic entry into the AI infrastructure landscape aimed at competing directly with Nvidia and AMD. The Atlas 950 system integrates up to 8,192 Ascend NPUs into a single logical AI system using Huawei’s UnifiedBus interconnect, mitigating communication delays during large training scenarios. Capable of delivering up to 8 exaflops of FP8 performance and 16 exaflops for lower precision, it spans approximately 1,000 square meters with over a petabyte of memory, designed explicitly for massive model training and high-throughput inference tasks. This launch is significant as it positions Huawei as a formidable player in the AI data center market, challenging Nvidia’s dominance with its established CUDA software platform. With Atlas 950 using Huawei's Ascend chips and the open-source CANN architecture compatible with frameworks like PyTorch and Triton, developers gain an alternative route for large-scale AI workloads, bypassing Nvidia’s ecosystem. The TaiShan 950 extends this approach to general-purpose computing, further broadening Huawei's portfolio and its potential to disrupt the competitive landscape of AI infrastructure.