At MWC Barcelona 2026, Huawei introduced its latest SuperPoD portfolio, including the Atlas 950 SuperPoD, TaiShan 950 SuperPoD and a series of next-generation computing solutions for the global market.
The launch reflects Huawei’s commitment to open source innovation and open collaboration, aimed at building a resilient computing foundation and offering new choices for AI-driven industries worldwide.
SuperPoD Architecture Designed for Trillion-Parameter AI Models
As AI technologies evolve rapidly and foundation models scale to trillions of parameters, agentic AI is increasingly embedded into core production processes across industries. This transformation is driving demand for larger computing clusters, higher bandwidth and lower latency.
Traditional horizontal scaling approaches are struggling to keep pace. Expanding cluster size often leads to lower resource utilization and frequent interruptions during large-scale AI training tasks.
To address these challenges, Huawei has developed its innovative UnifiedBus interconnect technology for SuperPoD systems. The new “cluster + SuperPoD” architecture is purpose-built to support massive AI workloads while maintaining high efficiency and stability.
Atlas 950 SuperPoD Delivers Ultra-High Performance
At MWC Barcelona 2026, Huawei showcased its latest SuperPoD offerings, including:
Atlas 950 SuperPoD
Atlas 850E
Built on the UnifiedBus interconnect, these systems are designed for diverse AI training and inference scenarios.
The Atlas 950 SuperPoD can connect up to 8,192 NPUs through UnifiedBus, delivering ultra-high bandwidth, ultra-low latency and unified memory addressing. The architecture enables the entire cluster to operate as a single logical computer, supporting large-scale model training, reasoning and high-performance data processing.
This design improves cluster utilization and minimizes training disruptions, addressing key bottlenecks faced by hyperscale AI workloads.
TaiShan 950 SuperPoD Expands General-Purpose Computing
Huawei also unveiled TaiShan 950 SuperPoD, described as the industry’s first general-purpose computing SuperPoD. It is complemented by next-generation servers such as:
TaiShan 500
TaiShan 200
These platforms provide flexible computing capabilities across workloads ranging from high-intensity AI processing to standard enterprise applications. By covering a broad spectrum of computing needs, Huawei aims to deliver scalable and adaptable infrastructure for governments, enterprises and research institutions.
Open Source Strategy to Accelerate Ecosystem Growth
Huawei continues to promote open source and open systems to foster developer innovation and ecosystem expansion.
The company plays a leading role in advancing openEuler, which has rapidly grown into one of the world’s prominent open source operating system communities.
In addition, Huawei has fully open-sourced its CANN heterogeneous computing architecture. Through layered decoupling, developers gain open access to operator libraries, acceleration libraries, graph computing frameworks and programming language components.
CANN also supports mainstream open source AI communities and projects such as:
PyTorch
vLLM
By enabling compatibility and integration with widely used AI frameworks, Huawei lowers development barriers and enhances efficiency for AI engineers and researchers.
Building a Resilient Computing Future
As intelligence reshapes industries worldwide, computing infrastructure is becoming a critical strategic asset. Through its SuperPoD innovations, UnifiedBus architecture and open ecosystem strategy, Huawei is positioning itself to deliver a resilient, high-performance computing foundation for the AI era. At MWC Barcelona 2026, Huawei’s latest computing portfolio demonstrates its ambition to combine large-scale AI performance with open collaboration, offering global markets a new and competitive option in next-generation computing infrastructure.
