Summary:
1. Huawei demonstrated a breakthrough in AI infrastructure architecture at HUAWEI CONNECT 2025, showcasing SuperPoD technology that allows thousands of AI chips to work together as a single, massive computer.
2. The UnifiedBus 2.0 protocol addresses challenges in long-range communications and bandwidth-latency, enabling Huawei to create a seamless logical machine for AI computing.
3. The SuperPoD architecture, including the Atlas 950 and upcoming Atlas 960, offers impressive scale and performance for AI workloads, with implications for general-purpose computing applications as well.
Article:
At HUAWEI CONNECT 2025, Huawei unveiled a groundbreaking advancement in AI infrastructure architecture that has the potential to revolutionize how artificial intelligence systems are built and scaled. The company showcased their SuperPoD technology, which allows thousands of powerful AI chips to work together as if they were a single, massive computer. This innovative approach creates a seamless, logical machine that can “learn, think, and reason as one,” representing a significant shift in how AI computing power is organized and deployed across industries.
Central to Huawei’s infrastructure strategy is the UnifiedBus 2.0 protocol, which addresses key challenges in long-range communications and bandwidth-latency. By deeply interconnecting physical servers using UnifiedBus, Huawei has overcome historical limitations in large-scale AI computing. This protocol ensures reliability and high bandwidth over long distances, enabling the creation of a single logical server from thousands of separate processing units.
The SuperPoD architecture, exemplified by the Atlas 950 and forthcoming Atlas 960, offers remarkable scale and performance for AI workloads. The Atlas 950 SuperPoD boasts 8,192 Ascend 950DT chips delivering impressive EFLOPS and interconnect bandwidth, setting new standards for AI computing. Additionally, Huawei’s TaiShan 950 SuperPoD, built on Kunpeng 950 processors, extends the SuperPoD concept beyond AI workloads into general-purpose computing applications, particularly relevant for the finance sector.
Huawei’s open architecture strategy, exemplified by the release of UnifiedBus 2.0 technical specifications as open standards, aims to foster collaboration and innovation in the AI infrastructure market. By open-sourcing hardware and software components, Huawei invites industry partners to develop their own SuperPoD solutions, accelerating developer innovation and ecosystem growth. This approach not only addresses constraints in semiconductor manufacturing but also introduces an alternative to the proprietary hardware and software approach dominant among Western competitors.
In real-world deployment, over 300 Atlas 900 A3 SuperPoD units have already been shipped in 2025, demonstrating the practicality and viability of Huawei’s AI infrastructure solutions. By creating an open ecosystem around domestic technology, Huawei is paving the way for competitive AI infrastructure development in China and potentially reshaping global dynamics in the AI infrastructure market. The SuperPoD architecture represents a fundamental shift in how massive computational resources are connected, managed, and scaled, with the potential to drive innovation and collaboration in the AI industry.
