Huawei Introduces Atlas 950 SuperPoD Supercomputer for AI

17:56 / 17.07.2026·26·Technology

Chinese tech giant Huawei has officially announced its next-generation Atlas 950 SuperPoD supercomputer, designed for training and deploying AI systems. This system is one of the largest and most powerful computing platforms in the industry, serving to fundamentally accelerate the creation of Large Language Models (LLM). The new device is expected to outperform its main global competitors in both power and memory capacity. This is reported by Ixbt.com reports .

SuperPoD technology allows for the integration of thousands of Neural Processing Units (NPU) into a single logical block. This ensures high efficiency in training AI models. According to ixbt.com, the Atlas 950 system features terabyte-level throughput and an ultra-low latency of just 3 microseconds. These metrics enable lightning-fast data stream processing without interruptions.

Technical Capabilities and Competition with NVIDIA

The Atlas 950 SuperPoD incorporates 1024 Ascend chips and can combine up to 8192 neural blocks in total. Such immense power is sufficient for training and deploying multi-stage AI models with trillions of parameters. Huawei engineers utilized the proprietary Lingqu protocol and a modernized SuperPoD architecture for this cluster.

The most interesting aspect is that Huawei is comparing its product to NVIDIA's NVL144 server rack based on the Rubin Ultra architecture. According to the Chinese company's statement, the Atlas 950 SuperPoD is 6.7 times more powerful in computing performance and 15 times superior in memory capacity compared to the NVIDIA NVL144 system. These figures demonstrate the seriousness of Huawei's ambitions in the global semiconductor market.

The supercomputer's total memory capacity is 256 TB. In terms of computing power, the Atlas 950 demonstrates the following performance:

1 EFLOPS computing power in FP8 format;
2 EFLOPS computing power in FP4 format;
256 TB of unified high-speed memory;
Minimal latency and maximum throughput via the Lingqu protocol.

This news is also relevant for countries developing a digital economy. Such high-performance systems could become a core pillar of infrastructure in the process of implementing AI technologies and digitizing public services. Huawei's achievement reduces monopoly in the AI sector and paves the way for new technological solutions in the market.

In conclusion, the Atlas 950 SuperPoD stands out not only for its technical specifications but also for its software. By creating its own closed ecosystem, Huawei continues to produce high-tech equipment despite external restrictions. This indicates that the balance of power in the AI race may shift in the coming years.