JAKARTA - Huawei introduced the Atlas 950 SuperPoD, a computing system that connects 1,024 Ascend chips to work as a single unit. The device is claimed to be the largest scale AI supernode in the industry.
China Daily, quoted on Saturday, July 18, said Huawei showcased the Atlas 950 SuperPoD at the 2026 World Artificial Intelligence Conference in Shanghai.
Huawei is preparing the system to train very large-scale AI models, support data centers, and handle many requests simultaneously.
The process when an already trained AI model produces an answer or prediction is called inference.
According to Huawei, the Atlas 950 SuperPoD combines a large number of chips with a chip-to-chip interconnection technology that is claimed to be a breakthrough. The system is said to be able to boost computing capabilities for model training with trillions of parameters.
The company also associates the device with the era of agentic AI. The term refers to AI that can plan and execute multiple steps to complete a task with little human intervention.
The Atlas 950 SuperPoD was previously only discussed in a technical presentation. Now, Huawei states that the device is ready to provide commercial-grade computing power on an industrial scale.
Huawei also explained the development of the Ascend ecosystem. According to the company, the previous generation Ascend 384 SuperPoD has been used in more than 750 commercial projects.
Its uses include internet, telecommunications, finance, education, health, transportation, and manufacturing services.
Huawei said it has worked with more than 3,000 industry partners to develop more than 7,000 specific solutions. These solutions are used by more than 2,000 major customers from government and corporate sectors.
The company is not only highlighting hardware. Huawei is also strengthening its open source software ecosystem to make AI development easier.
At the end of last year, Huawei fully opened the source code of the CANN heterogeneous computing architecture and the MindSeries basic software series.
Heterogeneous computing combines multiple types of processors so that each can handle different jobs more efficiently.
According to Huawei, the CANN open source community has produced 67 projects with more than 12.44 million lines of code. The ecosystem also attracts more than 3,500 active developers every month.
In this year's WAIC, Huawei showcased more than 20 case studies covering more than 60 real business uses.
The company says the combination of hardware and software is driving AI from the stage of technological breakthroughs to large-scale and profitable commercial use in various sectors.
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)
