AI

Inside Huawei’s plan to make thousands of AI chips think like one computer

Imagine the connection of thousands of strong artificial intelligence chips scattered in dozens of server cabinets and make them work together as if it were one huge computer. This is exactly what Huawei at Huawei Connect 2025 has shown, as the company revealed a breakthrough in the infrastructure of Amnesty International that can reshape how the world build and measure artificial intelligence systems.

Instead of the traditional methods in which individual servers operate somewhat independently, the new SuperPod technology from Huawei creates what the company’s executive officials describe as a single logical machine made of thousands of separate processing units, allowing them, or “learning, thinking and mind”.

The effects of impressive technical specifications, which are a shift in how to organize, evaluate and spread artificial computing power in industries.

Technical Corporation: Unifybus 2.0

At the core of Huawei’s infrastructure approach is Unifiedbus (UB). “Huawei has developed the leading SuperPod structure based on our unified interconnected protocol.

Technical specifications reveal the scope of this achievement. The UNIFIEDBUS protocol addresses two challenges historically, with a large -scale computing of Amnesty International: long -term reliability and frequency range. Traditional copper connections provide a highly frequency range but only at short distances, usually binds tanks.

Supports optical cables the longer extent, but they suffer from reliability problems that become more problematic, the greater the distance and size. Erik Show, Vice Chairman of the Huawei Board of Directors and Chairman of the Rurani Board of Directors, said that solving these basic challenges of communication is necessary for the company’s infrastructure strategy.

Xu detailed the OSI model solution solutions: “We have built reliability in each layer of our interconnected protocol, from the physical layer link and data relationship, up to the network and transmission layers.

SuperPOD: scale and performance

ATLAS 950 SuperPod represents the main implementation of this structure, and it consists of up to 8,192 of ASCEND 950DT chips in the composition of its description Xu as “8 EFLOPs in FP8 and 16 EFLOPS in FP4.

Specifications are more than gradual improvements. Atlas 950 SuperPod occupies 160 wardrobes in 1000 meters2With 128 account cabinets and 32 COMMS wardrobe linked to visual interconnection. System memory capacity reaches 1,152 TB and maintains what Huawei claims is 2.1-microsecond cumin in the entire system.

Later on the production pipeline, ATLAS 960 SuperPod, which was appointed to merge 15,488 ascend 960 chips in 220 cabinets covering 2200 meters2. Shaw said that it will serve “30 EFLOPs in FP8 and 60 EFLOPS in FP4, and comes with 4,460 TB of memory and 34 PB/s interconnected frequency range.”

Beyond artificial intelligence: Computing applications for general purposes

The SuperPod concept extends beyond the burdens of artificial intelligence work to computing general purposes through Taishan 950 SuperPod. Built on Kunpeng 950 processors, this system deals with the challenges of the Foundation in replacing old central computers and medium -range computers.

Xu put this as especially suitable for the financing sector, as “Taishan 950 SuperPod, along with the distributor Gaussdb, can be an ideal alternative, replaced by one-time and forever-central computers, medium-range computers, and Oracle database servers.”

Open architecture strategy

Perhaps the most important thing for the broader infrastructure market than artificial intelligence, Huawei has announced the issuance of UNIFIEDBUS technical specifications as open standards. The decision reflects strategic locations and practical restrictions.

Shaw acknowledged that “the Chinese mainland will fail to hold the semiconductor manufacturing process for a relatively long period” and stressed that “the sustainable computing power can only be achieved with a practical operation.”

“We are committed to our open and open -minded approach, which will help more partners develop their industry -based super solutions.

The company will open the ingredients and programs open source, with devices including NPU units, air -refrigerated blade servers, AI cards, central processing unit panels and CASCAD cards. For programs, Huawei adhered to the entire Cann Open Open Open Canning program, Mind Series applications, and OpenPangu Foundation models by December 31, 2025.

Spreading the market and the impact of the ecological system

The realization of the real world provides checking these technical claims. More than 300 SuperPod Atlas 900 A3 units were shipped in 2025, which was published for more than 20 customers from multiple sectors, including Internet, finance, youth, electricity and manufacturing sectors.

The effects of the development of infrastructure from Chinese artificial intelligence are great. By creating an open environmental system on local technology, Huawei deals with the challenges of building an Amnesty International Competitive Infrastructure within the parameters that have been identified by manufacturing and availability semiconductors. Its approach allows a broader participation in the industry in developing Amnesty International’s infrastructure solutions without the need to reach the most advanced contract.

As for the global infrastructure market for Amnesty International, the Huawei architecture strategy provides an alternative to the integrated and integrated software and software integrated and integrated combination of integrated and integrated software between Western competitors. Whether the ecological system suggested by Huawei can achieve comparable performance and maintaining commercial feasibility remains widely.

In the end, SuperPod’s structure is more than just a gradual progress of computing artificial intelligence. HUAWEI suggests how to connect, manage and expand computer resources. The open version of its specifications and elements will test whether cooperative development can accelerate innovation in Amnesty International’s infrastructure in an ecosystem of partners. This has the ability to reshape the competitive dynamics in the infrastructure market of Amnesty International.

See also: Huawei is committed to training 30,000 artificial intelligence professionals with the expansion of the ecosystem of local technology

Do you want to learn more about artificial intelligence and large data from industry leaders? Check AI and Big Data Expo, which is held in Amsterdam, California, and London. The comprehensive event is part of Techex and is determined with other leading technological events, click here for more information.

AI News is supported by TechForge Media. Explore other web events and seminars here.

Don’t miss more hot News like this! Click here to discover the latest in AI news!

2025-09-25 09:23:00

Related Articles

Back to top button