Back to today’s briefing

Story perspectives

Battle of the Titans: Huawei, Nvidia, AMD Unveil AI Powerhouses

10/6/2025

29 3 Full Breakdown

Summary

Story summary

Huawei's Atlas 950 SuperPoD targets 8 exaFLOPS FP8 and 16 exaFLOPS FP16 with 8,192 NPUs for training.
Nvidia's DGX SuperPOD, already available, uses 160 A100 GPUs for mixed precision, prioritizing reliability.
AMD's MegaPod targets high performance with 256 MI500 GPUs and advanced networking, set for 2027.
The competition shows differing philosophies: Huawei’s brute-force bandwidth, Nvidia’s ecosystem maturity, AMD’s scalable disruption.

History

Historical Context

The Tianhe-1 supercomputer, developed by China's National University of Defense Technology, became the first to exceed 1 petaflop in 2010, showcasing China's advancements in high-performance computing. This milestone initiated a global race, leading to the emergence of competitors like Huawei, Nvidia, and AMD, each striving for dominance in computing power, which now influences sectors such as artificial intelligence and scientific research.

1 of 2

Story summary

Huawei's Atlas 950 SuperPoD targets 8 exaFLOPS FP8 and 16 exaFLOPS FP16 with 8,192 NPUs for training.
Nvidia's DGX SuperPOD, already available, uses 160 A100 GPUs for mixed precision, prioritizing reliability.
AMD's MegaPod targets high performance with 256 MI500 GPUs and advanced networking, set for 2027.
The competition shows differing philosophies: Huawei’s brute-force bandwidth, Nvidia’s ecosystem maturity, AMD’s scalable disruption.

1 / 2