Particle.news
Download on the App Store
17 ARTICLES
·
6h ago

Huawei Unveils Atlas 950/960 SuperPoDs, Touts World’s Most Powerful AI Clusters

The plan bets on massive clusters of homegrown chips to offset per‑chip shortfalls.

Image
Huawei has announced AI computing cluster for the first time.
SHANGHAI, CHINA - JUNE 29, 2020 - Huawei's largest flagship store in the world, Shanghai, China, June 29, 2020.- PHOTOGRAPH BY Costfoto / Barcroft Studios / Future Publishing (Photo credit should read Costfoto/Barcroft Media via Getty Images)

Overview

  • Huawei detailed Atlas 950 and 960 SuperPoDs and associated SuperClusters that link 8,192 and 15,488 Ascend processors per supernode and can scale to 500,000–1,000,000 chips, with the 950 targeted for Q4 2026 and the 960 for 2027.
  • A new Ascend roadmap sets the 950PR for Q1 2026 and 950DT for Q4 2026, followed by 960 in 2027 and 970 in 2028, and introduces in‑house high‑bandwidth memory—HiBL 1.0 at 128GB/1.6 TB/s and HiZQ 2.0 at 144GB/4 TB/s—integrated into upcoming chips.
  • Huawei says the systems will lead on scale and throughput, pursuing performance by using many more accelerators, a strategy reflected in SemiAnalysis findings that Huawei outperformed Nvidia in tests by deploying five times as many chips despite lower per‑chip performance.
  • The company did not disclose a foundry for the new Ascend parts; analysts point to SMIC as the likely partner and flag capacity, yield and per‑chip efficiency as key execution risks.
  • The announcements come as reports say Chinese regulators have urged tech giants to stop testing and buying certain Nvidia China‑focused GPUs, a move Nvidia CEO Jensen Huang described as disappointing.