Particle.news
Download on the App Store
18 ARTICLES
·
3h ago

Huawei Details Mega-Scale AI Clusters, In‑House HBM and a Multi‑Year Ascend Roadmap

Reported Chinese curbs on Nvidia‑focused chips elevate demand for Huawei’s domestically built Ascend systems.

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 introduced Atlas 950 and 960 SuperPoDs that support 8,192 and 15,488 Ascend processors, with SuperClusters designed to scale from 500,000 up to 1,000,000 chips.
  • The company published its Ascend roadmap: 950PR in Q1 2026, 950DT in Q4 2026, 960 in 2027, and 970 in 2028, integrating proprietary HBM (HiBL 1.0 at 128 GB/1.6 TB/s and HiZQ 2.0 at 144 GB/4 TB/s).
  • Huawei remains barred from TSMC under U.S. controls, and analysts say SMIC is the likely manufacturing partner, leaving real‑world yields, efficiency and software migration as key uncertainties.
  • Huawei positions Atlas 950/960 as the world’s most powerful cluster designs and cites peak figures up to 524 FP8 exaFLOPS for training and 1 FP4 zettaFLOPS for inference for the 950 SuperCluster, which are company claims not yet validated.
  • Financial Times reporting says Chinese regulators told tech firms to halt tests and orders of certain Nvidia China‑specific chips, increasing strategic pressure on Nvidia and bolstering Huawei’s pitch.