Huawei Ascend 950 · Ascend

1,000 teraflops of dense FP8 compute a chip

Shipped 17 Sep 2026; shipped by the first date promised. Epoch AI names it at 1 AI data center and no GPU cluster. Huawei across the site.

Embed

Ascend 950

<iframe src="https://jesseahawkins.com/chips/embed/accelerators/huawei-ascend-950/" title="Huawei Ascend 950: 1,000 teraflops of dense FP8 compute a chip" width="100%" height="520" style="border:0;max-width:100%" loading="lazy"></iframe>

The frame credits Huawei, Epoch AI, AI data centers and Epoch AI, Data on GPU clusters and links back to its page. Preview it.

Cite

This page

Jesse Hawkins, “Huawei Ascend 950: 1,000 teraflops of dense FP8 compute a chip”, jesseahawkins.com, data retrieved 23 Oct 2026. https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/. Data: Each vendor’s own datasheets and releases, quoted; Epoch AI, “AI data centers” and “Data on GPU clusters”, CC BY 4.0.

BibTeX

@misc{hawkins2026chipsacceleratorshuaweiascend950,
  author = {Hawkins, Jesse},
  title = {{Huawei Ascend 950: 1,000 teraflops of dense FP8 compute a chip}},
  year = {2026},
  howpublished = {\url{https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/}},
  note = {Data: Each vendor’s own datasheets and releases, quoted; Epoch AI, ``AI data centers'' and ``Data on GPU clusters'', CC BY 4.0, retrieved 2026-10-23}
}

The data · Epoch AI

Epoch AI, 'AI data centers'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/ai-data-centers' [online resource]. CC BY 4.0.

The clusters · Epoch AI

Epoch AI, 'Data on GPU clusters'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/gpu-clusters' [online resource]. CC BY 4.0.

CSV Post on X

Each figure in Huawei’s own words

one chip, dense compute; memory is high-bandwidth memory

FP8, dense: 1,000 TFLOPS They will be able to deliver 1 PFLOPS in FP8, MXFP8, and HiF8, and 2 PFLOPS in MXFP4. Keynote speech by Eric Xu at Huawei Connect 2025 (1 PFLOPS on the page)
FP4, dense: 2,000 TFLOPS They will be able to deliver 1 PFLOPS in FP8, MXFP8, and HiF8, and 2 PFLOPS in MXFP4. Keynote speech by Eric Xu at Huawei Connect 2025 (2 PFLOPS on the page)

Huawei gives one compute figure for the 950 die that the 950PR and 950DT share. It gives 144 GB and 4 TB/s for the 950DT only, so memory and bandwidth are left out. The dates are the 950DT’s: Huawei promised the 950PR for the first quarter of 2026, and no Huawei page this site could read says the 950PR shipped. Huawei states no chip power.

1 dated promise and the shipment

each date links to the page it is quoted from

18 Sep 2025 Promised Ascend 950DT available in the fourth quarter of 2026. The Ascend 950DT chip will be available in the fourth quarter of 2026. Keynote speech by Eric Xu at Huawei Connect 2025
17 Sep 2026 Shipped: the Atlas 950 SuperPoD, built with Ascend 950DT chips, in large-scale commercial use. Atlas 950 SuperPoD is seeing large-scale commercial use. Advancing the Agentic World, Building a Solid Silicon Foundation

Named at 1 AI data center

H100-equivalents and power are the whole site’s today, with every chip it holds

AI data centers that name this chip
Site	Owner	Country	H100e	MW	Chips
Envision Galaxy Ulanqab	Undisclosed	China	0	0	Ascend 950PR, Ascend 950DT

In no GPU cluster of Epoch AI’s list

Epoch AI’s list of AI supercomputers, which stops in July 2025

Epoch AI’s names for this chip: Ascend 950PR; Ascend 950DT.

Other Huawei chips

Ascend 910C Huawei shipped, no dated promise on record

Sources

accelerators.csv Every dataset → Status → Glossary →

Huawei’s own pages, each linked beside the figure or date it gives. Each page was read and each quote checked on 23 Oct 2026.
Where it runs: Epoch AI, AI data centers and Epoch AI, GPU clusters, CC BY 4.0.