DeepSeek V4 is reportedly delayed again and shifting to Huawei chips

aiai-modeling 2 posts · 1 accounts

DeepSeek V4 is said to have slipped twice and now be targeting a late-April release on Huawei Ascend 950PR chips instead of Nvidia hardware, signaling a sharper domestic-stack pivot.

From the sources (2 posts)

@jukan05

@deepwhitman DeepSeek is indeed delayed. Please understand that I can’t go into detail.

@jukan05

@OmerCheeema But aren't those Chinese labs walking away from the playbook? DeepSeek V4 has slipped twice and is now targeting a late-April release on Huawei's Ascend 950PR chips instead of Nvidia silicon, while Alibaba and Zhipu are pivotin

Market impact

38 medium

NVDA SOXX

Why it matters: A leading Chinese lab reportedly training its next frontier model on Huawei Ascend chips instead of Nvidia silicon (with Alibaba and Zhipu also pivoting) signals domestic-stack substitution that erodes Nvidia's China accelerator demand.

At publish time: Headline states the pivot from Nvidia hardware to Huawei Ascend 950PR chips and a domestic-stack shift; the delay itself is backward-looking and lower-signal.

model confidence: low

What happened next

1–2 weeks later

played out

Follow-ups confirm DeepSeek V4 preview shipped on Huawei chips, advancing China's domestic AI stack; substitution thesis materialized, but NVDA's directional move not cleanly attributable to this single low-magnitude signal.

based on follow-up coverage

Follow-up coverage

No follow-up coverage matched this story.

Fed tells big banks not to fight upcoming capital rules aggressively SK Hynix is expected to post record first-quarter earnings