The Information reported DeepSeek is helping shape future Huawei Ascend chip designs and released open-source programming tools for Ascend, including optimization of a 128-chip Ascend 950 supernode.
ConcludedRan 179 days3 sourcesupdated 19 Oct 2026
DeepSeek and Huawei Co-Design Chips and Software for V4 Launch
The gist
The collaboration represents a working hardware-software integration between a leading Chinese AI model developer and a domestic chip maker, reducing dependence on Nvidia and CUDA. It provides a concrete test of whether Ascend chips can support cutting-edge large model inference at scale under US export restrictions.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
Huawei's Ascend 950PR and 950DT chips achieved 'day zero' adaptation to DeepSeek's V4 model on its release day, with performance described as 'significantly improved' due to close collaboration before launch. Huawei's full Ascend SuperNode product line is described as 'fully adapted' for V4 inference. The co-design included Huawei preparing its CANN software with fused operators and optimizations for the dual-core chip design ahead of the V4 launch. DeepSeek V4 is a 1.6 trillion parameter Mixture of Experts model with 1 million token context support. DeepSeek is also actively helping shape the design of Huawei's future AI chips and has released open-source programming tools for Ascend chips, including work optimizing a supernode built around 128 Ascend 950 chips. DeepSeek acknowledged V4 will face throughput constraints until Ascend 950PR supernodes ship at scale in the second half of the year.
How it developed
Jensen Huang argued that AI sovereignty is defined by owning data, culture, and knowledge, not by the nationality of the underlying hardware.
South China Morning Post reported Huawei's Ascend 950PR and 950DT chips achieved 'day zero' adaptation to DeepSeek V4 at launch, with the full Ascend SuperNode line fully adapted for V4 inference.
Sources
Milk Road AI on X
huaweicentral.com
Related
- Related storyDHS's DeepSeek federal exclusion order
- DeepSeek's fundraise and Huawei chip cluster
- DeepSeek's fundraising for Huawei accelerators
- DeepSeek's DeepInfer inference engine
- DeepSeek's reported China funding round
BackgroundCompute: the chips, clouds and data centers behind AI
Get the daily email
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.