DeepSeek V4 triggers rush for Huawei AI chips in China

DeepSeek V4 has triggered stronger demand for Huawei Ascend 950 AI chips in China.

Reuters reported on April 29 that ByteDance, Tencent and Alibaba are among the major Chinese internet companies reaching out to Huawei about new AI chip orders. The report cites three people familiar with procurement discussions. Chinese cloud providers and GPU rental companies are also trying to secure capacity.

The facts make this more than another model launch. DeepSeek has optimized V4 for Huawei's Ascend infrastructure. Huawei says its Ascend SuperNode product line has been adapted for V4 inference, meaning production use of the model. Reuters also reports that Ascend 950PR significantly outperforms Nvidia's H20 in China, while still trailing the more advanced H200, which remains caught in regulatory limbo between the US and China.

The leadership implication is that AI infrastructure can no longer be treated as a simple capacity decision. Model choice, chip access, cloud architecture and geopolitical risk are now linked. When Chinese platforms converge on a model that runs well on domestic chips, a more regional AI ecosystem starts to form, with its own cost curves, bottlenecks and lock-in effects.

For Norwegian CIOs, this does not make Huawei a straightforward option. Security, sanctions exposure, data location and public-procurement constraints remain material issues. The point is broader: dependence on Nvidia and US hyperscalers is also a risk. It belongs in architecture and board-level discussions before large AI platforms become hard to unwind.

Practical advice: separate model strategy from infrastructure strategy. Require exit plans for critical AI services, test real portability across models, and ask suppliers to document how price, performance and availability would change if the chip market fragments further. This is not only a China story. It is a signal that enterprise AI is becoming as much a supply-chain question as a software question.