A sharp rise in demand for Huawei’s Ascend 950 AI chips is rippling through China’s technology sector following the release of DeepSeek’s latest V4 artificial intelligence model, which has been optimized to run on the Shenzhen-based company’s hardware. The development has triggered an urgent wave of procurement activity among major Chinese internet and cloud computing firms, signaling a notable acceleration in the country’s push toward semiconductor self-reliance.
According to multiple people familiar with ongoing procurement discussions, leading Chinese tech companies—including ByteDance, Tencent, and Alibaba—have begun actively engaging Huawei regarding new orders for the Ascend 950 series. The interest extends beyond traditional internet platforms, with cloud computing providers and GPU rental firms also reportedly scrambling to secure supply, underscoring the broader ecosystem-wide impact of DeepSeek’s latest model rollout.
DeepSeek V4 reshapes domestic AI infrastructure demand
At the center of the surge is DeepSeek’s V4 model, released last week, which has been specifically optimized for Huawei’s Ascend architecture. Industry observers say this represents a strategic inflection point: rather than adapting models to foreign chips, Chinese AI developers are increasingly tailoring their systems to domestic semiconductor platforms.
Huawei has said its Ascend SuperNode infrastructure—built around the Ascend 950 series—fully supports V4 inference workloads, meaning the system is designed to efficiently handle real-time AI tasks such as query responses and automated decision-making. The company also stated that its entire Ascend SuperNode product line has been upgraded to accommodate the model.
The 950PR variant of the Ascend 950 series is currently regarded as the most advanced domestically developed AI chip capable of supporting a compressed numerical computation technique that improves processing efficiency and reduces operational costs. This capability is particularly significant for large-scale AI inference workloads, where speed and cost per computation are critical.
Competitive positioning against Nvidia amid export constraints
While the Ascend 950PR is reported to outperform Nvidia’s H20 chip—the most capable processor previously authorized for export to China before restrictions tightened—it still trails Nvidia’s more advanced H200 model. However, geopolitical and regulatory tensions have complicated the availability of the H200 in China, leaving a supply gap that domestic players are increasingly positioned to fill.
Despite having received regulatory approval for export in both the United States and China, shipments of the H200 have yet to materialize due to unresolved disagreements between Washington and Beijing over export conditions. This ongoing uncertainty has further strengthened Huawei’s position in the domestic AI chip market.
Rapid adoption across cloud platforms
The momentum created by DeepSeek V4 has already translated into immediate deployment across China’s major cloud service providers. Alibaba Cloud integrated the model into its Bailian platform on the day of release, offering both V4-Pro and V4-Flash versions aligned with DeepSeek’s official pricing structure. Tencent Cloud followed closely, launching preview access via its TokenHub platform and deploying the model across both domestic infrastructure and international nodes in Singapore.
This rapid rollout has significantly expanded access to the model for developers and enterprise users, effectively increasing the volume of AI workloads being processed across Chinese infrastructure. As a result, demand for high-performance AI chips like Huawei’s Ascend 950 has surged in tandem.
Supply constraints and production outlook
Despite strong demand, supply constraints remain a significant challenge. DeepSeek has acknowledged that production bottlenecks will persist in the near term, even as it offers discounted access to V4 until early May to encourage adoption. The company also indicated that pricing for its V4-Pro model could decline substantially in the second half of 2026, once Huawei’s Ascend 950 supernode systems reach full-scale production.
Current projections suggest Huawei may ship approximately 750,000 units of the Ascend 950PR this year, with mass production beginning in April and broader deployment expected in the latter half of 2026. However, analysts and industry sources caution that output may still fall short of demand, largely due to restrictions on advanced semiconductor manufacturing equipment that limit China’s ability to scale production of cutting-edge chips.
Strategic shift toward domestic AI ecosystem
DeepSeek’s decision to open-source its V4 models under the permissive MIT license further amplifies the potential for widespread adoption across China’s AI ecosystem. The model, which includes V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion parameters, supports a context window of up to one million tokens, making it suitable for large-scale enterprise and developer applications.
By aligning its model architecture with Huawei’s hardware, DeepSeek is reinforcing a broader national strategy aimed at reducing reliance on foreign semiconductor technology, particularly from U.S. firms such as Nvidia. For Beijing, the development represents a step toward strengthening domestic capabilities in advanced computing infrastructure amid ongoing technological competition with Washington.
As adoption accelerates and supply chains adjust, the convergence of DeepSeek’s AI models and Huawei’s Ascend chips is increasingly shaping the trajectory of China’s artificial intelligence landscape—one defined by rapid localization, constrained supply, and intensifying global competition in advanced computing.