Huawei Ascend 950 chip demand surges after DeepSeek V4 launch
Key Points
- Huawei's Ascend 950PR outperforms Nvidia's H20 chip (banned in China) but still trails the H200, creating market opportunity as H200 shipments remain stalled due to US-China disagreements
- Huawei plans to ship approximately 750,000 units of the 950PR chip in 2026, with mass production starting in April, though output will likely fall short of demand due to US equipment restrictions
- DeepSeek V4 deployed immediately on Alibaba Cloud Bailian and Tencent Cloud platforms at launch, with pricing expected to decline in H2 2026 once Huawei supernodes ship at scale
AI Summary
Summary
Key Development: Demand for Huawei's Ascend 950 AI chips has surged following DeepSeek's V4 model launch, which is optimized to run on Huawei hardware. Major Chinese tech companies including ByteDance, Tencent, and Alibaba are actively placing chip orders with Huawei.
Technical Specifications: The Ascend 950PR chip outperforms Nvidia's H20 (previously sold in China before Beijing blocked imports) but trails Nvidia's H200. DeepSeek V4 includes two versions: V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion parameters), both supporting a one-million-token context window under MIT open-source licensing.
Market Implications: This represents a significant shift toward China's domestic semiconductor ecosystem amid ongoing U.S. export restrictions. Huawei's Ascend 950 is currently the only domestic Chinese chip capable of supporting compressed numerical processing for AI workloads. The development fills a gap created by regulatory disputes between Washington and Beijing that have delayed H200 shipments to China.
Production & Supply: Huawei plans to ship approximately 750,000 units of the 950PR chip in 2026, with mass production beginning in April and full-scale shipments expected in the second half. However, supply constraints will persist due to U.S. restrictions on advanced chipmaking equipment limiting China's manufacturing capacity.
Cloud Deployment: Alibaba Cloud's Bailian platform and Tencent Cloud's TokenHub both launched DeepSeek V4 services on release day, expanding access to millions of users and increasing demand for AI processing capacity. Pricing for V4-Pro may decline in late 2026 once Huawei's supernodes scale production.
Model Analysis Breakdown
| Model | Sentiment | Confidence |
|---|---|---|
| GPT-5-mini | Bullish | 75% |
| Claude 4.5 Haiku | Bullish | 78% |
| Gemini 2.5 Flash | Bullish | 88% |
| Consensus | Bullish | 80% |