DeepSeek V4 drives Huawei’s Ascend 950PR to the center of China’s AI infrastructure race

Summary

The launch of DeepSeek V4 has sharply accelerated demand for Huawei Technologies’s Ascend 950PR AI chips, with major Chinese technology companies moving quickly to secure supply for upcoming model training and inference needs. The development marks a significant shift toward domestically optimized AI infrastructure in China, as firms reduce dependence on restricted foreign chips. However, supply constraints at local foundries continue to pose a major challenge.

BEIJING, April 29, 2026 — The release of DeepSeek V4 has triggered a major demand surge for Huawei’s Ascend 950PR AI processors, with China’s largest internet and cloud companies accelerating orders for domestic compute infrastructure. Recent reports indicate that companies including Alibaba Group, Tencent, and ByteDance are rushing to secure supply for their 2026–27 AI expansion roadmaps.

shift away from restricted foreign hardware

For the past two years, many Chinese firms relied on restricted versions of Nvidia chips, particularly the H20, due to U.S. export controls. DeepSeek V4 is among the first major Chinese frontier-scale AI models designed to run natively on Huawei’s Ascend architecture, signaling a deeper move toward domestic hardware ecosystems. Reuters reports that the Ascend 950PR significantly outperforms Nvidia’s H20 in China-available deployments, although it still trails higher-end global chips such as Nvidia’s H200.

large-scale enterprise demand builds

According to multiple reports, demand is now extending beyond model developers to cloud service providers and GPU rental companies. Huawei is reportedly targeting shipments of around 750,000 Ascend 950-series units in 2026, though demand may exceed this level. Major cloud platforms have already begun rolling out DeepSeek V4-compatible instances to enterprise clients.

supply remains the key bottleneck

Despite the momentum, production remains constrained by domestic manufacturing capacity. The chips are widely reported to be produced by Semiconductor Manufacturing International Corporation, where advanced-node capacity remains limited under ongoing equipment restrictions. This has created a supply-tight environment similar to the global AI chip shortage seen in other markets.

Why this matters

• Domestic AI ecosystem growth: China’s leading AI companies are increasingly building around local silicon rather than relying on foreign chips.
• Strategic self-reliance: The DeepSeek-Huawei stack strengthens China’s push for semiconductor and AI independence.
• Supply-side pressure: Manufacturing bottlenecks may slow deployment despite strong enterprise demand.

FAQs

Q1. Is the Ascend 950PR better than Nvidia’s H20?

In China’s currently available market, reports suggest it delivers stronger performance for DeepSeek V4 workloads and is emerging as the preferred domestic option.

Q2. Are Chinese tech firms placing bulk orders?

Yes. Major firms including Alibaba, Tencent, and ByteDance are reportedly seeking large allocations.

Q3. What is the biggest challenge now?

The main issue is production scale, as foundry capacity and advanced chip-making equipment remain constrained.