In a powerful demonstration of China’s accelerating drive toward technological self-reliance, Tencent Cloud has officially announced the deployment of the DeepSeek V4 foundation model on Huawei’s Ascend 950 artificial intelligence processors. The integration, confirmed by Reuters and detailed by SimplyWall.st, represents a milestone for the domestic AI ecosystem. It proves that China’s leading cloud providers can now offer state-of-the-art generative AI capabilities using a technology stack that is entirely free of US-origin silicon or software frameworks.
The deployment is not merely a technical proof of concept; it is a commercial reality. Tencent Cloud is now offering DeepSeek V4 inference services to its enterprise customers via its Model-as-a-Service (MaaS) platform, powered by clusters of Ascend 950 chips. This move signals a definitive shift in the market dynamics of China’s AI infrastructure, moving away from a reliance on Nvidia’s hardware and CUDA software ecosystem toward a consolidated, homegrown alternative.
The Significance of the Ascend 950 Integration
The successful integration of DeepSeek V4, currently one of the most capable open-weight models globally, onto the Ascend 950 architecture is a complex engineering feat. Historically, the primary barrier to adopting domestic AI chips in China has not been the hardware’s raw computing power (FLOPS), but rather the immaturity of the accompanying software ecosystem. Nvidia’s CUDA platform has enjoyed a near-monopoly because it is highly optimized, deeply entrenched in the developer community, and seamlessly supports popular frameworks like PyTorch.
(Related: Huawei Expects AI Chip Revenue to Jump 60% to $12 Billion in 2026)
To make DeepSeek V4 run efficiently on Huawei’s silicon, Tencent Cloud engineers had to deeply optimize the model using Huawei’s CANN (Compute Architecture for Neural Networks) software stack. This work involved migrating the model from Nvidia’s CUDA ecosystem to Huawei’s CANN framework, rewriting core operators and fine-tuning the model’s parallelization strategies to match the Ascend 950’s architecture. The successful deployment demonstrates that Tencent Cloud can deliver inference services on domestic hardware at commercially viable speeds and cost-efficiencies, a significant milestone for the maturation of Huawei’s software ecosystem.
A Strategic Pivot for Tencent Cloud
For Tencent Cloud, the decision to aggressively deploy the Ascend 950 is a strategic necessity. As US export controls continue to tighten, restricting access to even downgraded Nvidia chips like the H20, Chinese cloud providers face an existential threat to their AI infrastructure expansion plans. By proving it can deliver top-tier AI services on domestic hardware, Tencent Cloud is mitigating supply chain risks and assuring its enterprise customers of long-term service continuity.
Furthermore, this deployment positions Tencent Cloud as a leader in the “domestic substitution” (Xinchuang) market. Many Chinese state-owned enterprises, government agencies, and critical infrastructure operators are under strict mandates to transition their IT systems to domestic suppliers for security reasons. By offering a fully localized AI stack, from the underlying silicon to the foundation model, Tencent Cloud is highly competitive for these lucrative government and enterprise contracts.
The Consolidation of the Domestic Stack
The Tencent-Huawei-DeepSeek collaboration highlights a broader trend: the rapid consolidation of China’s domestic AI stack. Rather than dozens of companies attempting to build everything from scratch, a clear division of labor is emerging among the national champions. Huawei is solidifying its position as the dominant provider of AI hardware and foundational software (CANN/MindSpore). Startups like DeepSeek and established players like Alibaba (with Qwen) are focusing on training world-class foundation models. Meanwhile, cloud giants like Tencent and Alibaba Cloud are providing the massive infrastructure and MaaS platforms needed to deliver these models to end users.
This collaborative ecosystem is essential for China to remain competitive in the global AI race. By pooling their resources and expertise, these companies can overcome the manufacturing bottlenecks and software challenges that no single entity could solve on its own. The successful deployment of DeepSeek V4 on the Ascend 950 proves that this collaborative model is working, yielding commercial products that rival those built on the US technology stack.
(Related: Chinese Tech Giants Scramble for Huawei Ascend Chips as DeepSeek V4 Triggers Supply Crunch)
Implications for the Global AI Landscape
The consolidation of China’s domestic AI stack has profound implications for the global technology landscape. It demonstrates that US export controls, while causing short-term pain and forcing expensive workarounds, have ultimately accelerated China’s pursuit of technological independence. The captive market created by these restrictions has provided Huawei and its partners with the capital, real-world testing environments, and sheer necessity required to rapidly iterate and improve their products.
As the Ascend 950 and subsequent generations of Huawei chips are increasingly deployed across China’s cloud infrastructure, the total addressable market for Western chipmakers like Nvidia will inevitably shrink. Moreover, as China’s domestic software ecosystem (CANN) matures, it may eventually become an exportable alternative for countries seeking to build their own AI infrastructure without relying on US technology. The Tencent Cloud deployment is a clear signal that the bifurcation of the global AI technology stack is no longer a theoretical possibility; it is an operational reality.