Huawei’s CloudMatrix Cluster Achieves 30% Utilization After Training DeepSeek’s 1.6-Trillion-Parameter AI Model
Key Points
Huawei CloudMatrix achieved ~30% utilization while training DeepSeek’s 1.6T parameter AI model.
Over 1,000 Ascend 910C chips were used in large-scale post-training.
DeepSeek V4-Pro uses MoE architecture with 1M-token context window.
Milestone highlights rising competition between Huawei’s Ascend ecosystem and Nvidia AI infrastructure.
Huawei’s AI ambitions reached a new milestone in June 2026 when reports revealed that its CloudMatrix cluster achieved 30% utilization while post-training DeepSeek V4-Pro, a massive 1.6-trillion-parameter AI model. The achievement highlights the growing capabilities of Huawei’s Ascend chip ecosystem at a time when global competition in artificial intelligence infrastructure is intensifying.
As China pushes to reduce reliance on foreign technology, this development raises important questions about the future of AI hardware, performance, and the balance of power in the industry.
Huawei CloudMatrix Reaches a New AI Training Milestone
What Was Achieved?
Huawei and its research partners have reported a major advance in large-scale AI training. In June 2026, a Huawei-led team announced that it had successfully completed full-parameter post-training of DeepSeek V4-Pro, a 1.6-trillion-parameter artificial intelligence model, using more than 1,000 Ascend 910C processors.
The project reportedly ran on Huawei’s CloudMatrix infrastructure and achieved approximately 30% Model FLOPs Utilization (MFU). This means a significant share of the cluster’s theoretical computing power was converted into useful training work. Sources: Tom’s Hardware, June 2026; Yahoo Tech, June 2026.
Why the 30% Utilization Figure Matters?
Model FLOPs Utilization is one of the most important efficiency metrics in AI training. A higher percentage means less computing power is wasted.
For large-scale AI systems, maintaining strong utilization across thousands of chips is difficult because communication bottlenecks often slow performance. Reaching 30% MFU suggests Huawei’s software stack, networking system, and hardware coordination have improved significantly. Industry observers view this as an important step in closing the gap with leading AI infrastructure providers.
DeepSeek V4-Pro: The 1.6-Trillion-Parameter Model Behind the Breakthrough
Key Specifications
DeepSeek released its V4 family on April 24, 2026. The flagship V4-Pro model contains 1.6 trillion total parameters and uses a Mixture-of-Experts architecture. However, only around 49 billion parameters are activated for each token processed, reducing computational costs.
The model also supports a context window of up to one million tokens. This allows it to analyze and remember extremely large amounts of information during a single interaction. According to DeepSeek, the model competes with leading frontier AI systems while maintaining lower operating costs.
Why DeepSeek Chose Huawei Hardware?
DeepSeek’s latest generation was built with Huawei’s Ascend ecosystem in mind. This marks a notable shift from earlier dependence on Nvidia-based infrastructure.
The move reflects China’s broader strategy to strengthen domestic AI capabilities amid ongoing export restrictions on advanced semiconductor technology. Huawei’s cloud division has already worked with several Chinese AI companies to expand local computing options. The DeepSeek partnership represents one of the largest examples of that effort so far.
How Huawei’s CloudMatrix Challenges Nvidia’s AI Dominance?
CloudMatrix Architecture Explained
CloudMatrix is Huawei’s large-scale AI computing platform designed to connect vast numbers of Ascend processors into a unified training cluster. Instead of relying on a small number of extremely powerful chips, Huawei focuses on scaling performance through tightly connected hardware and software systems. This architecture aims to maximize efficiency while supporting increasingly larger AI workloads.
Comparing Huawei and Nvidia Approaches
NVIDIA remains the global leader in AI training hardware. Its CUDA ecosystem, software tools, and high-performance GPUs continue to dominate many frontier AI projects. Huawei, however, is pursuing a different strategy. Rather than matching Nvidia chip-for-chip, the company is optimizing entire clusters. The reported DeepSeek training result demonstrates how Huawei is attempting to compete through system-level efficiency and large-scale deployment.
Strategic Importance for China
The achievement goes beyond technology. It also supports China’s long-term goal of reducing dependence on foreign AI infrastructure.
Successful deployment of a 1.6-trillion-parameter model on domestic hardware could encourage wider adoption of Ascend chips across research institutions, cloud providers, and enterprise AI projects.
Industry Reactions, Challenges, and What Comes Next
Why Experts Remain Cautious
Despite the announcement, some analysts are urging caution. Huawei has not publicly released detailed benchmark comparisons against Nvidia systems. Independent validation of training speed, energy efficiency, and scalability remains limited.
Experts also note that post-training is less demanding than full pre-training, which requires substantially more computing resources.
Future Implications
Even with unanswered questions, the announcement signals rapid progress in China’s AI infrastructure sector. Companies are investing heavily in domestic computing ecosystems, software frameworks, and advanced networking technologies.
Many organizations now use an AI stock analysis tool and broader AI forecasting systems to evaluate how infrastructure developments may affect semiconductor and cloud-computing companies. If Huawei continues improving CloudMatrix performance, competition in the global AI hardware market could intensify throughout 2026 and beyond.
Conclusion
Huawei’s reported 30% CloudMatrix utilization during the post-training of DeepSeek V4-Pro represents a meaningful milestone for China’s AI industry. While questions remain about independent benchmarking and full-scale training performance, the achievement highlights growing confidence in Huawei’s Ascend ecosystem.
As AI demand continues to surge worldwide, CloudMatrix is emerging as a platform worth watching closely in the race to build the next generation of advanced AI infrastructure.
Disclaimer:
The content shared by Meyka AI PTY LTD is solely for research and informational purposes. Meyka is not a financial advisory service, and the information provided should not be considered investment or trading advice.
What brings you to Meyka?
Pick what interests you most and we will get you started.
I'm here to read news
Find more articles like this one
I'm here to research stocks
Ask Meyka Analyst about any stock
I'm here to track my Portfolio
Get daily updates and alerts (coming March 2026)