Jakarta, INTI - Chinese AI startup DeepSeek has released a preview of its latest AI model optimized to run on Huawei's Ascend chips. The DeepSeek V4 AI model is a "gateway" and a strong signal that China's AI ecosystem is shifting away from its reliance on Nvidia GPU systems.
Most global AI models, including those in China, were developed using Nvidia's CUDA ecosystem for training and inference. Now, DeepSeek is breaking this dependency pattern, claiming that its V4 model has successfully adapted inference to Huawei's Ascend chip, signaling the readiness of domestic chips for practical applications.
DeepSeek V4 Has Two Versions
DeepSeek V4 comes in two versions: Pro and Flash. The Pro variant is said to be the most advanced version, while Flash is the more lightweight. DeepSeek integrates Nvidia’s GPUs and Huawei's Ascend NPUs into a single hardware validation framework. The company states that its fine-grained expert parallelism scheme has been verified to run on both platforms.
He Hui, director of semiconductor research at Omdia, said that the technology is a big deal for China, especially the country's AI industry.
"Huawei's Ascend chips are the country's best homegrown alternative to Nvidia, and supporting DeepSeek V4 shows that top Chinese AI models can now run on Chinese hardware," said He Hui.
V4 Model Performance
In terms of performance, DeepSeek claims that the V4 model can process over one million tokens, equivalent to the context capacity of models like OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6.
The Pro version is claimed to outperform other open-source models and is second only to Google's Gemini-Pro-3.1 closed-source model. The company also said that V4 is very suitable agent-based AI, as it can execute more complex tasks than chatbots. However, despite its range of advanced capabilities, DeepSeek's latest AI model reportedly still cannot create images and videos.
The introduction of the V4 model comes amid restrictions on advanced chip exports by the United States, making China accelerate the development of its own technology. DeepSeek is also reportedly facing criticism from the US and rival companies, who attribute its success to its use of US-made technology.
However, the company denies these allegations and states that it does not intentionally use synthetic data generated by its rival, OpenAI. DeepSeek also acknowledged that it uses Nvidia chips, but did not specify whether those chips were affected by the export restrictions.
Conclusion
DeepSeek released a preview of its latest AI model, DeepSeek V4, optimized for Huawei's Ascend chips, marking a significant step toward reducing China's reliance on the Nvidia GPU ecosystem. V4 is available in Pro and Flash versions and supports high performance with large context capacity. DeepSeek V4 does not yet support image and video creation features. The model introduction was released amid pressure from US chip export restrictions.
Read more: Google will Invest US$10 Billion in Anthropic with an Additional of US$30 Billion if Targets are Achieved