DeepSeek AI's R2 Model Launch Delayed Due to Huawei Chip Training Setbacks

Beijing, China – Chinese artificial intelligence startup DeepSeek AI has postponed the release of its highly anticipated R2 model, originally slated for May, after encountering significant technical challenges in training the model using Huawei's domestically produced Ascend chips. The delay, widely reported on August 14, 2025, underscores the persistent gap between Chinese and U.S. chip technology for advanced AI training, despite Beijing's strategic push for technological self-sufficiency. A social media user, Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞), reacted to the news, stating, > "The Mask has been ripped off! as my aunties in Kaifeng say: oy vey."

DeepSeek, known for its cost-efficient and high-performing R1 model, was reportedly encouraged by Chinese authorities to adopt Huawei's Ascend processors for its R2 development. However, the company faced "persistent technical issues" during the training process with the Ascend chips, ultimately compelling it to revert to using Nvidia chips for the critical training phase. Huawei's chips will now primarily be used for inference, the stage where the AI generates results, rather than the more demanding training.

The setback highlights the technical hurdles Chinese companies face in developing cutting-edge AI without access to the most advanced U.S. hardware. Industry insiders point to stability issues, slower inter-chip connectivity, and inferior software as key reasons why Chinese chips still lag behind Nvidia's offerings. Huawei reportedly sent engineers to DeepSeek's offices to assist, but a successful training run on the Ascend chip could not be achieved for the R2 model.

DeepSeek's founder, Liang Wenfeng, has reportedly expressed dissatisfaction internally with the R2's progress, emphasizing the need to build an advanced model to maintain the company's lead in the competitive AI field. This delay follows DeepSeek's R1 model having previously "shocked" the US market in January 2025 by demonstrating high performance at a fraction of the cost of Western counterparts, briefly impacting Nvidia's market valuation. The current situation, however, reveals the ongoing reliance on foreign technology for foundational AI development.

The incident also comes as the U.S. government continues to implement export controls on advanced chips to China, aiming to slow Beijing's AI advancements. While China is committed to fostering a domestic chip ecosystem, DeepSeek's challenges with the R2 model illustrate the complex and time-consuming nature of achieving true self-reliance in high-performance computing necessary for frontier AI development. The company continues to work with Huawei to optimize the model for Ascend inference, with a potential release still anticipated in the coming weeks.