DeepSeek delays new AI model amid Huawei chip issues- FT

Chinese AI startup DeepSeek delayed its R2 model launch due to persistent technical issues encountered while training with Huawei's Ascend chips, compelling a pivot back to Nvidia chips for core training. This incident highlights the significant practical challenges Chinese AI developers face in reducing reliance on advanced U.S. semiconductor technology, particularly Nvidia's dominant AI chips, despite Beijing's strategic push for domestic alternatives and recent scrutiny over Nvidia chip orders. The situation underscores the ongoing difficulties in achieving chip independence amidst U.S.-China tech tensions, reinforcing Nvidia's continued market leadership in advanced AI chip training.

Analysis

The delayed launch of DeepSeek's highly anticipated R2 AI model highlights a significant setback for China's semiconductor self-sufficiency ambitions. The failure to effectively train the model using Huawei's Ascend chips, necessitating a pivot back to Nvidia (NVDA) hardware for the critical training phase, underscores the persistent technological gap between Chinese domestic offerings and leading U.S. technology. This event demonstrates that despite strong political encouragement from Beijing and U.S. export controls designed to foster local alternatives, Chinese AI developers, including major players like Tencent (HK:0700) and Alibaba (BABA), remain reliant on Nvidia's superior performance for computationally intensive tasks. The decision to relegate Huawei's chips to less demanding inference workloads, while continuing to use Nvidia's H20 for training, provides a clear, real-world benchmark of the current performance disparity. This reliance persists even as Beijing begins to question local firms' orders of Nvidia chips, indicating that practical necessity is currently overriding political directives.

Key Decisions for Investors

• The failure of a key domestic competitor's chip reaffirms Nvidia's (NVDA) technological moat and validates the bull case, suggesting continued, non-discretionary demand from the Chinese market despite geopolitical headwinds.
• Investors with exposure to Chinese AI developers, including Tencent and Alibaba, should treat reliance on U.S. chip imports as a material risk factor, as access remains vulnerable to both U.S. export policy and Chinese domestic regulation.
• This event provides tangible evidence that China's path to semiconductor independence in high-end AI is fraught with significant technical hurdles, warranting caution against investment theses predicated on a rapid displacement of established U.S. technology leaders.