China has granted conditional approval for DeepSeek, its top artificial intelligence startup, to purchase NVIDIA's powerful H200 chips, marking a significant shift in Beijing's AI strategy that prioritizes rapid innovation over strict technological self-reliance. The decision, confirmed by two sources familiar with the matter, comes as part of a broader approval that includes tech giants ByteDance, Alibaba, and Tencent, who have collectively been authorized to purchase more than 400,000 H200 chips.
The H200, NVIDIA's second most powerful AI chip, has emerged as a major flashpoint in US-China technology relations. While the United States earlier this month formally cleared the way for NVIDIA to sell the H200 to China where demand remains extremely strong, Beijing's hesitation to allow imports had been the main barrier to shipments until now. China's industry and commerce ministries have granted approvals but stipulated conditions that are still being finalized by the National Development and Reform Commission.
NVIDIA CEO Jensen Huang told reporters in Taipei on Thursday that his company had not received official information about the approvals and added that he believed China was still finalizing the licensing details. Any purchases of H200 chips by DeepSeek could draw scrutiny from US lawmakers, following allegations that NVIDIA had helped DeepSeek refine AI models later used by the Chinese military.
Meanwhile, DeepSeek continues to keep the AI world guessing about the release of its next-generation models. According to The Information, the company is set to unveil its V4 model in mid-February, featuring enhanced coding capabilities that are expected to rival and potentially surpass closed-source models like Claude and the GPT series. The V4 model promises major leaps forward in code generation, long-context comprehension, and training stability.
Speculation has swirled about potential delays in the launch of both V4 and R2 models, which would succeed V3 introduced in December 2024 and R1 released in January 2025. Industry insiders suggest that difficulties in training models on domestically produced Huawei Ascend AI chips, reportedly encouraged by Chinese authorities, may have caused extended delays. These chips presented issues with instability and performance, forcing DeepSeek to pivot back to NVIDIA hardware for critical training phases.
DeepSeek kicked off 2026 with a groundbreaking research paper signaling a push to train bigger models for less computational cost, a method that analysts describe as a breakthrough for scaling AI capabilities. This approach aligns with the company's reputation for achieving frontier AI performance with significantly fewer resources than American competitors, a factor that disrupted global AI markets when DeepSeek first gained international attention.
Comments