The domestic large model company DeepSeek has topped the free app download charts in the Apple App Store for both China and the United States, surpassing ChatGPT.
This achievement has not only garnered widespread attention in the market but has also significantly impacted the stock price of AI chip leader NVIDIA. Nevertheless, a spokesperson from NVIDIA has highly praised DeepSeek, calling it an outstanding advancement in the AI field, and noted that DeepSeek's work demonstrates how to utilize testing time scaling technology, widely available models, and computation resources that comply with export controls to create new models.
The NVIDIA spokesperson added that the inference process requires substantial support from NVIDIA GPUs and high-performance networks. NVIDIA currently masters three scaling patterns: ongoing pre-training and post-training, as well as new testing time scaling technology. This statement also implies that the GPUs used by DeepSeek fully comply with U.S. export control requirements.
Testing time scaling technology has been a hot topic in recent years, indicating that if a fully trained AI model invests more computational resources during prediction, text generation, or image generation, the model will produce higher quality answers. This technology has been applied in models such as OpenAI's o1.
On January 27, the three major U.S. stock indices closed mixed, with NVIDIA's stock price plummeting by 16.86% to $118.58 per share, hitting the lowest point in the past ten months. NVIDIA's total market value is $2.90 trillion, evaporating $590 billion in a single day, setting the record for the largest single-day loss in market value for an individual stock. This drop also significantly reduced the fortune of founder Jensen Huang by $21 billion.
Under NVIDIA's leadership, the U.S. semiconductor chip sector faced collective setbacks. DeepSeek, a large model company under the quantitative giant Huanfang Quant, officially released the inference large model DeepSeek-R1 on January 20. R1 has gained widespread attention for its outstanding performance, open-source nature, significantly reduced price, and training costs. The pre-training cost of R1 is only $5.576 million, completed over 55 days on a cluster of 2048 NVIDIA H800 GPUs.
Key Points:
📈 DeepSeek's app tops the free app download charts in the Apple App Store for both China and the U.S.
💬 NVIDIA praises DeepSeek as an outstanding advancement in AI, stating that inference requires a large number of GPUs.
💸 NVIDIA's stock price plummets by 16.86%, evaporating $590 billion in a single day, setting the record for the largest single-day loss in market value for an individual stock.
