At the beginning of the new year of 2025, DeepSeek, a Chinese AI lab, released an inference large language model called DeepSeek-R1, whose performance is comparable to OpenAI's latest o1 model. In addition, DeepSeek also announced that its pre-trained version, DeepSeek-V3, will cost only $5.5 million to train. The news sparked a strong reaction from the market, with Nvidia's stock price plummeting 17% on January 22, wiping off about $589 billion in market value. The drop was Nvidia's worst one-day decline since the early days of the pandemic in March 2020. This panic not only affected Nvidia's own stock price, but also spread to the entire technology stock market, with the Nasdaq Composite Index falling 3.07% on the day, and the stock prices of AMD, TSMC and other related companies falling simultaneously, and the market evaporated more than $1 trillion in a single day.
Technology Disruption: How DeepSeek is Rewriting the Game?
DeepSeek's breakthrough technology hits the industry's pain points:
Hardware efficiency revolution: Only 2,048 NVIDIA H800 GPUs (castrated version of the H100) were used to complete the training, and the DualPipe parallel pipeline algorithm was used to reduce communication overhead and optimize bandwidth utilization to the limit.
Cost control: FP8 precision is used to compress computing requirements, avoid tensor parallelism and reduce memory consumption, so that the cost of a single pre-training is reduced to $5.5 million (only 1/30 of the training cost of GPT-3); Through the DualPipe parallel pipeline algorithm, optimized communication cores, adjusted memory usage, and reduced accuracy, Deepseek greatly improves training efficiency and reduces dependence on NVIDIA's high-end chips. This has led to fears that the long-term demand for Nvidia's GPUs will decrease, shaking investors' confidence in Nvidia stock and causing the stock price to fall.
The open-source ecosystem weakens NVIDIA's advantage: Deepseek adopts the MIT open-source license to attract global developers to participate in optimization and promote the democratization of AI technology. This open model weakens the moat of NVIDIA's CUDA ecosystem. The CUDA ecosystem is an important competitive advantage for NVIDIA in the field of GPU computing, and developers can develop various applications based on CUDA. However, DeepSeek's open-source model provides opportunities for other chip manufacturers, such as domestic chips (such as Huawei's Ascend), making the market more competitive and further affecting investors' expectations for Nvidia stock.
Figure: Nvidia's stock price has been hit, and the artificial intelligence chip bubble has burst?
The decline in Nvidia's stock price is mainly due to the new AI model released by DeepSeek that triggered a market reassessment of the cost-effectiveness of AI chips.
Despite the short-term volatility, three major trends underpin the long-term value of the sector:
Technology symbiosis: DeepSeek technology is completely based on the development of NVIDIA GPU clusters, and its efficiency improvement will stimulate more enterprises to purchase NVIDIA chips to build computing power pools;
Software ecological moat: NVIDIA's CUDA platform occupies 90% of developers' minds, and the migration cost far exceeds the hardware procurement cost.
Exponential explosion of demand: The global demand for AI computing power is increasing by 10 times every year, and 100,000 H100-level chips are needed to train GPT-5 alone, while the current global H100 number is less than 500,000.
In addition, the competition in the artificial intelligence chip market is diversified. In addition to international giants such as Nvidia, Intel, and AMD, local companies in China and other countries and regions are also on the rise. The Chinese government regards AI chips as the core infrastructure in the field of artificial intelligence, and has provided financial support through a series of policies, promoted technological research and industrial chain integration, and promoted the development of domestic AI chips. Huawei, Cambrian, Horizon and other companies have made significant progress in the field of AI chips, forming an industrial layout covering the cloud, edge, and terminals. With technological innovation and in-depth understanding of the local market, these companies occupy a certain share in the domestic market and gradually expand to the international market. The diversification of market competition prompts enterprises to continuously innovate, improve product performance and service quality, and promote the development of the entire industry.
The decline in Nvidia's stock price is the result of a combination of factors, but this does not mean that the development prospects of the artificial intelligence chip industry are bleak. In the long run, artificial intelligence chips still have broad development space in the future due to factors such as technological innovation, application scenario expansion, market competition diversification, and policy support. Whether it is an international giant or a local enterprise, they need to continue to innovate and improve their technical strength and market competitiveness to adapt to the development trend of the industry and occupy a place in the market.
