DeepSeek Timeline: How It Stirred Up the AI Industry

From Stanford to MIT, DeepSeek - R1 almost overnight became the preferred model for researchers at top US universities, said Anjney Midha, a partner at Silicon Valley investment firm a16z. This model has become the focus of the industry with its disruptive computational efficiency and high applicability.

On January 27th, NVIDIA dropped 16.97%, with its market capitalization evaporating nearly $600 billion. Hours later, DeepSeek - R1 engineers open - sourced the Janus - Pro model on the Hugging Face platform, and its performance surpassed OpenAI's DALL - E3 model. This shockwave seems to have pushed the triple resonance of technological revolution, geopolitical games, and the capital cycle to a climax. Is the power game in the AI industry being rewritten?

Other tech giants were not spared either. Microsoft's share price fell 2.14%, ASML dropped 5.75%, and AMD declined 6.37%.

The next day, US tech stocks rebounded. Major tech stocks such as NVIDIA (+8.93%), Apple (+3.65%), TSMC (+1.34%), Microsoft (+2.91%), and Meta (+2.19%) all saw significant increases. This indicates that the market remains optimistic about the AI and technology industries.

Looking at the timeline of this technological surprise attack, this disruption can be divided into two waves: The first wave began on January 20th when DeepSeek officially released the R1 model and simultaneously open - sourced the model weights. Its performance is on par with the official version of OpenAI - o1. Even more remarkably, it was revealed that the training cost of the DeepSeek - R1 model was only $5.57 million, much lower than its peers, and its performance is comparable to OpenAI's GPT - 4o. On January 27th, DeepSeek further launched an iOS application, which quickly surpassed the official ChatGPT application and topped the Apple Store, setting off a technological tsunami that swept through the global capital market.

The industry marvels that DeepSeek - R1 has achieved significant breakthroughs in chip resource utilization, algorithm complexity, and inference speed. Its performance indicators comprehensively exceed existing cutting - edge AI models in multiple key areas, including OpenAI's GPT - 4.5 and Anthropic's Claude series.

The second wave of shock came from the Janus - Pro model. At 3:00 am on January 28th, Beijing time, DeepSeek open - sourced this multimodal model on the Hugging Face platform. According to the measurement of GenEval (an object - centered framework for evaluating the alignment capabilities of text - to - image generation models), the accuracy rate of the Janus - Pro model reached 84.2%, significantly surpassing DALL - E3's 79.7%. This result directly shattered the market's psychological defense.

In a report to clients, institutional analysts pointed out that the pillars of the US stock market are large - cap tech stocks represented by tech giants, and the valuations of these tech stocks are highly dependent on the optimistic expectations of the AI industry. Although the rise of DeepSeek has raised some doubts, the company may become a key variable that shakes the optimistic sentiment in AI.

In the same week that DeepSeek - R1 was released, US President Trump announced the launch of the Stargate project. According to a statement released by the White House, the project plans to invest up to $500 billion over the next four years to build up to 20 new data centers to support AI - related research and development. This move is seen as an important strategy for the US to maintain its technological leadership in the AI race.