Introduction
Yesterday, Monday, January 27, 2025, the technology company Nvidia plummeted 17%, losing $589 billion in market capitalization in a historic stock market drop. The decline extended to the rest of the tech market, dragging the Nasdaq Composite Index down by 3.1%. The damage was concentrated on stocks related to artificial intelligence.
What caused this crash? Why has the tech world been shaken so dramatically?
The culprit behind this stock market earthquake has a name: DeepSeek.
Let’s set the stage. Who is Nvidia?
The American giant has invested colossal sums in recent years in its quest to dominate the AI sector. Nvidia’s ecosystem, spanning from its software to its hardware supply (FPGA and GPU processors), has made it exceptionally valuable to companies dedicated to AI development.
And the arrival of DeepSeek has shaken this monopoly and the future of Nvidia and other major tech companies worldwide.
Let’s analyze the situation.
Introducing DeepSeek
DeepSeek is a Chinese startup operating as a research-focused organization and characterized by its independence from external investments. Its founder, Liang Wenfeng, funds the startup through his own investment fund, HighFlyer. This allows him to attract talent and maintain an academic focus.
The launch of its AI model has stunned the world, propelling its models to the forefront of the AI sector. In fact, it has surpassed ChatGPT in downloads on mobile app stores.
Moreover, they are free to use, requiring no subscriptions for access to updates or improvements.
What is DeepSeek?
It is a large-scale LLM (Large Language Model) that surpasses most AI models. It boasts 671 billion (671B) parameters, almost 66% more than Llama by Meta (405B).
Its training cost approximately $5.6 million, compared to nearly $80 million spent by OpenAI.
It specializes in reasoning and can outperform OpenAI’s models in math and programming tests.
It can communicate in several languages, though it excels in English and Chinese.
The model has been developed using pure reinforcement learning, which means it doesn’t rely on pre-labeled data. The model can learn independently through trial and error, discovering and developing its own reasoning strategies for continuous improvement. Additionally, the R1 version combines supervised training to refine it and provide greater consistency.
And it is an open-source model.
Open Source Model
Unlike its Western rivals, which have developed private models, DeepSeek uses open source. This allows researchers and developers worldwide to examine, use, and modify its models. This approach has made a significant impact on the global developer community.
DeepSeek has thus initiated the path toward the “democratization” of advanced AI. It can be utilized not only in large infrastructures but also in smaller and more accessible setups.
Using DeepSeek
Its operation is identical to that of other language models:
- You can ask the chatbot about any topic, and it responds in real time.
- It can explain complex topics in simple terms—just ask.
- You can share documents for it to extract information or answer questions about their content.
- It allows you to write or debug programming code and solve math problems.
- It can review and correct texts, as well as generate them: social media content, poems, or specific articles.
In the text box where you type your question, you’ll find three buttons:
- A paperclip to attach files.
- A “DeepThink” button for more detailed information on any topic.
- A “Search” button to fetch information from the internet with source references.
Is There Censorship in DeepSeek?
There is some controversy surrounding this issue, as public opinion has questioned potential censorship of the model by the Chinese government.
However, we must acknowledge that all AI models are subject to the values and regulations of the countries where they were developed. This means no model answers sensitive questions about politics, religion, race, or gender, for instance. ChatGPT has avoided such topics since its inception.
In DeepSeek’s case, it has been observed to avoid answering questions related to especially sensitive topics for the country, such as the Tiananmen Square events or Taiwan’s autonomy, to name a few.
The difference with other models is that DeepSeek’s censorship occurs at the chatbot’s online level. If you run the model on your own computer—remember, it is open-source—you won’t encounter these limitations.
Conclusion
The emergence of DeepSeek in the market has triggered a true earthquake in the artificial intelligence sector, threatening the dominance of giants like Nvidia and OpenAI.
Its innovative approach, based on an open-source model and the democratization of access to advanced technologies, has opened new possibilities for researchers and developers worldwide.
Its success could mark the beginning of a new era. The story of DeepSeek is just starting, and its impact on the industry will undoubtedly remain a subject of close monitoring and analysis.
