Nvidia lost half a billion in one day: record
In the last 48 hours, the financial markets have experienced a real earthquake caused by the irruption of DeepSeek, a Chinese startup from artificial intelligence with less than two years of life. This company has surprised the world with the launch of an advanced language model capable of competing directly with the giants of the sector, but developed with a surprisingly low investment.
Nvidia, until recently the jewel in the technological crown and the main engine of artificial intelligence hardware, has been the hardest hit. In a single day, its shares fell by 16.86%, which is equivalent to a loss of 589,000 million dollars in its stock market value. This figure sets an all-time high as the largest destruction of value of a company in a single day.
This drop not only affects Nvidia, but also calls into question the stability of the big tech names that dominate the artificial intelligence market. Companies that until now seemed unbeatable are seeing how more agile and economical models, such as DeepSeek's, can reconfigure the rules of the game.
Why does this impact go beyond Nvidia?
Although Nvidia is the most prominent case, the impact of DeepSeek has shaken all the big tech companies. Collectively, the sector has lost more than $1.5 trillion in market value, reflecting widespread uncertainty among investors. Surprisingly, Apple, considered until now one of the least committed to the artificial intelligence race, saw a 3% rise in its shares. This contrast suggests that investors are starting to price in other strategies that are less dependent on heavy infrastructure.
In this context, Nvidia, which was once worth more than Apple, faces a double challenge: justifying the need for its specialized hardware against cheaper alternatives and adapting to increasingly agile competition. This crash not only highlights Nvidia's vulnerabilities, but also marks the beginning of a potential restructuring in the tech sector.
DeepSeek has opened the door to a future in which artificial intelligence might not necessarily be led by the largest companies or the most powerful models, but by those who are able to do more with less.
Who is DeepSeek and what does it bring back to artificial intelligence?
DeepSeek is the name that has been on the lips of all artificial intelligence experts this week. Although this Chinese startup is barely two years old, it has already shaken up the game board. Its recent model, DeepSeek-V3, has proven to be as capable as solutions from giants like OpenAI, but with one key difference: it has done so at a fraction of the cost and with a completely different philosophy.
If we were to compare artificial intelligence with the automotive industry, we could say that DeepSeek has not manufactured a more expensive and powerful luxury car, but an efficient, fast and accessible electric vehicle that is beginning to conquer the roads. All this while allowing other engineers to inspect its engine and use its innovations to create their own cars.
An open source approach: the big differentiator
While OpenAI, despite its name, keeps the doors closed on its models and advances, DeepSeek has taken the opposite path. Its model is open source , which means that any company or researcher can study it, adapt it and use it. This approach fosters collaboration and accelerates innovation in the sector. It's as if DeepSeek is inviting other companies to join its mission instead of competing against them.
This approach raises an important question for business leaders: what does this shift toward openness mean for the industry? On the one hand, it encourages healthier and more dynamic competition; on the other, it forces traditional companies to rethink their intellectual property protection strategies.
What makes DeepSeek so special?
The success of DeepSeek is no coincidence. Their model uses an innovative architecture called Mixture of Experts (MoE) . To understand this, imagine an orchestra where each musician is an expert in an instrument. Instead of everyone playing at once, the DeepSeek orchestra activates only the musicians that are needed for a specific piece. Not only does this approach save resources, but it allows each note (or task) to be executed accurately and efficiently.
In addition, DeepSeek has implemented techniques to reduce the energy consumption and training costs of its models. In business terms, this means they can deliver powerful artificial intelligence without needing the huge investments that companies like Nvidia and OpenAI require to maintain their operations.
Expert opinions
Experts stress that DeepSeek is not just a "new player" in the sector; It is a model that challenges traditional assumptions. According to some analysts, DeepSeek's efficiency could usher in a new era in artificial intelligence, one where it's not just giants with deep pockets who can lead.
However, there is no shortage of skeptical voices. Some argue that the real battle in artificial intelligence is not just in the performance of the model, but in the amount of users and data that a system can accumulate over time. In that sense, companies like OpenAI, which have millions of users training their models daily, have an advantage that is difficult to match.
What this means for us
For business leaders, the case of DeepSeek raises key lessons:
- Efficiency matters as much as innovation. DeepSeek has shown that it is possible to compete with the giants by being more agile and reducing costs.
- Collaboration can be more powerful than secrecy. DeepSeek's open-source approach could inspire other industries to adopt more transparent models.
- The game is not always won by the greatest. In an ever-changing industry, adaptability and disruptive thinking can outweigh financial resources.
In short, DeepSeek isn't just changing the way we look at artificial intelligence; It's teaching companies to think differently. This is an inflection moment for the industry, and understanding how to respond to these changes will be key to staying competitive in the years ahead.
Beyond euphoria and pessimism
In the dynamic world of artificial intelligence, it's easy to get swept up in shocking news and market fluctuations. However, at Proportione we believe that it is essential to maintain a balanced and analytical perspective.
The recent emergence of DeepSeek has generated both excitement and concern. While its innovative and open-source approach is worthy of attention, we must not forget that success in the field of artificial intelligence does not depend solely on the power or novelty of a model. A crucial factor is the amount of data and the diversity of users that contribute to the training of these systems. In this sense, platforms such as OpenAI's ChatGPT have built a significant advantage by being continuously refined by a wide global user base.
In addition, it is important to note that artificial intelligence is evolving beyond traditional language models. An example of this is " Operator ", the artificial intelligence agent recently launched by OpenAI. This agent is designed to automate web tasks, interacting with on-screen elements such as buttons and text fields, which represents a step towards more practical and autonomous applications of AI.
On the other hand, recent fluctuations in the stock market, such as the notable drop in the value of Nvidia, can be influenced by multiple factors. Among them, President Donald Trump's announcement of a private investment of up to $500 billion in artificial intelligence infrastructure in collaboration with companies such as OpenAI, Oracle and SoftBank.
In conclusion, while innovations like DeepSeek are exciting and bring new perspectives, it's critical not to succumb to rash reactions. True transformation in the business world will come from practical applications of artificial intelligence, such as autonomous agents, which have the potential to redefine processes and improve efficiency. Maintaining a balanced vision focused on real value for businesses and people will be key to navigating this ever-changing landscape.
Added January 29, 2025: Growing doubts about DeepSeek's consumption, efficiency, and quality
In the last few hours, new analyses have emerged that cast doubt on the initial claims about the energy efficiency, the number of chips used and the actual performance of the DeepSeek artificial intelligence model. What was initially touted as a technological feat—with extremely low power consumption and surprisingly low GPU usage—now appears to be, at best, a Optimistic exaggeration , and at worst, an attempt to Hide key information about its infrastructure and performance.
Is it really as efficient as it is claimed?
Since DeepSeek's announcement, it has been repeated that its training required less than 10% of computational resources used by models such as Meta Flame . However, several AI experts have expressed skepticism about this claim. According to the researcher Martin Vechev of INSAIT , the $6 million training cost is "misleading" and probably does not reflect actual spending in hardware, energy and model optimization.
Moreover, recent reports from the International Energy Agency (IEA) warn that the information on DeepSeek's electricity consumption is Too opaque as if to draw definitive conclusions. While the company has insisted on its efficiency, it has not provided verifiable data on actual energy usage in its data centers.
Another critical point is the number of GPUs Nvidia H800 allegedly used in DeepSeek training. The Chinese company claimed to have used 2,048 units , a figure that, if true, would imply revolutionary computational efficiency. However, industry experts have pointed out that Training a model of this caliber with so few chips is highly unlikely , unless extremely aggressive compression and optimization techniques have been applied, which could affect the quality of your responses.
This raises a key question: was the Reasoning ability of the model to achieve these supposed levels of efficiency?
Quality of responses and strong ideological bias
Beyond the doubts about its infrastructure, the first tests with DeepSeek have revealed Significant deficiencies in the quality of their responses . Unlike models such as OpenAI's ChatGPT , DeepSeek has shown:
- Less detailed and less nuanced answers on complex issues.
- Difficulty handling open consultations , generating vague or imprecise results.
- A marked ideological bias , especially on issues related to China, geopolitics and human rights .
Media such as The New Indian Express have documented how DeepSeek deliberately avoids certain questions , especially those that could be sensitive to the Chinese government. In some cases, the answers seem designed to align with the official narrative of the Chinese Communist Party , which raises doubts about its impartiality and its usefulness as a general-purpose AI.
As DeepSeek's capabilities are further analyzed, it becomes increasingly apparent that Their accomplishments might have been overestimated . The lack of transparency about its energy consumption and hardware, added to the questionable quality of their answers and their obvious ideological bias , suggests that we are dealing with a product that still not up to industry leaders , such as ChatGPT or Claude.
If DeepSeek wants to position itself as a serious competitor in the global AI market, it will need to demonstrate with verifiable data that its efficiency is real, that its number of GPUs has not been underestimated, and that it can offer unbiased, high-quality answers. Until then, it will continue to be viewed with skepticism by the tech community.