The world of artificial intelligence (AI) has entered a new phase of global competition. While the United States maintains restrictions on the export of cutting-edge technology to China, the Asian giant seems to be doing more with less. With the launch of Deepseek R1 just over a week ago, it did better than other intelligences such as Llama and the famous ChatGPT. Deepseek, despite having been created with NVIDIA chips of lower quality than those used by its American peers, is an open source AI with a development cost significantly lower than ChatGPT, which has put Silicon Valley' s technological benchmarks in check and has accelerated the race for the development of artificial intelligence.
NVIDIA and Deepseek: the Goliath and the David of AI
NVIDIA is a leading developer of graphics chips and artificial intelligence processing units, playing a crucial role in the evolution of global technology. Its microchips have been fundamental in the development of modern AI, enabling complex models to be trained thanks to their high computational power. Demand for these chips has grown exponentially in recent years, with large companies and governments competing to secure a stable supply.
NVIDIA's dominance in the industry is due to its ability to build high-performance hardware that optimizes machine learning and simulation tasks. However, with the emergence of models such as Deepseek, which have demonstrated efficiency with more limited hardware, reliance on its chips could be in question. Given this uncertain outlook for NVIDIA and other U.S. technology companies, the U.S. government's response was swift.
As early as the end of 2022, the U.S. had already imposed tough restrictions domestically and internationally on the export of high-end microchips, such as those from NVIDIA, to Chinese companies. These chips, particularly the H100 models, are essential for training advanced artificial intelligence systems such as ChatGPT due to their high performance in parallel computation, optimization of deep learning models, and efficiency in processing large volumes of data. Unlike less powerful chips, such as the A800 models that China has used as an alternative, the H100s enable significant reductions in the time and cost of training complex models, making them a key tool in the race for AI dominance. However, China seems to have found a way around these restrictions: according to Reutersreports , it is secretly importing NVIDIA's H100 units from Singapore, thus defying the veto imposed by the United States.
In the face of the recent drop in the value of tech stocks, with NVIDIA suffering the biggest loss of assets for a company in history (18% of its market value), a key question has arisen: are NVIDIA's chips really essential to lead the AI race? The launch of Deepseek has challenged this premise, demonstrating that it is possible to develop advanced models using less powerful hardware , such as the A800 chips, rather than the H100s used by ChatGPT. While more advanced chips offer advantages in speed and efficiency, Deepseek's success suggests that dependence on cutting-edge technology or chips may not be as absolute as previously thought. If this trend is confirmed, the decline in the company's value could continue, reflecting a change in the market's perception of the need for its products in artificial intelligence development.
The "Stargate Initiative": the U.S. response
With the recent inauguration of Donald Trump as President of the United States, government policy on artificial intelligence has taken a significant turn. Trump has announced the ambitious "Stargate Initiative", a $500 billion program aimed at boosting AI development in the United States. This project seeks to ensure that the country maintains its leadership in emerging technologies by securing massive funding for companies in the sector, including NVIDIA, OpenAI and Google DeepMind. The Stargate Initiative involves not only direct financial support, but also tax incentives and priority access to key infrastructure, which could represent a strategic advantage over the Chinese advance.
The effectiveness of this initiative will depend on how it is implemented and whether it succeeds in curbing the growing influence of open source AI models such as Deepseek. Deepseek's open nature raises questions about the future of AI competition. If companies can use its code to improve their own models, they could reduce their dependence on OpenAI and other U.S. companies. However, disadvantages also arise. An open source model could make it easier for malicious actors to access advanced technologies without restrictions.
Deepseek R1 represents an attractive alternative to closed models such as ChatGPT. Its open source allows AI companies to enhance and adapt their systems without restrictions. In addition, its lower development cost and free usage make it a more accessible option for researchers and developers. Compared to ChatGPT, Deepseek has key differences in its training. While OpenAI uses supervised fine-tuning, R1 relies on reinforcement learning, an approach that could offer improvements in certain types of tasks. This model has also proven to be more economical: the cost of its first version, R1, was significantly lower than that of ChatGPT, which could attract more investors and users in the future.
Biases and narrative control in AI
On the other hand, the emergence of Deepseek triggered multiple discussions in social networks on the question of the biases inherent in each model. As an example, while ChatGPT has been accused of having a Westernist and white supremacist bias in its responses, Deepseek has been criticized for omitting information that attacks the image of the Chinese government. These biases not only reflect cultural and political differences among AI developers, but also raise questions about neutrality and ethics in AI development.
An example of this situation can be seen in the way ChatGPT describes Deepseek. According to a response obtained from their platform:
"Bias in artificial intelligence is not accidental, but the result of the data on which these models are trained and the design decisions made by their developers". OpenAI, the company behind ChatGPT, has implemented filters and rules to avoid misinformation and hate speech, but this has led to accusations of favoring certain political narratives over others. On the other hand, Deepseek, developed in a context of increased state control, is designed to align with Chinese government guidelines, which could limit its ability to provide unbiased responses on sensitive topics such as human rights, democracy or international conflicts."
Interestingly, upon entering this input to Deepseek the AI responds to us, "Deepseek, as a model developed in China, is designed to comply with Chinese government regulations and guidelines. This includes avoiding content that may be considered sensitive or contrary to Chinese Communist Party policies. As a result, the model is likely to have limitations on topics such as human rights, democracy or international conflicts, especially if these topics conflict with official narratives."
The bias in artificial intelligence is not accidental, but a reflection of the priorities and constraints imposed by the entities controlling its development. AIs not only generate answers based on data, but also on the perspective of their developers. While OpenAI may face criticism for its focus on content moderation, Deepseek is subject to a stricter regulatory framework given its Chinese provenance, which may affect its ability to address certain issues in a neutral manner.
Artificial intelligence and global balance
The dynamics at issue pose a fundamental challenge for users and businesses that rely on these technologies. The perception of impartiality is crucial to trust in AI, and as these models become integrated into decision-making processes and information dissemination, the need to develop more transparent mechanisms to identify and mitigate these biases becomes increasingly apparent. In this context, competition between AI models is not only a question of accuracy or efficiency, but also of control over the overall narrative. This control not only affects the perception of AI, but also has direct geopolitical implications. In a world where information is power, AI supremacy can redefine the global balance.
The U.S.-China trade war has moved from hardware to software. With the acceleration of AI development in China, the question is no longer whether the Asian country will catch up with the West, but how close it is to overtaking it. Time is running out for OpenAI and other companies trying to consolidate their lead. The longer they take to establish themselves as dominant companies, the faster investment will flee to more accessible and efficient alternatives. The AI revolution is already underway and, rather than a bursting of a market bubble, we seem to be witnessing an acceleration in the technology race for global supremacy.
History has shown that it is not always those who lead in the early stages of a technological competition who end up prevailing. The Soviet Union dominated the early stages of the space race, but it was the United States that reached the moon first. With its vast investment capacity and innovation ecosystem, it is too early to consider the United States defeated. The competition for artificial intelligence is no longer just technological, but also strategic, geopolitical and hegemonic: will the United States be able to maintain its leadership, or will China take the lead in the age of artificial intelligence?
As final thanks, I think it is essential to highlight the work of my colleague, friend and co-author of this article, Gerónimo Becker who, from his position as a student of Systems Engineering at the National Technological University (UTN), provided the most important technical concepts when talking about the particularities of each AI.
Comments