Chinese Artificial Intelligence: strategy, competition, and practical lessons

There is an old Chinese proverb that goes like this: “May it not be your turn to live in interesting times.I hope you don’t have to live in interesting times“. It is a kind wish that contains a veiled warning. Interesting times will mean having to navigate through times of profound change, with great balances being broken, certainties being blurred, and the obligation to develop a greater tolerance for danger. These are times of uncertainty, transition and risk.

Antonio Gramsci around 1930 in his Quaderni del carcere left written another warning that today we cannot overlook: “The old world is dying. The new one is slow to appear. And in that chiaroscuro the monsters emerge. “. Although the Sardinian thinker may have had other kinds of monsters in mind at the time, the truth is that the idea he conveyed in his work enjoys the gift of universality and will remain valid for many years to come. Moments of change, whether a technological revolution, the disruption of existing geopolitical hegemonies, or both, bring upheaval. I suppose to the reader in mid-2025 the description sounds familiar at the very least.

The widespread diffusion of artificial intelligence has amplified geopolitical dynamics that were already underway years ago: the trend towards a more multipolar world necessarily leads to technological competition between power blocs. We are navigating amidst symptoms of a de-globalization where different powers are building their own digital ecosystems and we should not take for granted that these will be interoperable in the future. We know where we are today but we cannot be sure that tomorrow will look too much like what we are seeing today. Two clear examples of this are Google’s situation in China and that of Tiktok in the United States.. The Internet also has its frontiers, and they are likely to grow in the coming years.

China is not an anomaly in this global context. Nor is it any of those monsters that appear in the chiaroscuro of history. It is one of the central players in the technological process, not only because of its economic or demographic weight, but because it has formulated with clarity and determination a project of its own.based on self-sufficiency, internal scale, strategic control of data and a strong alignment between industrial policy and geopolitical ambition. We tend to live with our backs turned to this obvious reality, but this attitude leads us to miss opportunities and suffer from a partial understanding of the world around us. The digital society facilitates the creation of personal information bubbles that isolate us. It is good to be aware of this and make an effort to try to see the whole picture.

There is a lot of literature on the subject and what we can focus on here is how we can benefit from the tools we have available today. Beyond ChatGPT or Claude there are models that are worth knowing, testing and taking advantage of as much as we can.

The arrival of Deepseek

The model Deepseek-V2 model was launched publicly in December 2023. It was presented as a significant evolution over its previous version, with remarkable improvements in mathematical reasoning, technical understanding and outstanding multilingual capabilities. Anecdotally or not, I was able to speak with this LLM in perfect Catalan from the very first moment and other western models with a large market share in Europe, such as Gemini, took months to be able to work in that language in a more or less decent way.

Although it is not as popular in our environment as other models such as OpenAI, Google or Anthropic, it is undoubtedly a competitive alternative in specialized fields such as engineering or programming. It has been well received in technical and academic communities. It should be added that it is a model that helps to bring cultural contrast to the literature generated in collaboration with the more mass dissemination models. It is therefore a useful tool for contrasting political and cultural biases. contrast the political and cultural biases of other models.. In other words, it is a good idea to test the correctness of texts generated with ChatGPT against Deepseek.

One of the aspects that has attracted the attention of Western audiences is its performance in solving tasks that require step-by-step reasoning, technical translations or detailed explanations of complex processes. This makes it especially interesting for professionals in industry, engineering, technical maintenance, in-house training or process documentation. Experience tells us that this is not an LLM designed to compete in creative storytelling or to chat in a colloquial tone. Its value lies primarily in its ability to rigorously answer questions that are not open to digression..

The model also stands out in terms of cost-performance ratio, both in its public versions and in its commercial models. Its launch caused NVIDIA’s stock to fall by 17% on January 27, 2025, or in other words, a reduction in capitalization of 589 billion US dollars. It didn’t take the hardware company too long to recover to its previous trading level, but everyone took note of the warning in the logbook..

Qwen: integration, scalability and corporate purpose

Another model that deserves attention is Qwendeveloped by Alibaba Cloud and released in different versions starting in 2023. It must be said that this is not simply an LLM: it is part of a much broader strategy by Alibaba to consolidate its technology ecosystem by integrating artificial intelligence at all levels of its digital infrastructure. From e-commerce to cloud computing.

Qwen is an enterprise foundational model. This means that it not only seeks precision or efficiency, but has been conceived to scale in volume, adapt to multiple use cases and fit as one more gear within a service architecture. One of its differentiating elements is its multimodal approachIt is not limited to text, but has been designed to work also with images, complex instructions and tasks that combine different types of input data. This orientation makes it an attractive LLM for industries where computer vision, visual technical documentation or process analysis are important.

In the European context, Qwen may not seem as obvious a choice as Deepseek. However, its accessibility, adjustability and multimodal approach make it an interesting LLM for integrating R&D environments, multilingual technical documentation or processes requiring text and image processing. In addition, for more editorial or research tasks, it is a model that can also accompany us in our mission to bring cultural contrast to our writing and avoid the biases of Silicon Valley models.

An ecosystem that continues to grow

The map of Chinese artificial intelligence is not limited to Deepseek and Qwen. In recent times, multiple projects have appeared with different characteristics but sharing a common goal: to provide the country with technological autonomy in one of the most strategic development fields of the 21st century. Some of these models are already publicly available, while others are being developed in more restricted environments, but they are all part of a rapidly growing ecosystem of tools.

An example of this is InternLMa model developed by Shanghai Artificial Intelligence Laboratory and recognized as one of the most ambitious models in terms of scientific scope and long-term vision. Since its inception, it has been supported by the country’s public institutions and major Chinese industrial players. The vision behind the project is very clear: to build a “made in China” alternative to the big Western LLMs. The performance of its latest versions has pleasantly surprised analysts and its architecture stands out for being oriented towards extreme efficiency in terms of energy consumption.

It is also worth mentioning Yi, a tool developed by the company 01.AI by former Google ex-Google Kai-Fu Lee. It burst onto the scene in 2023 with a bang, offering high-quality results and a strong focus on machine translation and multilingual use. It is a reference for developers and teams looking to create lightweight and well-documented models. It does not pretend a priori to compete with anyone in power, its ambition lies in becoming a practical, scalable and accessible development tool.

Finally, it is worthwhile not to forget about Baichuan. Its positioning has been clear from the outset: it is a model that aims to meet the needs of companies looking for ready-to-integrate artificial intelligence solutions. It has launched several versions optimized for fast inference, which have achieved competitive performance and good compatibility with production environments. Beyond what we can know from the technical side, the company has been able to weave a good policy of strategic alliances that have led it to collaborate with universities, startups and relevant cloud computing players. It is therefore a model focused on developing its full potential in the field of sector adaptability.

Is it worth trying Chinese artificial intelligence tools?

Undoubtedly, experimenting with Chinese models is worthwhile. First of all, because we will discover that we are paying for licenses of western models when for common and not very intensive use cases, we could work with Chinese alternatives for free. We are not in a static environment and it is important to take our time to adopt solutions that make sense, but we must be clear that if we learn where to go in each case, the world of LLMs is offering us many opportunities and not all of them are paid. Without going too far, Qwen offers us the possibility of generating images of more than acceptable quality at a speed that is currently not within the reach of Chat GPT itself.

Secondly, it is worth keeping a curious attitude at a time of such effervescence. Looking only to one side of the technological development means, at the very least, missing half the story and being unaware of the existence of resources that can be of help in our day-to-day professional lives.

To conclude: China’s economy is already the largest in the world.. It makes no sense to ignore this reality and refuse to understand the profound repercussions this situation will have in the long term. We do not know whether this circumstance will fit in with the globalization model we have known so far or whether it will be the trigger for a new era of more isolated power blocs, but we can be sure that looking the other way and pretending that nothing is happening is not an intelligent attitude.