When, over the past twenty, thirty or perhaps fifty years, we thought of cutting-edge films, songs or products, we pictured the United States.
In a sense, without knowing the country or ever setting foot there (until not long ago, going there was a privilege for the few), we all felt we somehow knew the States.
That subtle influence seeping into culture and the collective imagination is called soft power. And here is the point: today it is China that is building the same form of attraction, albeit in a different way.
More than through films and entertainment (areas in which Beijing has, in any case, been investing heavily for years), it is in Artificial Intelligence that a large share of global influence is being contested today.
Consider this: a language model that costs less can be downloaded, modified and installed by a developer in Indonesia, a bank in Kenya or a public administration in Brazil. Its value grows exponentially every time someone manages to use it without requesting access from OpenAI, Anthropic or Google.
Admittedly, at first glance this might look like an industrial strategy, but it is geopolitics. The pure geopolitics of modernity.
On 17 July 2026, at the World Artificial Intelligence Conference in Shanghai, Xi Jinping described the opening up of models as a rare historic opportunity. He presented global collaboration as the route for distributing the benefits of artificial intelligence and pointed to open systems as an alternative to dependence on the large American technology companies.
A few days later, an article in the New York Times gave the project a name: soft power, precisely.
China is offering the world more accessible artificial intelligence. Meanwhile, it is building the conditions for companies, developers and entire countries to start working within its ecosystem.
And this, in the long run, will change the world. Let us see how.
Soft power you can download
Soft power works when influence is perceived as desirable.
Hollywood made a way of life desirable. American universities turned education into prestige. Silicon Valley spread tools that millions of people adopted before they even understood the power accumulated by the companies that made them.
As one might imagine, the Chinese strategy starts from the same principle, but applies it to an asset suited to the digital economy: AI infrastructure.
DeepSeek, Qwen, Kimi and GLM can be adapted, run locally and integrated into products built by others. The advantage for those who adopt them is immediate. Costs fall, dependence on proprietary APIs decreases and the scope for customisation increases.
According to estimates gathered by the New York Times, for some tasks Chinese alternatives can cost between 50 and 90 per cent less than American models. A difference that transforms a developer’s prospects, determining whether or not they can enter a given segment of the market.
And this has consequences that are not only practical, but emotional: gratitude.
Yes, because when a technology becomes accessible to those who were previously excluded from it, it is only natural that the country that made it available should gain gratitude, familiarity and, who knows, perhaps even a privileged position in any future discussion.
Pure soft power.
Necessity becomes strategy
China’s race towards efficient models stems from a constraint. US restrictions on the export of the most advanced chips have reduced Chinese companies’ access to the computing power used by American laboratories. The response came through lighter architectures, better optimisation and greater attention to the cost of training models.
In practice, the Chinese chose to make a virtue of necessity, creating a strategy that makes efficiency the core value of their model.
And so, while OpenAI and Anthropic keep improving their flagship products, making them ever more cutting-edge (and expensive), Beijing is advancing on diffusion and integration.
According to Brookings, the United States still holds an advantage in computing power and frontier performance, but China is accelerating through efficiency, distribution and application. And if it keeps doing so, it will soon become the standard.
Low cost breeds habit
Technology adoption creates habits.
A developer who learns to build on Qwen will go on using the libraries, documentation and tools that have grown up around Qwen. A company that trains its processes on DeepSeek will accumulate data, skills and procedures compatible with that architecture. And so, even if switching remains possible, over time it becomes ever more costly.
This inertia holds an extraordinary power: every choice made today makes tomorrow’s more likely.
The signs are already visible. In 2025 the number of Chinese organisations active on Hugging Face grew rapidly following the international success of DeepSeek. Baidu went from no publications in 2024 to more than a hundred repositories in 2025. ByteDance and Tencent multiplied their releases.
By mid-2025, Qwen had already become the basis for more than 113,000 derivative models on Hugging Face. According to an analysis by the platform itself, that figure was many times higher than the number of derivatives built on Llama and DeepSeek.
Every derivative represents someone who decided to build their own product on a Chinese foundation.
And that is how influence accumulates. One technical choice at a time.
The Global South is the real playing field
In his Shanghai speech, Xi Jinping announced that over the next five years China will offer developing countries 5,000 training opportunities and seminars on artificial intelligence.
The plan provides for cooperation centres with ASEAN, the Arab League, the African Union, the Community of Latin American and Caribbean States, the Shanghai Cooperation Organisation and BRICS. Thirty countries will then receive access to the MAZU weather system for forecasting and early warning of extreme events.
The same event saw the launch of the World Artificial Intelligence Cooperation Organization. Twenty-nine countries joined as founding members of a body created to promote development, training and shared governance principles.
These programmes respond to a real need. Many countries have researchers, businesses and problems that artificial intelligence could help to solve. What they lack, however, is (i) stable access to computing power, (ii) capital comparable to that available in the United States and (iii) models suited to local languages and conditions.
According to Brookings, the Global South is focusing its attention on smaller, efficient models applicable to specific contexts. Exactly the kind that China is creating. One model at a time, one small strategic move after another, China is steadily consolidating its global sphere of influence.
What changes for European companies
For businesses, choosing a model is becoming a strategic decision.
Cost advantages may already be pushing towards a Chinese system, even though the decision carries hidden risks.
A model enters workflows, shapes content, takes part in decisions and can become part of essential infrastructure. Its licence may change. Support may be discontinued. Geopolitical tensions may restrict updates, supplies or collaborations. The model’s origin may become a reputational issue in certain markets.
This is why adoption should go through an AI Governance framework able to assess costs, results, security, compliance, continuity and dependencies before integration.
After all, a model may be cheap today, but what would it cost in three years’ time?
Answering this question requires a map of the tools, the data transferred, the dependent processes and the skills developed in-house. All things that are rarely considered until it is too late.
What you don’t pay today, you will pay tomorrow
American soft power asked the world to desire America.
China’s, for now, simply makes its technology convenient. But this is only the first phase.
The rest will come with time: skills, standards, ecosystems, diplomatic relations and a growing difficulty in breaking away from the foundation on which everything has been built.
When a country becomes the source of this normality, its influence lives inside the software others use. China has understood that the future will be controlled not so much by whoever builds the perfect model first, but by whoever succeeds in making artificial intelligence normal for billions of people.
This is also why free access at the outset is only the first step.
The real value will accumulate later, when the chosen model becomes the habitual way in which a company or a country learns to use artificial intelligence.
And that is where access will begin to turn into power.
New connections (FAQ)
Why do Chinese models cost less?
Restrictions on access to the most advanced chips have pushed Chinese companies to invest in architectural efficiency, inference optimisation and models capable of running on fewer resources. Publishing the weights also reduces the margin demanded by a single supplier. Companies and developers can run the model on their own infrastructure or choose among several providers, increasing price competition. The real cost, however, still depends on hardware, energy consumption, customisation and maintenance. A model that is free to download still requires resources to run and people capable of governing it.
Does adopting a Chinese model create dependence on China?
The availability of the weights allows a company to run the model locally and can reduce operational dependence on the developer. The risk grows when processes, data, applications and in-house expertise are built around a single family of models. Dependence also concerns the frequency of updates, compatibility with tools, licences and the organisation’s ability to replace the system without disrupting work. An exit plan, interoperable formats and a periodic assessment of alternatives make it possible to retain the cost advantage without turning a convenient choice into a permanent constraint.
How should a company choose between an open model and a closed one?
The choice should start from the business process and the level of risk. Open models offer customisation, control over infrastructure and potentially lower costs. Closed systems can provide simpler integration, centralised support and vendor-managed updates.
Data sensitivity, financial return, required quality, regulatory compliance, the availability of in-house expertise and the cost of a future migration all need to be assessed. AI Governance turns these elements into criteria applicable to every new adoption. In this way, the choice of model follows the organisation’s strategy rather than depending on the most visible novelty of the moment.
BLISS®
© 2026. ALL RIGHTS RESERVED