“Whoever wins in artificial intelligence, wins.” With these words, American President Donald Trump responded last September to the calls of some of the big tech bosses in Silicon Valley to slow down the dizzying development of AI to focus on security. The president of the United States (in the photo near the title, in a moment of prayer) is also among the main private investors. Between fear of the apocalypse and almost palingenetic promises for the radical change of the world, the sector is perhaps the biggest bet of the 21st century. However, what is often ignored is the economic stakes at stake, which could decree the rise or decline of entire nations and production models.
Record investments in AI and the risk of a new bubble
The American race for artificial intelligence now mobilizes enormous capital. According to the Ai Index 2026, the annual report from Stanford University, between 2013 and 2025 US companies in the sector attracted around 757 billion dollars in private investments, of which almost 286 billion in 2025 alone. Another figure gives the measure of the infrastructural effort: according to the Financial TimesFrom 2023 to June 2026, Amazon, Microsoft, Google, and Meta incurred approximately $1.1 trillion in capital expenditures. These are global investments, also intended for the cloud and other activities.
The two figures cannot be added, because they measure funding and expenses that can overlap, but they show the economic dimension of the American bet on AI.
These impressive numbers are accompanied by an essential question: when and how much will this money yield? The earnings must justify the capital employed, also covering the costs of operating and updating the infrastructure. If subscriptions, business services and new applications do not produce sufficient profits, investor expectations could fall, dragging down company valuations and stock market prices.
Apocalypse or marketing? What’s true about the AI alarms – by Daniele Tempera
The precedent is the dot-com bubble, which burst at the beginning of the 2000s. The Internet would indeed transform the economy, but many companies into which investor money had poured in still lacked solid business models to deliver on their promises. And from the digital world the crisis also had repercussions on the real economy.
A risk that is still present today, as Giuliano Noci, professor of Management Engineering and vice-rector for the Chinese territorial hub of the Polytechnic University of Milan, recalls: “The expectation created by artificial intelligence is excessive, because as has happened with other innovations such as electricity and railways, for example, time is needed to digest the transformative factor of this new technology”.
And the repercussions can be serious: “The big tech giants have announced investments that cannot be repaid. Furthermore, there is no supply chain that supports the growth of data centers that they have planned. The risk of a bubble is real”, underlines Noci. Also because to complicate the picture there is also the challenge coming from the Far East.
More flexibility and attention to companies: the Chinese challenge
The Chinese challenge also became evident to the general Western public with the advent of DeepSeek. What was called into question was the idea that increasingly large investments were needed to develop competitive AI and that “low cost” models could not emerge in this sector too. The immediate reaction was a plunge in technology stocks: on January 27, 2025, Nvidia, the leader in AI chips, lost approximately $593 billion in market value in a single session. The collapse was then partially recovered, but the question remains: how much can American AI-related companies earn if services are available at increasingly lower prices?
The artificial intelligence war that Europe has already lost – by Andrea Falla
“The American approach is centralized, predominantly power-based and consumer-oriented,” underlines Giuliano Noci. Beijing’s strategy, however, is different: “The Chinese model is not based so much on power, but on sharing and applications and is oriented towards industry and manufacturing”, he adds. Furthermore, China is the largest global market for industrial robots: in 2024 it accounted for 54% of new global installations.
The difference also concerns the distribution of the models. American companies such as OpenAi and Anthropic focus above all on proprietary systems, accessible through online services. Several Chinese developers, including DeepSeek and Alibaba with Qwen, have instead made “open weight” models available: the parameters learned during training are downloadable, allowing other companies to run them on their servers and adapt them to their own applications, within the limits of the license.
This also introduces a distinction on the privacy front. Using a supplier’s online chatbot means entrusting it with the data entered: running the model yourself can instead allow you to keep it in the company infrastructure. Translated: If the system is configured to run entirely locally, data is not sent to the Chinese manufacturer, reducing concerns about access to company and personal information.
What are the major Chinese Ai and how do they work
Today, regardless of strategies, many Chinese models compete with their Western counterparts. But there is a dispute over training methods. Anthropic and OpenAi have accused some Chinese companies, including DeepSeek, of unauthorizedly using their models’ responses to improve their own through “distillation”: the use of one model’s responses as examples to train another. Anthropic reported over 16 million trades with Claude via approximately 24,000 fraudulent accounts. According to American companies, this would allow competitors to benefit from Western investments by reducing their own development costs.
But, beyond the controversies, Beijing seems to have long applied the maxim of Lao Tzu’s Tao Te Ching: “The weak conquers the strong”. Since 2022, Washington has introduced and strengthened restrictions on the export of the most advanced chips and the technologies needed to produce them to China, with the declared aim of preventing their use to strengthen Beijing’s military capabilities. These constraints in China, however, have incentivized the search for greater efficiency in training and the development of national alternatives, with Huawei aiming to compete with Nvidia and AMD in AI chips.
Meanwhile, several Chinese chatbots offer free access to consumers with increased usage limits, complemented in some cases by affordable subscriptions for advanced features.
Among these is Qwen, developed by Alibaba, one of the world leaders in e-commerce. The consumer version offers virtually unlimited free access to basic functions, while access to higher functions starts from 4 euros per month. Also worth mentioning is the aforementioned DeepSeek, supported by the Chinese hedge fund High-Flyer, which offers the chatbot for free on apps and the web and is known for its logical and mathematical reasoning capabilities.
“The consumer product remains a free business card”, the chatbot replies when we question it: the app and site would serve to attract users, create familiarity with the model and feed the ecosystem.
Kimi, the assistant launched by “Moonshot Ai” in 2023 and known for managing long documents and using agents, also offers free access to various subscriptions (pro versions start from 13 euros per month). The revenue game, moreover, does not end with consumer services: it also passes through APIs and the solutions offered to companies.
The issue of costs: this is how Beijing’s strategy challenges Silicon Valley
One of the main revenue channels are APIs, interfaces that allow companies and developers to connect an artificial intelligence model to their programs: for example, a customer service that automatically answers questions or software that summarizes documents. Payment is often based on tokens, the small units into which the system divides the material to be processed: words, parts of words or punctuation marks, for example. Both the text sent and the response generated are counted, generally at different rates. And it is precisely on these tariffs that several Chinese operators offer prices lower than those of their American competitors and their widespread diffusion could constitute a significant problem for overseas big tech companies.
From big tech layoffs to the lower demand for graduates: this is how AI has already begun to change work – by Daniele Tempera
“The cost of a Chinese token at the moment is often equal to a tenth of that of an American one, it is no coincidence that many US companies are also using Chinese models” explains Giuliano Noci. To get an idea of the differences, just look at the table below. Introduction: it is a comparison between list prices, with different performance and service conditions, which does not measure the cost necessary to obtain the same result. But the comparison still gives an idea of the difference.
Another difference concerns the possibility of downloading some models and running them on your own servers, rather than using the manufacturer’s APIs. In this case, according to the conditions of the license, the company can adapt the model to its needs and keep the data in its own infrastructure, without sending it to the supplier. Distributing a model for free does not necessarily generate immediate revenue: those who download it onto their servers may pay nothing to the manufacturer. The commercial challenge is to encourage its diffusion and sell services to those who prefer not to deal with the infrastructure, through APIs, cloud, assistance and customizations.
War and business thanks to AI: what Amodei and Pignataro say and why we should worry – by Daniele Tempera
The effects? “All this can certainly make the return on investment in the West much more complicated and certainly influence the possible explosion of a bubble linked to AI”, observes Giuliano Noci.
And the fear, on the other side of the Atlantic, is perhaps that the race for artificial intelligence will repeat what happened in the electric car, where Chinese competition has put the European industry under pressure. If cheaper Chinese models undermine expected returns from colossal American investments, the consequences could extend far beyond the tech sector, and the crisis could soon move from financial markets to the real economy.
Dossier is the exclusive subscription investigative section of The Vermilion. If you want to support our journalistic work and subscribe, Click here.