DeepSeek and the end of the era of expensive AI models
For the past two years, the global capital market has operated under an almost dogmatic model of “AI exclusivity.” Investors, tempted by the vision of the digital revolution, paid a huge premium for the shortage without blinking an eye, accepting the limited supply of chips, the astronomical costs of training models and the technological gap separating Silicon Valley from the rest of the world.
– However, behind this hype there is another, much more dangerous risk – not macroeconomic, but structural. It is not born in Washington or Frankfurt, but in Shenzhen, Beijing and Shanghai – says Marek Górniak, Network Development Manager at Freedom24.
This is where a new reality is being forged, in which artificial intelligence ceases to be a luxury good and becomes a mass tool with a low entry threshold.
The breakthrough came with the debut of Chinese models such as DeepSeek, which shook the foundations of Silicon Valley at the end of 2023. The shock did not result from the fact that the Chinese had created something qualitatively better than OpenAI or Google solutions, but from how drastically they reduced production costs. The American model of AI industry development is based on brute force: more graphics processing units (GPUs) mean more powerful data centers, which creates a high barrier to entry and allows for sky-high margins. The Chinese have proven that there is an alternative path. Instead of adding thousands of integrated circuits, they focused on deep optimization of the algorithm architecture, which resulted in fewer computational operations and drastic price cuts. As a result, the long-term viability of the entire Western AI infrastructure chain has come into question.
Why China is winning over the USA
Although we usually think of artificial intelligence in terms of immaterial algorithms, in reality this sector has become a modern version of heavy industry, the foundation of which is not steel, but electricity. The scale of demand for electricity is gigantic: training one large language model requires consumption of from 10 to even 30 million kWh of electricity. In this context, the competition for dominance in the field of AI is moving from research laboratories to the field of pure energy utility economics.
– China has a structural cost advantage in this field, which means it can compete on price, not quality. This does not change the technology, only the investment logic of the entire sector – explains Marek Górniak.
This paradigm shift means that existing models of valuing technology companies, based on pure innovation, must be verified through the prism of operating costs.
Analyzes of the International Energy Agency (IEA) and the national energy agencies of China and the USA (as of 2025) reveal the deep cost disproportion between the superpowers. The estimated cost of training one model in the Middle Kingdom ranges from $0.6 to $2.4 million. For comparison, the average for the United States is almost twice as high and ranges from $1.2 to $4.8 million. The situation is even worse in American technology hubs, such as Texas or Arizona, where due to the specificity of local energy markets, this cost may increase up to USD 6.6 million. This is a gap that allows Chinese entities to aggressively expand prices.
– There is a tired cliché that says that history repeats itself. China took it to heart because it had used the same scheme before. The mechanism is the same every time: a new technology is created that gives hope for high margins, China adapts the idea, increases production, often developing technology or processes, which translates into a faster drop in prices compared to expectations and kills the excess profits of companies creating the infrastructure base – explains Marek Górniak.
Will Nvidia share Cisco’s fate?
The current situation in financial markets bears a striking resemblance to the euphoria at the turn of the century, when investors believed that hardware manufacturers would be the ones dealing the cards in the digital economy for life. In 1999, the symbol of this domination was the Cisco concern, providing infrastructure for the construction of the Internet, just as today Nvidia is perceived as the irreplaceable foundation of the AI revolution. However, Marek Górniak warns against repeating the mistakes from a quarter of a century ago.
– The current situation of the AI market resembles that of the Internet bubble in the 2000s. In 1999, it was believed that hardware would take the lion’s share of the market. However, history shows that as soon as a technology becomes popular, the main profits shift from infrastructure producers to those who create the final product based on it – says the Freedom24 expert.
Today’s valuations of sector leaders, led by Nvidia, are in fact a bet that the state of permanent shortage of computing power will last for years to come. Chinese script “artificial intelligence under the roof” brutally tests these assumptions, turning deficit into abundance. In such a system, we do not have to witness the collapse of technology itself. The business can still generate revenues and the technology can develop, but the share price will fall, adapting to the new, much lower margins of the sector, forced by Chinese competition.
Beneficiaries of cheap artificial intelligence
If predictions about a sharp reduction in the costs of training and operating AI models come true, we will face a fundamental rotation of capital on global markets. In a “cheap AI” scenario, economic value will inevitably shift from those who perform the calculations to those who deliver the finished result and real business value to the end user. The greatest growth potential lies in the service sectors, SaaS platforms and vertical solutions in industries such as medicine, logistics and finance. For them, cheaper computing resources mean a step-change in unit economics: margins increase and the barrier to entry for the mass customer drops dramatically.
– Perhaps the biggest risk in today’s market is not that AI will be overvalued as a technology, but that it will be too effective and too cheap. And then it will not be artificial intelligence itself that will be overvalued, but investors’ expectations regarding its beneficiaries – summarizes Marek Górniak.
About Freedom24
Freedom24 is a stock broker connecting European clients with major international capital markets. Using its own platform, it offers investors access to a wide range of stocks, ETFs, corporate and government bonds and US stock options across US, European and Asian markets.
Freedom24 is CySEC licensed and operates within the MiFID II regulatory framework, serving all EU/EEA countries, with a physical presence in 10 countries. Freedom24 is the European subsidiary of Freedom Holding Corp., an international financial group operating in the United States, Europe and Central Asia, listed on the Nasdaq stock exchange. The Freedom24 platform is available to Polish users from 2021, and the representative office in Warsaw offers them support in Polish.
More information at https://freedom24.com/
Reservation:Your capital is at risk. Past performance is no guarantee of future results and there is no guarantee that any investment strategy will achieve its objectives. Before making an investment decision, you should analyze the situation yourself and consider consulting a qualified financial advisor. Access to certain financial instruments may require a suitability test. The Company may participate in trading in the instruments described in the material as a transaction party or liquidity provider. At the same time, the recommendations and information presented are independent and not related to its positions
