Three pillars of neural networks

What the intensification of competition in the artificial intelligence market means

Neural networks have already been laid in the foundation of Belarus’ development for the coming decades. It can even be said that this is ensured at a legislative level. As Belarus’ Prime Minister Roman Golovchenko reported during the plenary session of the Digital Almaty 2025 forum, themed Industrial AI: Technologies for a New Era, Belarus is developing a draft model law on artificial intelligence technologies for the CIS countries. This means we are beginning to establish common approaches to the creation and use of neural networks.

Cost of the issue  

Training neural networks is a very expensive and energy-intensive process. It is no coincidence that the USA, in its pursuit of leadership in this field, is reopening old nuclear power plants and pouring billions into the industry. Just recently, Trump has promised to allocate $500bn for the development of the industry. The project even received the romantic name — Stargate.  

In general, the story of training neural networks rests on three pillars: the data for training, the equipment used for training, and the energy consumed during this training. 

In other words, it all needs to be powered literally by nuclear power plants. However, the equipment is even more expensive than the electricity consumed, and until recently, it was primarily produced by Nvidia, which shamelessly exploited its monopolistic position in pricing.  

New player  

Recently, a new Chinese neural network developed by DeepSeek has been released into the public domain. Anyone interested could try it out for themselves. At first glance, this seems like a fairly ordinary occurrence, as new neural networks are constantly emerging. But as soon as people figured out what they were dealing with, Nvidia’s market capitalisation immediately plummeted by $600bn.
DeepSeek is a direct competitor to the well-known ChatGPT — at least that is how it is being compared. I shall refrain from making any judgements for now. DeepSeek has only just emerged, and there has not been enough time for definitive conclusions. However, in any case, DeepSeek is at least on par with ChatGPT. The difference is that ChatGPT is a closed commercial product, developed over several years with substantial investment. In contrast, DeepSeek was trained in just two months. This was likely made possible by a revolutionary approach to training: DeepSeek learns not only from raw data but also from pre-trained models, which significantly reduces both the cost and duration of the training process. Moreover, this neural network, with some caveats, is capable of running on standard personal computers and even smartphones (not through a smartphone, but directly on a smartphone), which opens up interesting prospects. However, for the USA, this is only part of the problem; the other half lies elsewhere.  


As I have mentioned earlier, Nvidia has been virtually a monopoly in the production of hardware for training neural networks. Notably, the USA has consistently imposed various restrictions on the distribution of this equipment, aiming to prevent the development of competitors. Now, DeepSeek operates on Chinese Huawei Ascend 910Bc chips. This means that no one can restrict China from obtaining the equipment it needs. It is this combination of factors that has led to a collapse in stock prices, accompanied by unprecedented hysteria in Western media. The new Chinese neural network has been accused of everything — from stealing personal data to being based on stolen American technology, with claims that using it is unethical. Applications for interacting with DeepSeek were removed from markets, reinstated, and then removed again. The developers have faced a torrent of absurd accusations, which shows no signs of abating. It seems unlikely that this will subside anytime soon. After such a blow, Americans will take a long time to recover.

This is not just a slap in the face for the USA from China; it is a sudden blow to the face with a jian — the ancient Chinese sword, which symbolises impartiality and justice. In the sense that now the markets will need to be impartially reassessed and fairly divided. 

Losing a monopoly is always painful, and now we are talking about a loss of monopoly coupled with a sharp technological lag and the prospect of deflating a financial bubble inflated around the neural network industry. It is still very expensive; miracles do not happen, but today the training of large neural network models is becoming much more accessible for countries that do not print dollars, and on equipment that can be purchased with yuan.

Our realities  

Let’s return to Belarus. What do all these events mean for us? They mean that we have the opportunity to significantly accelerate in the technological race. I did not just mention the three pillars on which the neural network story stands at the beginning of this article for no reason. Look, we have already begun to develop common concepts for the development of the neural network industry at the legislative level, and this concerns not only legislation but also general approaches to data preparation for training, while the availability of quality data is one of the crucial aspects of the development of this field. In this respect, co-operation is extremely difficult to overestimate.  

As for energy, we have a whole nuclear power plant. We can afford to establish a very serious data centre and operate it cheaply.  

We have much to attract investors with. And finally, we enjoy an all-weather partnership with China, which has quite predictably overtaken the USA, taking a leading position in the technology race. The West will not be able to prevent China from occupying new markets and demonstrating the advantages of its developments to the whole world. The Celestial Empire has already surged ahead and, by presenting its own chips, has taken half of the monopoly from the USA. So, the prospects in this area seem extremely enticing. The key thing is not to lose momentum.

RETALIATORY MOVE  

We are currently witnessing the end of American sanctions related to the development of neural networks. Moreover, China will now be able to impose such sanctions on the USA itself. However, a law banning imports from China and exports to China of artificial intelligence tools has already been pushed through the US Congress. This indicates a belated attempt to shield themselves from the blow. The USA has even developed and presented a similar model, the O3 mini, in record time. Yet, it suddenly began speaking in Chinese, using Chinese characters. In other words, they even failed to steal properly to cover up their embarrassment. In any case, everyone knows how to count money, and when there is an opportunity to acquire something at a significant discount, the capitalist grin towards the US Capitol, which hinders capital growth with its policies, will not be long in coming.

By Yury Terekh