I have noticed something important lately worth discussing. Eight years ago, the heart of a major Chinese telecom company stopped due to a single American ban. But what is happening now is completely different. Instead of surrendering, Chinese companies have chosen a harder and more creative path.

The truth that many have overlooked is that the core problem is not the chips themselves, but NVIDIA’s CUDA development platform. This platform accounts for about 90% of the global AI development market. Millions of developers have learned on it, and millions of applications are built on it. The more developers there are, the more tools and libraries are available, and as the environment thrives, it attracts even more developers. It’s a closed loop that’s very hard to break out of.

But in 2024-2025, a radical shift occurred. Chinese companies began focusing on improving algorithms rather than directly fighting the ban. Hybrid expert models became the new trend. DeepSeek is a clear example: 671 billion parameters, but only 37 billion are used during operation. Training costs just $5.6 million compared to $78 million for GPT-4. The price difference caused their model to spread rapidly.

By February 2026, the use of Chinese models on the world’s largest aggregation platform increased by 127% in just three weeks. A year ago, their share was less than 2%, now it’s approaching 60%. This is no coincidence. Emerging markets in India, Indonesia, and Brazil have started heavily relying on these models.

As for chips, the story is even more exciting. Local chips like Loongson and Taichu Yuanqi have begun training large-scale models. In January 2026, Zhipu AI launched the first fully Chinese-trained image model. This is a qualitative shift from inference capability to training capability.

The most important point here relates to energy. China produces 10.4 trillion kWh annually compared to 4.2 trillion in the US. Industrial electricity in China is 4-5 times cheaper than in America. While the US faces a real electricity crisis, China has enormous production capacity that can be directed toward computing.

What is coming out of China now is not products or factories, but tokens themselves. The units of information processed by AI models have become a new digital commodity. They are produced in computing factories and then transmitted over the internet worldwide.

Data on DeepSeek user distribution tell the story: China 30.7%, India 13.6%, Indonesia 6.9%, America only 4.3%. 26,000 global companies have accounts. In China, they captured 89% of the market.

This is very similar to the war on industrial independence that happened with Japan 40 years ago. Japan was at the top in 1988 with 51% of the semiconductor market, but accepted being a better producer in a system dominated by others. When circumstances changed, it collapsed.

The difference this time is that China is building a truly independent ecosystem. From algorithm improvements, to local chip leaps, to 4 million developers in the Ascend environment, and finally the global spread of tokens. Every step builds real independence.

On February 27, 2026, three Chinese chip companies announced their results on the same day. Revenues soared by huge percentages: (453%, 243%, 121%), but some also incurred massive losses. These losses are not management failures but a tax of war to build an independent ecosystem. Every dollar lost is an investment in R&D and human support.

The market needs an alternative to NVIDIA. This is a very rare structural opportunity resulting from geopolitical tensions. The war over computational power has changed its shape. Eight years ago, we asked: Can we survive? Now, the question is: How much do we need to pay to survive? And the answer itself is real progress.