Eight years ago, ZTE was suffocating under U.S. sanctions. Today, the picture is completely different.

I noticed something very important happening in the Chinese artificial intelligence industry. Not only is DeepSeek undercutting OpenAI’s prices by a factor of 25, but more importantly, the infrastructure is changing radically.

The truth everyone had been ignoring: what’s choking China isn’t the chips themselves, but CUDA from NVIDIA. This software platform has monopolized 90% of the global AI development market. Every developer, every framework, every project connected to it. Trying to get out of this system is like rewriting contracts built on years of accumulated experience.

But Chinese companies chose the harder path. Instead of direct confrontation, they broke through on three fronts:

First, algorithms. Mixture-of-experts models reduced training costs from 78 million دولار (تكلفة GPT-4) إلى 5.5 million to just 5.5 million. A huge difference.

Second, local chips. In Shaanxi, a single production line worth 1.1 billion yuan started producing complete computing servers with Loongson processors and Taichu Yuanqi cards. Most importantly: these chips moved from inference to training. This is a qualitative shift.

Third, the ecosystem. 4 million developers are now working on the Huawei Ascend platform. In January, the first advanced image-generation model fully trained on Chinese chips was trained. In February, the massive “Stars” model was trained on a local computing pool.

The seawater tides are flowing. The United States is facing a real electricity crisis. Electricity costs rose 267% in data-center regions. While China produces 2.5 times more electricity than the United States, industrial electricity prices are 4 times cheaper.

Now the (Tokens) are leaving China quietly. 30% of DeepSeek users are local, but the remaining 70% are distributed worldwide. 26,000 global companies use it. In sanctioned countries, its market share ranges from 40–60%.

This is similar to a war for industrial independence. In the 1980s, Japan controlled 51% of the semiconductor market, but it was willing to be the best producer in a global system dominated by an external power. When the tide receded, all it had was manufacturing.

This time, China is building a fully independent system—from algorithms to chips to software to global distribution.

On February 27, three Chinese chip companies released their results on the same day. Revenue surged by huge percentages (453%, 243%, 121%). Most of it was massive net losses. But this isn’t failure. It’s the price of war.

Every loss is an investment in building the ecosystem: research and development, software support, field engineers. These losses are the cost of independence.

The question is no longer “Can we survive?” but “What price must we pay?” And the price itself is progress.