I've noticed something very important happening behind the scenes in the AI industry right now, and I believe most people are missing the big picture.

Eight years ago, the ZTE story was a harsh lesson about technological dependence. A single American ban, and a company with 80,000 employees was completely shut down. But what’s happening now in the AI field is entirely different, and the picture is much more complex.

The real problem wasn’t always chips. When restrictions on NVIDIA A100 and H100 began, everyone thought this was the choke point. But the actual choke point was something called CUDA. This software platform from NVIDIA became the foundation of everything in global AI. Over 90% of AI developers worldwide operate within this ecosystem. It’s a sustainable wheel— the more people use it, the more valuable it becomes.

But here’s the interesting part. Instead of trying to compete directly with NVIDIA, Chinese companies have chosen a completely different path.

First, algorithms. Chinese companies have shifted to hybrid expert models— instead of running everything, they only activate the relevant parts. DeepSeek V3 is an ideal example: 671 billion parameters, but it only uses 37 billion during inference. The result? Training costs are 14 times lower than GPT-4, and API prices are 25 to 75 times cheaper than competitors.

Second, local chips. What’s truly remarkable is that Chinese local chips have now surpassed simple inference. In January 2026, the first fully advanced image model was trained entirely on Chinese local chips. In February, a large model was trained on a complete Chinese computing pool. This is a qualitative shift—from just running models to actually building new ones.

As for the prices of new water chips and infrastructure, the situation is extremely concerning for the United States. Electricity is the fundamental limit of computational power. China produces 2.5 times more electricity than the US, and its industrial costs are 4-5 times lower. Meanwhile, the US faces a real electricity crisis—Virginia and Georgia have halted approvals for new data centers.

The result? Chinese AI is quietly going out into the world. Not as a traditional product, but as (tokens)— the basic units of processing. They are produced in Chinese computing factories and transmitted online everywhere. DeepSeek now serves 30% of China, 13.6% of India, 6.9% of Indonesia. In sanctioned countries, market share ranges between 40-60%.

There’s a strange resemblance here to the story of Japanese semiconductor industry in the 1980s. Japan was the best in a global system dominated by others, but it didn’t build an independent ecosystem. When the wave receded, it had nothing left. This time, China is choosing a different path—building a truly independent ecosystem from scratch.

Recent earnings reports from local companies tell the real story. Half are fire, half are water—huge revenues but significant losses. But this isn’t management failure. It’s a war tax for building real independence.

The question now isn’t “Can we survive?” but “How much do we need to pay to stay independently alive?” And the answer is the same: progress.