What happened in the past few years in the AI race is quite interesting. In 2018, ZTE almost faced cardiac arrest due to the US export ban. But now, the story is completely different.

The real bottleneck isn’t just chip hardware. The true chokehold is NVIDIA’s CUDA ecosystem. Did you know that over 90% of AI developers worldwide depend on CUDA? It’s not just a tool; it’s a flywheel that’s almost impossible to replace. Ten years of development, millions of developers, thousands of applications—all built on top of CUDA. It’s like an ecosystem with no alternative.

But in China, the approach is different. Instead of direct confrontation, they pivoted to algorithm optimization. DeepSeek V3 is a perfect example—671 billion parameters but only using 5.5% per inference. The training cost is just $5.6 million, compared to GPT-4’s $78 million. Plus, API pricing is 25-75 times cheaper. That’s the power of smart engineering.

Now, local computing infrastructure production is accelerating. New server production lines are emerging, using homegrown chips like Loongson and TaiChu Yuanqi. The key milestone is this—the local chips have transitioned from inference-only to full training capability. It’s a qualitative leap. The Huawei Ascend ecosystem now has 4 million developers, and 43 major models have been pre-trained using Ascend chips.

The energy situation is another huge factor often overlooked. China produces 2.5 times more electricity than the US, and industrial electricity costs are $0.03 per kWh compared to $0.12-0.15 in the US. That’s a massive advantage. Meanwhile, US states like Virginia, Georgia, Illinois have paused new data center permits due to power grid constraints.

So what happened? Tokens—the basic unit of AI output—have become a new commodity. Produced in China’s computing factories, distributed globally. DeepSeek’s user distribution is 30.7% China, 13.6% India, 6.9% Indonesia, 4.3% US. It’s not just about technology; it’s about market shift.

There’s a historical parallel here. Japan’s semiconductor industry in the 1980s was similar—they were dominant but relied on a US-controlled ecosystem. After the US-Japan Semiconductor Agreement, everything changed. Japan lost market dominance because they didn’t build an independent ecosystem.

In China, the approach is more strategic. Algorithm optimization, local chip development, Ascend ecosystem building, and global token distribution are all coordinated. Each component strengthens the others. It’s not an overnight success—there are losses, challenges, but a systematic building of an independent industrial ecosystem.

Latest earnings reports from local chip makers are telling—Cambrian revenue up 453%, Moore Threads up 243%, Muxi up 121%. Half fire, half water in performance, but market demand is clear. The world needs an alternative to NVIDIA’s monopoly.

The real cost isn’t just technology; it’s ecosystem building—software subsidies, developer support, on-site engineering. These are the expenses needed for independence. But that’s the price of having options.

It’s been eight years since the ZTE incident. Back then, the question was survival. Now, the question is different—what’s the price of independence and long-term competitiveness? The answer is clearer now—there’s a need for a comprehensive ecosystem, not just chips. Algorithms, energy, developers, and global market access. China is building all of this in parallel, and it’s fundamentally changing the game.