China's AI Counterattack — Honestly, I'm Envious

To state the conclusion up front: Chinese AI can no longer be ignored. It is not just DeepSeek — companies like Qwen, Kimi, GLM, MiniMax, and Manus are all moving at once, and cloud, chips, and data centers have begun to connect to them. Security is still a problem that cannot be ignored, but the conclusion "don't use it because it's Chinese AI" treats risk and use as the same thing. The criterion for judgment should shift from the country to what I am sending to that service right now.

You can no longer talk about DeepSeek alone

The impression when DeepSeek first appeared was, "So China has produced a considerable AI too." Now the situation is different.

Alibaba's Qwen is expanding beyond chatbot models into coding, document understanding, images, voice, video, tool calling, and agent execution. The latest family includes Qwen3.8-Max, Qwen3.8-Omni-Flash, and Qwen3.8-LiveTranslate, and Qwen3.8-Max claims a large-scale MoE structure and a context of up to 1 million tokens. Alibaba has announced it is already training Qwen 4.

Moonshot AI's Kimi drew attention with large open-weight models like K3, released in 2026, and Z.ai's GLM family set its direction toward agents and coding. Manus aims at an agent that performs real work — when a user assigns a task, it searches, gathers material, analyzes, runs code, creates files, and writes reports.

China's characteristic is not one company's dominance but that a structure of multiple companies competing at the same time is maintained. This is a pattern that has continued since DeepSeek appeared.

There is one shared direction as well. Rather than remaining a "chatbot that answers questions" as before, they are moving toward coding, searching, using tools, and going through multiple steps to produce actual output. The next competition in AI will likely be not about who talks well but about who does the work better.

What I am really envious of is not the models but the infrastructure

Qwen, Kimi, GLM, DeepSeek, and Manus are appearing, Alibaba is expanding cloud and AI infrastructure, and Huawei is building an AI chip and software ecosystem. These move simultaneously. It goes beyond the performance of a single model — a structure is being built that pushes AI across an entire industry. Honestly, this part is enviable.

So the AI competition is no longer about a single model. Who makes a better model, who makes it cheaper to use, who attracts more developers, who builds a better agent, who has better infrastructure, who puts AI more deeply into real industry and daily life. All of it is becoming the competition.

"There is a security issue" must not become the conclusion

Whenever you talk about Chinese AI, a discussion of security attaches to it. That is correct. Trade secrets, personal information, passwords, financial information, and important source code must be handled carefully.

But "there is a security issue" and "it cannot be used" are different problems.

In South Korea in 2025, there were 193,889 traffic accidents, and 2,549 people died in those accidents. The number of injured exceeded 270,000. Looking at the numbers alone, it is frightening. Cars are clearly dangerous objects.

So should we get rid of cars? Nobody says that. You get a driver's license, obey traffic laws, wear a seatbelt, and do not drink and drive. It is not because cars are not dangerous. It is because the existence of risk and the impossibility of use are different problems.

Chinese AI may also carry risk factors. Even so, ending at "Chinese AI has security issues, so don't use it" is similar logic to saying we should get rid of cars because accidents happen. What matters is how you use it.

What does an ordinary person ask AI?

"Make me some elementary school math problems," "Translate this English sentence," "Show me some Python code," "What is this Linux command," "Summarize this post," "Plan me a trip," "Organize this document."

Is there any need to treat even things like this as the same level of security risk? I do not think so. You just need to judge what to entrust to AI.

Concretely, it goes like this. Do not put in trade secrets. Do not put in personal information. Do not put in passwords. Use local AI for important source code. Handle sensitive work in a separate secure environment. Handle ordinary questions with ordinary AI services.

There is an apt saying in our proverbs. Fear of maggots keeps you from making soy sauce. It means that rather than not using AI because of security issues, you should distinguish what to entrust to AI. It is about using it while knowing the risks.

The yardstick is the service, not the country

Pointing out the security issues of Chinese AI is surely necessary. But that yardstick should apply equally to all AI services. If you say "personal information is at risk" about Chinese AI while thoughtlessly uploading important documents and data to other overseas AI services, that too is strange.

The important question should not be "Is it Chinese AI?" but "What information am I sending to this service right now?"

Of course, this is not a story of "trust Chinese AI unconditionally." Security and data-handling issues should rather be checked properly. But the conclusion "there is a security issue → Chinese AI is nothing special" is a completely different story. A technology's competitiveness and security issues are separate problems. The fact that a car crashes does not make engine technology disappear, and the fact that a plane crashes does not make aviation technology worthless. Risk should be managed as risk, and technology evaluated as technology.

So I do not ignore it

I am not trying to say it is number one in the world. That judgment can only be known with time.

What I want to say is simple. Chinese AI can no longer be ignored. It is not a story about DeepSeek alone — Qwen, Kimi, GLM, Manus, and the infrastructure beneath them are all moving at once. All of this movement has grown too large to sum up with a single phrase: "security is a problem."

I do not like China. But AI technology is a separate matter, and in that field I have to admit it honestly.

The more important thing in the coming AI era may be not "who built the smartest model" but "who is spreading AI into reality the fastest."