For the longest time, the AI race seemed pretty simple...
America was winning.
OpenAI had GPT.
Google had Gemini.
Anthropic had Claude.
NVIDIA was selling the GPUs that basically everyone wanted to train AI.
And China?
China was supposed to be catching up...
But then something interesting happened.
DeepSeek happened.
Qwen kept getting better.
Kimi started getting serious.
And suddenly, people around the world started realizing something...
China wasn't just "catching up" anymore.
It was becoming one of the most important players in AI.
And personally, I think the most interesting part of China's AI industry isn't even necessarily its biggest models...
It's the open models.
Or, more accurately in many cases, open-weight models.
Models like Qwen, DeepSeek, Kimi, GLM, and others are doing something that I think could become extremely important in the next few years...
They are making powerful AI accessible.
Not just through an API...
But directly to developers, companies, researchers, and basically anyone who wants to build something with them.
And honestly... I think that could be a much bigger deal than people realize.
First... America and China are playing slightly different games
Right now, most of the biggest American AI companies have a pretty straightforward business model.
They build a ridiculously powerful model...
They put it on their servers...
And you pay to use it.
You want GPT?
Use OpenAI's servers.
You want Claude?
Use Anthropic's servers.
You want Gemini?
Use Google's infrastructure.
It's simple.
The company owns the model.
The company owns the servers.
The company controls access.
And, obviously, the company charges you.
This isn't necessarily bad. In fact, it makes complete sense.
Training frontier AI models costs billions of dollars. Running them costs money too. Companies need to make that money back somehow.
But China's AI ecosystem has been moving in another direction too...
Build powerful models...
Release the weights...
Let developers build on top of them.
And that's where things get interesting.
Because once you release a capable model, it stops being just your product.
It can become infrastructure.
Someone can run it locally.
Someone can fine-tune it.
Someone can build an AI assistant around it.
Someone can put it inside a coding tool.
Someone can deploy it inside a company.
Someone can literally build an entire startup around it.
That is a completely different kind of strategy.
Qwen is probably one of the best examples of this
I've personally found the Qwen ecosystem extremely interesting.
Alibaba isn't just trying to build one giant chatbot and call it a day...
There are different models.
Different sizes.
Different capabilities.
Different models for different use cases.
And that's important because the future of AI probably isn't going to be...
«Everyone opening one website and talking to one gigantic AI model.»
AI is going to be everywhere.
Inside operating systems.
Inside coding tools.
Inside apps.
Inside companies.
Inside robots.
Inside servers.
Inside phones.
Inside cars.
And for most of those things, you don't necessarily need the biggest and smartest AI model on the planet.
You need something that is...
- Fast
- Cheap
- Good enough
- Reliable
- Customizable
- Possible to run on your own infrastructure
That last one is huge...
Because companies don't always want their data leaving their infrastructure.
Governments definitely don't want to depend entirely on another country's AI companies.
And developers don't always want to pay an API bill forever.
That's where models like Qwen become extremely powerful.
And then there's Kimi...
Kimi is another reason why I don't think DeepSeek was just some random one-time event.
Moonshot AI and other Chinese labs are showing that there is an actual ecosystem developing in China.
Multiple companies are competing.
Multiple companies are training models.
Multiple companies are trying different approaches.
And that's important.
Because people often look at this as...
«America vs China.»
But there is another competition happening...
China vs China.
Alibaba.
DeepSeek.
Moonshot AI.
Z.ai.
MiniMax.
Baidu.
ByteDance.
And many more.
All of them want to build better models.
Better coding models.
Better reasoning models.
Better agent models.
Cheaper models.
Faster models.
That competition matters.
Because competition creates pressure...
And pressure creates innovation.
If you're the only company in a country building AI, you can take your time.
But if five other companies are about to release something better than you...
Well...
You move faster.
DeepSeek changed the entire conversation
Before DeepSeek, there was this assumption that frontier AI was basically impossible without infinite money.
You needed billions of dollars.
Thousands and thousands of GPUs.
Massive data centers.
The newest NVIDIA hardware.
And basically the infrastructure of Google or Microsoft.
Then DeepSeek showed up...
And suddenly, everyone started paying attention.
Now, I'm not saying DeepSeek magically solved AI.
And obviously, there are debates about training costs, hardware, compute, and all of that.
But that's not even the most important part.
The important part was what DeepSeek proved...
Chinese AI labs could release models good enough to make the entire industry pay attention.
And after DeepSeek...
People started looking more seriously at everything else coming out of China.
Qwen.
Kimi.
GLM.
MiniMax.
Suddenly, China wasn't just this country everyone expected to follow behind the US.
It was becoming one of the biggest sources of powerful AI models in the world.
Especially open ones.
And I think that's where the real story begins...
Open AI creates something closed AI cannot easily create...
An ecosystem.
Think about Linux.
Linux didn't become important because one company forced everyone to use it.
It became important because developers could build on it.
Companies could build on it.
Researchers could experiment with it.
Entire companies could create products around it.
Now think about Android.
Google created the base.
Then Samsung built on it.
Xiaomi built on it.
OnePlus built on it.
Thousands of developers built apps for it.
And suddenly, you don't just have an operating system...
You have an ecosystem.
I think open AI models could work similarly.
Imagine a powerful model like Qwen being available to developers.
One developer builds a coding assistant.
Another builds a local AI chatbot.
Someone else builds an AI operating system.
A company fine-tunes it using their internal data.
Another company uses it for customer support.
Someone builds an AI agent.
Someone builds a robot around it.
And eventually...
That model becomes infrastructure.
This is the part that I think people underestimate.
The company that builds the smartest AI isn't necessarily the only company that wins.
The company whose AI becomes the foundation for millions of other things could win too.
And China is also making AI cheap...
This is probably one of the biggest reasons Chinese models are becoming popular.
Because let's be honest...
Most people don't need the smartest AI in the universe.
A company doesn't necessarily care if GPT is 3% smarter than another model.
What they care about is...
«Can it do the job?»
And then...
«How much does it cost?»
Imagine two models.
Model A is slightly better.
Model B is almost as good for your use case but costs significantly less.
If you're just using AI occasionally...
Maybe you choose Model A.
But what if you're processing millions of requests?
What if you're building an AI product?
What if you're running thousands of AI agents?
Suddenly, cost becomes extremely important.
And I think this is where Chinese companies have understood something very important...
AI isn't only a research competition anymore... It's becoming an economics competition.
The question isn't just...
«Who has the smartest model?»
It's also...
«Who can give people powerful AI at the lowest possible cost?»
That is a completely different competition.
And China seems very serious about winning it.
The US restrictions might have created an unexpected result...
The United States has restricted China's access to some of the most advanced AI chips.
And obviously...
That creates serious problems for Chinese companies.
I'm not going to pretend otherwise.
Access to advanced GPUs is extremely important.
NVIDIA has an incredible ecosystem.
And the United States still has major advantages when it comes to frontier compute and AI infrastructure.
But there is another side to this...
Restrictions can create pressure.
And pressure forces efficiency.
If you don't have unlimited access to the newest hardware...
You have to figure out how to do more with what you have.
You need better architectures.
Better training techniques.
Better inference efficiency.
Better memory management.
Better software.
Better optimization.
Basically...
You can't just throw more GPUs at the problem.
You have to become smarter.
Again, I'm not saying export restrictions helped China.
They clearly created major challenges.
But they may have also pushed Chinese AI companies to focus heavily on efficiency.
And that efficiency is now becoming one of their strengths.
China also has something incredibly important...
Scale.
China has...
- A massive population
- Huge technology companies
- Massive manufacturing
- A huge number of engineers
- Massive cloud infrastructure
- A huge consumer technology market
- Serious government interest in AI
That creates an environment where AI isn't just another startup trend.
It's becoming infrastructure.
AI can be integrated into...
Manufacturing.
Robotics.
Phones.
Apps.
Education.
Transportation.
Government systems.
Companies.
And that's important because the future AI race might not be won by whoever builds the best chatbot.
It might be won by whoever actually deploys AI everywhere.
Think about it...
Building a powerful AI model is one thing.
Actually getting millions or billions of people to use AI is something completely different.
And China is extremely good at scaling technology.
I think the future could look a little like Android vs iPhone...
Not literally...
But philosophically.
On one side, you could have extremely powerful closed AI systems.
OpenAI.
Anthropic.
Google.
These companies build the model...
They run the infrastructure...
And you access the AI through their products or APIs.
It's controlled.
It's polished.
And probably extremely powerful.
Then on the other side...
You have powerful open-weight models.
Qwen.
DeepSeek.
Kimi.
GLM.
Llama.
And many others.
You can download them.
Run them yourself.
Fine-tune them.
Deploy them.
Build products around them.
And I don't think one side is going to completely destroy the other.
Both will probably exist.
Some people will always want the absolute smartest AI possible.
Others will want control.
Privacy.
Customization.
Lower costs.
Or independence from a specific company.
And those people will choose open models.
The really interesting part is the Global South...
I think this could become huge.
The US and Europe can afford expensive AI infrastructure.
But what about everyone else?
India.
Africa.
Southeast Asia.
Latin America.
Smaller countries.
Smaller companies.
Universities.
Individual developers.
Not everyone can afford expensive API calls forever.
But an open model?
That's different.
You can download it.
Run it locally.
Host it on your own server.
Customize it.
Fine-tune it for your language.
Build your own product around it.
And if Chinese models become the default AI infrastructure in developing countries...
That would be massive.
Not because China forced anyone to use them...
But because people chose them.
Because they were cheap.
Accessible.
Capable.
And customizable.
That's a completely different kind of technological influence.
And honestly...
I think this possibility is heavily underestimated.
But China isn't winning everything...
Let's be clear here...
China still has serious challenges.
The United States still has huge advantages.
Especially when it comes to...
- Advanced AI chips
- NVIDIA's ecosystem
- Frontier compute
- Venture capital
- Cloud infrastructure
- Some of the world's best AI research labs
- The biggest closed frontier models
And there are also concerns about Chinese models.
Not every model described as "open" is actually fully open source.
Sometimes the model weights are available...
But the training data isn't.
Sometimes the full training process isn't reproducible.
There are also concerns about censorship and how some models handle politically sensitive questions.
So no...
Chinese AI isn't some perfect open-source paradise.
Far from it.
But pretending that China is still massively behind everyone else is also becoming increasingly difficult.
The gap is getting smaller.
And in some areas...
There might not even be a meaningful gap anymore.
My prediction for the future...
I think the next few years are going to be absolutely insane.
And I think three things are going to happen...
- Closed models will probably remain at the absolute frontier
OpenAI.
Anthropic.
Google.
And maybe other companies will continue building the most powerful AI models.
They have massive amounts of compute.
Massive data centers.
The latest chips.
Huge research teams.
If you want the absolute best AI model for the most difficult tasks...
Closed models might continue leading for quite some time.
- Open models will become good enough for almost everything
And this...
This is where things become interesting.
If an open model is 50% as good as GPT...
Most people will still use GPT.
If it's 70% as good...
Same thing.
But what happens when an open model becomes 90% as good?
What happens when it becomes 95% as good for your particular use case?
And you can run it yourself?
Customize it?
Control your own data?
Avoid paying API costs forever?
Now companies start asking questions...
«Why are we paying so much?»
«Why are we sending our data somewhere else?»
«Why are we dependent on one company?»
«Why can't we just run this ourselves?»
That's where open models become dangerous for closed AI companies.
They don't necessarily need to become better than GPT or Claude.
They just need to become...
Good enough.
Cheap enough.
Fast enough.
And accessible enough.
And honestly...
They're getting there very quickly.
- The AI race will become an ecosystem race...
Right now, everyone is obsessed with benchmarks.
Who's number one?
Who scored higher?
Which model can solve more problems?
And yeah...
That stuff matters.
But eventually, I think ecosystem will matter even more.
Imagine this...
One model is 3% smarter.
But another model has millions of developers building tools around it.
Thousands of companies deploying it.
Governments using it.
Startups fine-tuning it.
Researchers experimenting with it.
Who is actually winning?
The answer becomes complicated.
Because technology leadership isn't just about having the best product.
It's also about...
- Adoption
- Developers
- Cost
- Infrastructure
- Distribution
- Integration
- Ecosystem
And China is becoming very good at those things.
So... Why is China doing so well in AI?
I think it's because of a combination of things...
Massive engineering talent.
Huge technology companies.
Internal competition.
Government support.
Scale.
Efficiency.
A massive domestic market.
And increasingly...
A willingness to release powerful models and let developers build on top of them.
That last part could become extremely important.
Because the real AI war might not be...
«OpenAI vs DeepSeek.»
Or...
«America vs China.»
The bigger question might be...
What happens when powerful AI becomes cheap enough and accessible enough that almost anyone can use it?
What happens when a startup can download a model instead of paying millions for API access?
What happens when governments can run their own models?
What happens when your laptop can run an AI that would have required a data center a few years ago?
What happens when open models become the foundation of millions of products?
I don't think we have the answers yet...
But I do think one thing is becoming clear.
China isn't just trying to catch up anymore.
It's helping shape what the next generation of AI could actually look like.
America might continue building some of the smartest AI models in the world...
But China could play a huge role in making powerful AI available to everyone else.
And if that happens...
The AI race won't be decided by one model.
It won't be decided by one benchmark.
And it definitely won't be decided by one company.
It will be decided by ecosystems...
And honestly...
I don't think we've even seen the real AI war yet...
Maybe it isn't far away tho.
Top comments (0)