Transcription
The most downloaded AI on the planet isn't American. It's Chinese. It's called Qwen. It has over 700 million downloads. And most people in the West have never heard of it.
Same goes for DeepSeek, Kimi, GLM, MiniMax. You don't know these names, but you should. They're free. They're open source and they match or beat models that cost 40 times more. Airbnb runs it’s customer service on one. Pinterest built its recommendation engine on another. 80% of Silicon Valley startups working with open source AI are building on Chinese models right now.
Now, I already know what some of you are thinking. Chinese AI? Isn't that dangerous? Won't they steal my data? Aren't these models censored by Beijing? Fair questions. Ask DeepSeek about Tiananmen Square and it will politely change the subject. You’re right… But ChatGPT trains on your conversations by default. Gemini merges your prompts with your Gmail, your YouTube, your entire Google History. Claude quietly changed its privacy policy and opted users into five years of data retention. The idea that Chinese AI is the threat and Western AI is the safe choice is complete bullsh*t.
The people who actually understand this space, the engineers, the investors, the people running real companies, they already switched quietly without announcing it because they know something you don't. The best AI tools on the Earth right now. On Chinese money is pouring into the Chinese AI sector like it was for crypto five years ago. Except these companies have actual revenue. That's what this video is about. Not the geopolitics, not the US-China rivalry. It's about return on investment. If you're interested in what most Western media isn’t covering, this channel is built for that. I've been on the ground in Asia for 26 years, and I share what I see from the inside. Subscribe.
Andreessen Horowitz is the most powerful venture capital firm in Silicon Valley. They backed Facebook before anyone believed in it. They backed Airbnb, GitHub, Coinbase. When they published data about where the tech industry is going, people listen. Their latest numbers say that 80 percent of American startups building on open-source AI are doing it on Chinese models. The US spent hundreds of billions of dollars to stay ahead in AI, and its own startups are quietly building their products on technology made in China. Nobody's talking about it because you don't see the switch happen. It's buried in infrastructure decisions and pricing spreadsheets that never make the news.
When Airbnb moved its customer service AI to Alibaba's Qwen, there was no press release. The CEO said it in a Bloomberg interview with 3 words: good. Fast. Cheap. When Pinterest rebuilt its recommendation engine on Chinese open-source models, their CTO said they're 30 percent more accurate and 90 percent cheaper than the American alternatives. Cursor, the most popular AI coding tool in America, got caught in a more embarrassing way. When they released their latest version, users dug into the code and found the engine underneath wasn't American, it was Kimmy K 2.5 made by a Beijing startup called Moonshot AI. The co-founder had to admit publicly that they had built their flagship product on Chinese AI and hadn't told anyone.
Even Mira Murati left. She was OpenAI CTO, the person most responsible for ChatGPT becoming the product the world knows. She raised $2 billion. The biggest seed round in history, started her own lab. And the first thing she shipped was a tool that helped developers fine tune Alibaba’s Qwen. Even Stanford had to admit it. China has nearly erased America's lead. The gap between the best US model and the best Chinese model is 2.7%. That's it. Except Stanford measured performance. They didn't measure value. When the models are equal and one of them is 40 times cheaper, “nearly erased” is a polite way of saying you lost.
But let's talk about what cheaper actually means, because the numbers are almost absurd. If you run a business and you're using Claude or ChatGPT through their API, you're paying somewhere between $2.50 and $25 per million tokens depending on the model. DeepSeek’s budget model charges $0.28. Alibaba’s Qwen 3.5 plus comes in around $1.20. Moonshot’s Kimi K2 charges $0.15 on input. That's not a discount. That's a different economic reality. A company processing 100 million tokens a month pays about $35 on DeepSeek. The same workload on Claude Opus runs about $1,500. Same input, same output quality on the benchmarks. And one bill is way larger than the other.
And the pricing gap doesn't come from cutting corners. It comes from a fundamentally different engineering philosophy. When the US banned China from buying Nvidia's best chips, the assumption in Washington was that Chinese AI labs would fall behind. No chips, no compute, noncompetitive models. What actually happened was the opposite. DeepSeek couldn't buy the hardware, so it rewrote the software. Its engineers sliced the model into 256 ultra specialized expert clusters so that when you ask a coding question, only eight of those clusters wake up. The rest of the brain stays asleep. They compressed the model's working memory by over 90%. DeepSeek built a world-class model for reportedly $6 million dollars. And this month, the next version, v4, is expected to drop multimodal 1 million tokens of context. And now, according to multiple reports, trained entirely on Huawei chips.
So far, I've been talking about who's using Chinese AI. Now let's follow the money, because what's happening on the capital side is just as wild. My company operates in Hong Kong. I've been here for two decades. And in the last few months, the energy around AI companies in the city has been unlike anything I've seen since the pre-2020 tech boom. Bankers, investors, lawyers, everyone is working on the same type of deal: Chinese AI IPOs. Zhipu AI is a Beijing company that builds open-source AI models. Almost nobody in the West has heard of it. Inside China, it's one of the so-called AI Tigers. The generation of startups competing head to head with OpenAI. On January 8th, Zhipu listed in Hong Kong. The public offering was oversubscribed 1160 times for every single share available. Over 1000 investors were fighting to get in. Within 43 days, the stock had climbed 524%. The very next day, a Shanghai startup called MiniMax listed on the same exchange. Same story, different numbers. Investors borrowed 148 billion Hong Kong dollars in margin financing just to access the retail tranche. The stock doubled on day one. Within six weeks, both companies were worth over 40 billion Hong Kong dollars each. MiniMax even surpassed Baidu in market cap for a moment. A startup nobody has heard of overtook the Chinese equivalent of Google.
You're not hearing about this on CNBC. Your financial advisor isn't calling you about it. If you're a retail investor, chances are someone saw your, well, ETF and told you that was diversified enough. Meanwhile, the hottest IPO market on Earth right now isn’t the Nasdaq, it’s the Hong Kong Stock Exchange. I see this firsthand at Statrys. We help entrepreneurs set up and operate in Hong Kong and Singapore. People used to ask us about Hong Kong for tax efficiency. Now they're asking about Hong Kong because that's where the AI ecosystem lives. Different motivation, same destination. The city raised $14 billion in equity sales in the first quarter of 2026, the best Q1 in five years, ahead of every other exchange on the planet. And it's almost entirely driven by Chinese AI. The pipeline behind it is stacked. Moonshot AI, the company behind Kimi, is now preparing a Hong Kong IPO at $18 billion. Unitree Robotics, China's leading humanoid robot maker, filed for a $610 million IPO in Shanghai. By the way, Unitree actually turns a profit, $87 million last year. Behind them, over 400 companies are already in the Hong Kong listing pipeline. And nobody is telling you about these names.
Alibaba just opened a data center running entirely on its own domestically designed chips. ByteDance plans to spend $23 billion on AI infrastructure this year. Alibaba committed $53 billion over three years. Those numbers are still smaller than what Alphabet and Meta are burning in the US. But that's the whole point of this story. China keeps getting comparable results for less money. Right now, in cities across China, local governments are handing out subsidies to a new kind of business they call "one person companies." Cash incentives to get individuals building AI-powered businesses on their own, using open source models that cost almost nothing to run. They're not experimenting with AI adoption. They're industrializing it.
You can see the results in strange places. China's short drama industry now produces roughly 470 new shows every single day. That number sounds made up. But it's not. AI tools slashed the production cost of a short drama from around $150,000 to about $15,000, and cut the production timeline from a month to under five days. That's what happens when the tools are open, cheap, and an entire population of 1.4 billion people is comfortable using them.
So what does any of this mean for you? Wherever you are, if you're a student or developer, the tools that were locked beyond corporate budgets six months ago are now free to download. You can run a model on your laptop that performs at the same level as what billion dollar companies use. The playing field just got flattened in a way that hasn't happened since the early internet.
If you work in a company that uses AI, ask your team what models you're running and how much you're paying. You might be surprised by the answer. And you might be even more surprised by the alternative. If you run a business, this is more direct. Qwen, DeepSeek, Kimi – these aren’t research experiments. Fortune 500 companies run real workloads on them. You can host them on your own servers. Your data never leaves your infrastructure, and you can fine tune them for your exact use case at a fraction of what you're paying today. When you use ChatGPT or Claude through an API, you're renting someone else's brain, and they can change the price whenever they want. When you download Qwen, you own it. Maybe not the best update, but still.
And if you none of the above and you just come here because the thumbnails look interesting, here's what matters. The technology that's going to shape the next decade isn't coming from one country anymore. It's coming from two. And the one most people are ignoring is moving faster. Every time a technology shifts from expensive and closed to cheap and open, a new generation of businesses get built in the gap. Not the people who invented the technology, the people who figured out how to apply it, package it, distribute it, integrated into industries that the investors never thought about. That's the window that's open right now.
Manus AI understood this early. It was a Chinese-started building AI agents. The founders reincorporated in Singapore, positioned themselves at the intersection of Chinese AI technology and global markets, and Meta acquired them for $2 billion. They didn't build a model. They built on top of models that were free, and they picked the right geography.
Well, I need to be honest about the risks too, because there are real ones. Chinese models carry political censorship on certain topics. For most business applications, that doesn't matter. Nobody's using the customer service chatbot to discuss Chinese politics, but you should know it's there. There are open questions about data handling, regulatory uncertainty around US-China tech relations, and the possibility that Washington restricts the use of Chinese models at some point. Some of that is genuine concern. Some of it is protectionism dressed up as security. Either way, smart companies are already hedging, running Chinese models for cost-sensitive workloads and keeping Western models for tasks where they need maximum control or compliance.
The point is not that Chinese AI is perfect and the American AI is dead. The point is that the choice exists now, and most people in the West don't even know it's there. A year ago, there was one playbook: pay OpenAI. Or pay Google. That consensus aged badly. The US has the capital and the frontier labs. China has the efficiency, the open-source ecosystem, the adoption, and the manufacturing base. And Europe… Well… Europe isn’t even in the conversation. The continent trains more AI researchers per capita than either the US or China… And then watches them leave. The EU's AI Act, designed to set ethical standards, ended up raising costs and slowing deployment. Even Mistral’s CEO, the man running Europe’s best shot at a competitive AI company, wrote publicly that European developers operate under a “fragmented legal environment” while US and Chinese companies develop under permissive or non-existent copyright rules at the moment. Sadly, the continent produces brilliant researchers, the near-zero global AI products.
I've been in Asia for 26 years. I've watched cycles like this before. When something shifts at this speed, the window between “early mover” and “too late” is shorter than people think. This channel exists to give you that kind of edge. What's actually happening in Asia where the money moves, what the Western press is and covering. If that's useful to you, subscribe. If you're already started using Chinese AI tools in your work, I'm generally curious what your experience has been. Drop it in the comments. And if this is the first time you're hearing about any of this, well, now you know. The question is, what do you do with it? See you in the next video.