Can China Win the AI Race? Cheap Models, Open Weights, and a Price War with the US
TBS CROSS DIG with Bloomberg
As AI leaders warn of existential risk and call for slowing development, a Bloomberg video report asks whether China can win the AI race it is running against the United States. The piece argues that China, long seen as trailing, has gained ground mainly by being cheaper and by releasing open-weight models. It also notes that Chinese firms have not figured out how to make money this way, and that the US still leads at the frontier and has begun cutting its own prices.
A race neither government wants to slow
The report opens with the safety debate. Leaders of the biggest AI firms have called for slowing the technology down. One voice in the piece warns that technological change on this scale can completely reshape societies, and another says "we're at this extremely dangerous point."
Washington and Beijing have both rejected these calls. According to the report, President Trump has dismissed the fears as a hoax and insisted the US press ahead, and China has criticized the US tech leaders' warnings as "AI fear-mongering." The narrator explains the alignment by saying both are racing to dominate a technology that will redefine the world, summed up in the line "Whoever wins AI wins."
The report says the US had the early advantage, but that in June, global usage of Chinese AI models overtook usage of US models for the first time. Commentators describe AI as a potential major source of geopolitical influence for China. They argue that Beijing believes being an AI leader would give it more say over standards, rule-making, and how the technology is used worldwide, and that China presents itself as favoring a multilateral approach. Given this, one commentator says the race is unlikely to stop, "especially in China," despite the concerns AI leaders have raised.
China's crowded AI field
The report describes China's AI landscape as "extremely vibrant": almost every major Chinese tech company has a model released or in training. It sets up a parallel. In the US there are big tech companies like Google and Meta and startups like OpenAI and Anthropic. In China there are ByteDance, Alibaba, and Tencent on the big-company side, and startups such as Moonshot, DeepSeek, Zhipu, and MiniMax. One commentator calls these companies both innovative and very aggressive in how they approach the market.
Chinese firms had been developing AI for years without much traction, and the common view was that they were behind the US. According to the report, that changed in early 2025 when DeepSeek released a sophisticated reasoning model that was competitive with Silicon Valley's, reportedly built at very low cost. It rattled global tech stocks, and one clip in the piece calls it "a wake-up call."
The coffee shop test: same job, a fraction of the price
To show the price gap, the report runs an experiment. Imagine you have just opened a coffee shop and ask an AI to build its website. With Anthropic's Claude Fable 5, the result was a polished, fully functional site for about $50. Moonshot's Kimi K3 built a similar site for about $12. That is the same job at roughly one quarter of the cost.
A commentator says cost has become "a very, very important factor" in the race, and that China's offerings are currently much more attractive on that front. Another cites a Bloomberg article quoting someone who said "you don't need God to write your emails." The point is that most users are weighing cost against capability, not looking for the most powerful model available.
Closing the performance gap
The report says Chinese models are also catching up quickly in capability. Moonshot unveiled Kimi K3, a 2.8-trillion-parameter model the company says can compete with the best from OpenAI and Anthropic. A commentator says this made people realize the gap between the top Chinese and top US models was even narrower than they had thought.
The report's chart traces this history. After ChatGPT's release in late 2022, US frontier models kept improving, and Chinese models lagged for years. As of 2026, the gap has been narrowing quickly.
Adoption overseas has risen along with performance. The report cites data from OpenRouter, a platform that offers access to models from many companies. By that data, Singapore and Germany now use more Chinese models than US ones, and as of 2026 the United States does too.
Silicon Valley's quiet switch goes public
Commentators say Chinese open models have found a receptive audience in Silicon Valley because many tasks, such as AI-assisted customer service, do not need frontier capability. One describes a shift in attitude. For a while, companies used Chinese models "hush-hush," but that has changed. When a ban was threatened, hundreds of US startups publicly backed Chinese models, saying they needed them, had been using them, and found them much cheaper.
The report profiles Ben Sara, a San Francisco-based founder whose company, Polcia, uses AI agents to automate business workflows. He says Anthropic was his go-to model at the start. Even though it was expensive, he wanted to give his customers "the absolute best."
Then his platform went viral and costs got out of control. His monthly bill went from $10,000–$20,000 to $1 million, then $1.5 million. At that point, he says, he thought he was "just going to go bankrupt, straight up." Chinese models were getting a lot of online buzz at the time, so he switched to open-source models, and his costs fell by about 10x, to around $100,000 a month.
Why Chinese AI is cheap: power, talent, and open weights
The report gives several reasons for the low prices. China's electricity is cheaper, and it has abundant engineering talent that costs a fraction of what US companies pay top engineers. The report points to one more factor: open-weight models.
It explains the idea with the coffee shop analogy. A closed model is like buying a slice of cake at a café: you eat it there and can't take it with you. An open-weight model lets you take the cake home, add your own sprinkles, and get a general recipe for how it was made. You can customize it, slice it up, and even sell it yourself. The commentator argues that this lets many people examine the technology at the same time, which may give China a better chance of closing the gap.
The catch: nobody is making money
The report names a key weakness in this approach: "Nobody has quite figured out how to make money." A commentator calls it a race to the bottom, with companies undercutting each other on price for over a year. Another adds that Chinese consumers are not very willing to pay for AI services, which makes monetization hard.
One commentator compares the two ecosystems to smartphones. Anthropic and OpenAI have built the iPhone, and Chinese firms are more like Android. The iPhone makes by far the most money, but Android has the larger market share.
Beijing's strategy: affordability, the developing world, and growth
Commentators argue that low prices fit Beijing's economic goals. The government wants AI to be affordable so it can spread quickly through the economy. The report includes a clip of a Chinese official saying, in Chinese, that AI development should not be a solo performance by a single country.
The developing world is described as "a tremendous battleground," where Chinese firms can bring in their technology and achieve widespread adoption. One commentator adds that China offers an alternative to US Big Tech at a time when trade wars and tariffs make more countries want options.
At home, the report presents AI as a bet on economic growth. According to commentators, China's traditional economic pillars are faltering, and Beijing has made AI one of its new growth drivers. It also supports related fields like robotics, where AI is being built into humanoid robots. The report's data shows that GDP growth driven by property has declined since 2018, while high-tech and green industries have steadily grown, and the gap between them is expected to widen.
Washington responds: chip controls, ban talk, and theft accusations
The report says these advances have put Washington on notice. A commentator says US export controls on advanced chips and chipmaking equipment have definitely slowed China's AI development. US officials have also accused China of obtaining chips through back channels.
Washington has discussed banning some Chinese companies, especially Chinese open-source models. The report notes the irony that this is happening while greater threats from US AI companies are also being debated, threats Trump says the government can protect against. The US has also accused leading Chinese firms of "stealing" from top US models, or in one commentator's words, "shoplifting" their capabilities.
Where the US still leads
The report stresses that the US keeps important advantages. At the top end, "the biggest, smartest, baddest AI models out there," a commentator says the US is still ahead. It also has far more computing resources and better access to chips and hardware. Capital is another constraint for China: the report says Chinese AI companies are worth a fraction of the top US players, which limits their ability to fund their projects.
The price war reaches the US
US firms have started lowering prices too. In July, OpenAI launched GPT-5.6 Luna, which the report calls its most cost-efficient model so far. In the coffee shop test, Luna built the website for just over $4, matching the budget Chinese options.
Ben Sara has been trying Luna. He says its cost is starting to look like that of open-source models, and describes the 80% price cut as "a massive price drop." In his view, the Chinese open-source models showed the market that there was an alternative.
An open question
The report concludes that it is probably too early to say whether China or the US will win. One commentator argues that China's approach suggests the effective use of AI across industry to boost productivity may turn out to be a more important measure of national success than frontier capability alone. On safety, another says the world would benefit if the two sides could find some common ground.
The closing assessment is that the race is intensely competitive and unlikely to stop just because some CEOs have urged a slower pace, because the political leaderships in Washington and Beijing do not want it to slow down.
Is AI a threat to humanity? That's the question on many people's minds these days.
Leaders of the biggest AI firms call for slowing the technology's development.
When you have technological changes of this scale, societies can be completely reshaped.
We're at this extremely dangerous point.
But not everyone agrees.
President Trump is pushing back, dismissing those fears as a hoax, and insisting the US press ahead.
China has also criticized the calls from US tech leaders, calling it AI fear-mongering.
So, why are Washington and Beijing aligned? Well, they're racing to dominate a technology that will redefine the world.
Whoever wins AI wins.
The US had the early advantage, but that's now changing. In June, global usage of Chinese AI models overtook those from the US for the first time.
AI could become a major source of geopolitical influence for China.
Despite the very serious concerns the AI leaders have raised, this is a race that's unlikely to stop, especially in China.
They think that if they can be one of the AI leaders in the world, then they can have more say over the standards, over the rule-making, over really how this technology is used around the world.
China is advocating for a multilateral approach.
Against a background of existential threat, the day-to-day rivalry continues. And with so much momentum behind it, can China win the AI race?
China's AI landscape is extremely vibrant. Virtually any big Chinese tech company you can name has an AI model or a model in training.
In the US, you have Google, Meta, and then in China, you have ByteDance, Alibaba, Tencent. And then in the US, you have startups like OpenAI and Anthropic. In China, you have Moonshot, DeepSeek, Zhipu, MiniMax.
These companies are really coming up with breakthrough innovations, and they're very aggressive in terms of how they're approaching the market.
Chinese companies have been developing AI for years.
But they didn't get a lot of attraction, and there was an understanding that China's AI is relatively behind the US.
That changed in early 2025.
Chinese AI startup DeepSeek has rocked the global technology stocks.
The release of DeepSeek AI from a Chinese company should be a wake-up call.
It was a very sophisticated reasoning model that was competitive with what we saw in Silicon Valley, but they made it allegedly at a very low cost.
And right now, prices can vary greatly. Imagine you've just opened your dream coffee shop. Now, you need a website. So, like most people these days, you ask AI to build one. With Anthropic's Claude Fable 5, you get this polished, fully functional website for about $50. Not too bad, right? But China's Kimi K3 can create a similar site for just $12. Same job, one quarter of the cost.
Cost has become a very, very important factor in this AI race, and in that sense, China's offerings are a lot more attractive right now.
One Bloomberg article quoted somebody saying that you don't need God to write your emails. So, it comes down to really a balance between cost and capability.
And Chinese models are also rapidly catching up in performance.
AI startup Moonshot has unveiled Kimi K3, a 2.8 trillion parameter model it says can compete with the best from OpenAI and Anthropic.
That made people realize actually the gap was even narrower between the top Chinese models and the top US models.
And you can see that here. ChatGPT was released in late 2022. Since then, the performance of frontier US AI models has been increasing. China's AI models lagged behind for years, but as of 2026, the gap has been rapidly narrowing.
This rise in the performance of Chinese AI models has been accompanied by an increase in adoption overseas. OpenRouter is a platform that offers access to AI models from many different companies. According to its data, Singapore and Germany now use more Chinese AI models than US ones. And as of 2026, so does the US.
We've seen that China's open AI models have found a very receptive audience in Silicon Valley because you don't always need frontier capability to carry out, for example, AI-assisted customer service.
For a while, it was very hush-hush when people were using Chinese models. And now it's totally shifted. Hundreds of US startups actually rallied behind Chinese AI models when there was a threat of a ban. They all said, "We need them. We've been using them. They're a lot cheaper."
San Francisco-based Ben Sara is one of the startup founders who recently switched to using China-made AI. His company, Polcia, uses AI agents to automate workflows for businesses.
At the very beginning, Anthropic was the go-to model. And even though the models were really expensive, I was like, "You know what? I need to give my customers the absolute best."
But then, his platform went viral.
That's when cost got out of control. And so, at the beginning, I had my bill with like 10,000 bucks a month, 20,000 bucks a month. But when that became a million dollars a month, 1.5 million dollars a month, that's when I was like, "I'm just going to go bankrupt, straight up." And at the time, there was a lot of hype online about those Chinese AI models. And so, when I switched to the open-source models, it became pretty much 10x less, so around $100,000 a month.
So, how has China been able to make AI so cheap?
China's electricity is cheaper. It has abundant access to talent at also a fraction of the cost that US companies must pay top engineers.
Sure, cheap energy and talent are helpful, but there's something else.
Open-weight models.
Open-weight AI?
Open-weight.
To understand what this means, let's go back to our coffee shop.
Let's say an AI is sort of a slice of cake. If it's a closed model, you go to the bakery, or in our case, coffee shop. You buy a slice of cake, you eat it there, you can't take it out. When it's open weight, you can take the cake home, you can put different sprinkles on it. They give you a sort of general recipe of how they made it, and you can customize it to however you like. You can even slice it up, and you can sell it yourself. It allows a lot of different eyes to really look at the technology at the same time, and I think it gives them a better shot at really closing the gap.
But this model comes with a key disadvantage.
Nobody has quite figured out how to make money. It's really been sort of a race to the bottom. For more than a year now, companies are sort of undercutting each other on price.
Consumers in China are not willing to pay for AI services that much. So, the monetization prospects have been quite difficult.
Anthropic and OpenAI have built the iPhone. The Chinese companies are more like Android. iPhone makes by far the most money, but Android has the larger market share.
It fits the view of what Beijing wants in their economy. They want it to be affordable, so it can be adopted quickly within the economy.
The development of artificial intelligence should not be a solo performance by any one country, but should…
The developing world is a tremendous battleground. They're able to bring their technologies to these countries and see widespread adoption.
It also offers an alternative to sort of US Big Tech, which amid trade wars and tariffs, more countries I think want options.
China's investment in AI is also part of a bet that the technology will help generate much-needed economic growth domestically.
The traditional pillars of China's economy are faltering, and China has made AI one of its new growth drivers.
They're also supporting adjacent technologies like robotics, where you're seeing AI deployed into the humanoid robots that are being adopted.
Since 2018, GDP growth driven by the property sector has been on the decline. On the flip side, high-tech and green industries have steadily climbed, with the gap between the two expected to widen. And all of these advancements have Washington on notice.
The US's export controls on advanced chips and chip-making equipment have definitely had an impact on slowing China's AI development.
There have been accusations from US officials that China is able to get chips through the back door.
There has been discussion in Washington, D.C. about banning some of these Chinese companies, especially the Chinese open-source models. That's even as we talk about perhaps greater threats from the US AI companies. Threats President Donald Trump contends the government can protect against.
The US has also accused top Chinese companies of so-called stealing from top US models to basically shoplift their capabilities.
But to be sure, the US still has many advantages in the AI race.
When it comes to top-of-the-line frontier models, the biggest, smartest, baddest AI models out there, the US is still on top. The US also has vastly more computing resources and access to chips and hardware.
China's ability to keep up with the US is also limited by the relative scarcity of capital to keep their projects going. In terms of their market valuation, they are a fraction of some of the top US players.
US firms are also beginning to offer lower-cost AI models. In July, OpenAI launched GPT-5.6 Luna, its most cost-efficient model so far. In our coffee shop experiment, it was able to create a website for just over $4, matching the pricing of budget-friendly Chinese alternatives. And it's a model startup founder Ben Sara has been experimenting with as well.
The cost of Luna starts to be similar to that of open source. They decided to lower the price by 80%, which is a massive price drop. These open-source Chinese models showed that there was an alternative.
It's probably still too early to call whether China or the US will win the AI race. But China's AI development has demonstrated that the effective use of AI across industry to drive productivity may well turn out to be a more important indicator of a country's success in terms of AI.
When it comes to other aspects of AI like safety, I think that the whole world could really benefit from these two sides finding some way to have common ground.
The AI race right now is tremendously competitive, and that's unlikely to stop just because some CEOs have said we must slow down the pace, because the political leadership in the US and China don't really want that to happen.
Article published