The headlines have lately been full of signals of the progress China has been making in AI. Chinese open models’ share of downloads has overtaken the figure for US open models (17% vs. 15.8%). China’s flagship technical university, Tsinghua, has more patents awarded annually than the top 6 US universities (e.g. MIT, Stanford, Harvard) combined. Alibaba’s rebranded Qwen AI app just saw 10M+ downloads in its first week post-relaunch. (For comparison, ChatGPT had 606K downloads and Sora had 627K downloads in their respective first weeks.) There’s also been a steady drumbeat of model/product announcements from Chinese AI players like Alibaba, DeepSeek, ByteDance, and Huawei. And all of this AI progress is being rapidly incorporated into China’s manufacturing prowess and military might.

The efforts in China to use AI to drive down manufacturing costs has been particularly notable. Chinese president Xi Jinping has backed a pragmatic applications-oriented approach to national AI investment, with particular emphasis on advanced manufacturing. A recent WSJ report highlighted some eye-opening examples. One Chinese jacket maker reported using AI to cut the time to produce a sample by 73%, from 100 days to 27 days, lowering development costs by 60%. Steelmaker Baosteel AI has reduced the frequency of human intervention from every 3 min to every 30 min. Appliance company Midea saw the cycle time for certain processes plummet from 15 min to 30 sec, helping to increase revenue per employee by 40% (2015-2024). Cement producer Conch’s use of AI has reduced coal consumption by 1%, with a target of a 2% reduction by 2026, for potentially tens of millions of dollars in savings. The large Tianjin port is using AI to turn a typical 24 hours of schedule planning into 10 min, and the highly automated port now requires 60% fewer workers.

The co-founder of American open-model startup Reflection AI, Misha Laskin, recently described Chinese open models as “palpably close to the frontier.” A growing number of US startups are turning to Chinese open models, calling them faster, cheaper, and often just as good (depending on the use case). It’s not just parity among open models either. Lin Qiao, Fireworks AI CEO and co-creator of PyTorch, believes “[t]he gap is really shrinking” between American closed-source models (e.g. from OpenAI) and Chinese open models. The Chinese models’ value proposition is compelling enough that they are seeing this growing adoption despite indications they may have weaker safety protocols and a pro-China bias.

Some of the strongest Chinese models are coming from Alibaba and DeepSeek, which have been rapidly shipping models on a weekly or biweekly basis and offering an array of variations. In recent weeks, Alibaba has slashed the price of the closed-weight Qwen3-Max by nearly 50%; had Qwen3-Max rated as the best-performing model for financial trading (despite the model still seeing a loss); released its Qwen-powered Quark smartglasses; and had the relaunched Qwen app become “the world’s fastest growing AI app.” In Alibaba’s last earnings quarter, it reported triple-digit growth in its AI business for the 9th consecutive earnings period.

Late last month, DeepSeek revealed that DeepSeekMath-V2 is achieving gold-medal levels on the International Math Olympiad. Earlier this week, DeepSeek released DeepSeek-V3.2, which reportedly is performing on par with OpenAI's GPT-5 on reasoning benchmarks. It also released DeepSeek-V3.2-Speciale, which is reportedly performing on par with Google's Gemini-3 Pro.

There’s also a long tail of Chinese tech giants (e.g. ByteDance, Huawei) and up-and-coming AI startups. Startup Moonshot AI’s Kimi K2 Thinking has lately emerged as possibly the leading open model, reportedly outperforming OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5 on certain benchmarks.

Ex-Google CEO Eric Schmidt pointed out at TED2025: “You're six months ahead of me, and we're both on the same path for superintelligence. And you're going to get there, right? And I'm sure you're going to get there, you're that close. And I'm six months behind. Pretty good, right? Sounds pretty good. No. These are network-effect businesses. And in network-effect businesses, it is the slope of your improvement that determines everything.” Schmidt took this to mean that, if a player is ahead and being ahead helps them move faster than anyone else, then they may have already won. But as OpenAI – currently in “code red” as Gemini takes the lead – is discovering, being ahead isn’t enough. It is the trajectory that matters.

In the West, there’s often a focus on the companies that rise to the top of the leaderboards – the front-of-the-front frontier and the research breakthroughs needed to get there. China is betting on other mechanisms to reach a steep trajectory. Scale distribution, for instance, can help enact a strategic flywheel powered by user context and personalization – which is one of the routes that Google is taking. Distribution is somewhat easier in China, which has its own digital ecosystem, meaning there are fewer rivals vying to attract China’s large population of 1.4B. Chinese AI players are also looking to other countries for distribution. In regions like Africa, where Chinese giants like Huawei and ZTE have built infrastructure, DeepSeek’s lightweight, low-cost models have made deep inroads. Similarly, ByteDance’s Cici chatbot app has reportedly made inroads into the UK, Mexico, and Southeast Asia.

There’s also a strategic flywheel available that consists of lightweight low-cost models, swift implementation, deeper cost efficiencies, and rapid iteration made possible by the freed-up capital and less compute needed. As one industry watcher put it, “You let the technology leader, the US in this case, eat the cost of exploration, and then you try to be the fast follower or be the one who optimizes for implementation.” With this strategy, China may not always be on the bleeding edge but it might be able to stay very close to the cutting edge. And then, at whatever point China needs to leapfrog to the bleeding edge, that effort would be supported by a modern advanced-manufacturing system with an advantageous cost structure.

That leapfrog isn’t outside the bounds of possibility. After all, China has already ascended to become an innovation hub for global corporations in automotive, consumer and home goods, and biotech, among other sectors. For China, innovation-driven productivity is one avenue to combating the hyper-competitive “involution” and a deflationary race to the bottom caused by excess capacity. It could also be a way for China to escape the “middle-income trap” that has held back 100+ countries from progressing to high-income.

Jan 17 2025 (3 Shifts): China’s notable progress in AI models

Sep 6 2024 (3 Shifts): China's investment push in AI