It’s wild that, in 2025, people still say that only low-quality products come out of China’s factories. Some people even believe that China lags behind the US!
China is one of the biggest sources of talent there is. And big tech companies don’t only choose China for the supposedly lower costs, but for the huge concentration of skill and machinery. It’s one of the few places with the manufacturing ecosystem to produce state-of-the-art products requiring highly specialized precision tooling. China stopped being the low-labor-cost country a long time ago!1
Yes, China still produces low-value copies of Western products. However, they are ahead of or extremely competitive in certain strategic fields, such as robots, batteries, electric vehicles, solar energy, and even some factory automation.
Of course, China isn’t better in every way. The US, Europe, or even Japan have an edge in lithography and some precision machine tools.
This high concentration of skill is also what allows them to stay competitive in AI. The newly released Kimi K2 Thinking model from Moonshot AI just showed us that Chinese and open-source AI can be leading the race on key benchmarks, right next to the closed-source models from US frontier AI labs. And they’re doing all this with severely constrained access to the newest Nvidia chips!
These sanctions forced Chinese researchers to be more creative around efficiency. It pushed them to innovate. One illustration is the reported $4.6 million training cost for K2 Thinking, according to a CNBC source (not independently verified)2. If that number is even roughly accurate, it is much less than other equivalent trillion-parameter reasoning models, showing how efficiently Chinese labs are using their limited compute. That’s clearly not what was intended when the US introduced those sanctions.
A few months ago, it looked obvious that the US would dominate the open-source space with Meta and their incredible Llama family of models. But with Llama 4, they stumbled a little and lost a large part of their lead to Moonshot, DeepSeek, Zhipu AI and Alibaba. Meta is playing the benchmark game - in my experience, their models don’t feel as good as the scores suggest.
In early November 20253, Jensen Huang, CEO of Nvidia, stated that:
“As I have long said, China is nanoseconds behind America in AI. It's vital that America wins by racing ahead and winning developers worldwide.”
Tim Cook has publicly explained that Apple manufactures in China primarily for that very reason. Here is the clip “Tim Cook explains why Apple manufactures products in China,” YouTube, https://www.youtube.com/watch?v=2wacXUrONUY.
The reported ~$4.6M training cost for Kimi K2 Thinking comes from a CNBC article on Moonshot AI’s model launch, which cites an anonymous “source familiar with the matter” and notes that CNBC could not independently verify the figure. See Evelyn Cheng, “Alibaba-backed Moonshot releases new AI model Kimi K2 Thinking,” CNBC, 6 Nov 2025, https://www.cnbc.com/2025/11/06/alibaba-backed-moonshot-releases-new-ai-model-kimi-k2-thinking.html.
The quotation from Jensen Huang comes from a statement posted by Nvidia on X, see here https://x.com/nvidianewsroom/status/1986221177099194484.