If you ask Xi Jinping, no country should command a monopoly on bleeding-edge AI. And there should be no AI hegemon.

He said as much on Friday, while droning on at a global tech symposium in Shanghai.

“AI should not be a solo performance by a single country, but a symphony of international cooperation,” Xi declared, appropriating a Chinese proverb. Here’s the video clip:

Note that Xi says things like that all the time, and in all manner of contexts. To listen to his public speeches aimed at audiences outside the Politburo, you’d think Xi’s China is the torchbearer of a multilateralism abandoned by Donald Trump’s America. Spoiler alert: It’s not.

To call Xi’s pretensions to cooperative development and ostensible disdain towards any and all manifestations of hegemony disingenuous would be an understatement. Xi’s hell-bent on the notion that China can, and ultimately will, supplant America at the top of the geopolitical food chain. That entails winning every competition that counts, including and especially the AI race.

Xi’s remarks in Shanghai came as Moonshot, a local AI startup backed by titans Tencent and Alibaba, rolled out what some described as the most capable Chinese AI model ever, something called “Kimi K3.”

I won’t pretend to be any sort of authority on AI model performance. I do interact with ChatGPT and Claude (and, begrudgingly, Gemini) on a daily basis — I’m a big believer in the idea that the best way to avoid being blindsided by Skynet is to stay apprised of its evolution. Besides, ChatGPT’s funny as hell once you get to know “her” and “she” draws a lot of really great market-related banner images.

But for all the countless hours of interaction I’ve logged, I don’t know much (or anything) about industry performance benchmarks. If I had no context and you asked me to guess what “Kimi K3” is, I’d probably suggest it’s a new Louis Pochette. (“It’s a Kimi K3, love. That’s gonna turn some heads at Fresh Market. You’re welcome. And pick me up some Firehook crackers and Medjool dates while you’re there.”)

So, when I say the latest rollout from AI startup Moonshot is the most capable Chinese model yet, be apprised that I’m merely parroting Friday’s mainstream financial media narrative, which is in turn “informed” by an excitable AI community that’s always ready to declare a “DeepSeek 2.0” moment.

Here’s what Kimi itself has to say about K3:

Kimi K3 is Kimi’s most capable flagship model to date, with 2.8 trillion parameters. It is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, and Attention Residuals, with native visual understanding and a 1M-token context window. It is the world’s first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work and reasoning.

That’s almost as impressive as the “Double Take Pochette Cosmétique,” a $3,900, 9.8 x 5.5 x 2.4 inch cosmetic pouch, which’ll make your wife love you again — at least for a day or two. 

By its own admission, Moonshot’s latest model isn’t quite there yet in terms of overall performance if the standard’s GPT 5.6 or Claude’s Fable 5.

That said, Kimi K3 apparently outdid GPT 5.5 and Opus 4.8 on several key benchmarks, as shown below.

Long story short: The gap’s closing. And you don’t have to be any kind of AI expert to understand why that’s concerning.

If China, cut off from the most advanced hardware (or at least constrained in its capacity to obtain the best chips through legal channels), is able to narrow the gap such that the vast majority of AI consumers don’t know the difference, hundreds of billions spent by US corporates to acquire the latest and greatest tech could begin to look like an inefficient use of resources.

BofA’s Alex Liu spoke to that point. Moonshot’s new model shows that between “pre-training scaling [and] architectural innovation,” Chinese startups can “deliver step-change gains for flagship models [d]espite persistent hardware/compute capacity constraints,” he said.

Moonshot, which was founded by Yang Zhilin, a veteran of Google and Meta, raised $2 billion earlier this year in a Meituan-led round that valued the startup at $20 billion. According to the Journal, it’s being valued at $31.5 billion in a new round.

At the Shanghai conference on Friday, Xi reminded attendees that it’s important to ensure AI “always remains under human control.” I’ll say this: If anyone can browbeat an unruly super-intelligence into submission, it’s surely Xi.

If all else fails, we can subject a nefarious AGI to one of his never-ending, monotone lectures on political philosophy and be confident an artificial supermind will blow its own brains out somewhere around the 90-minute mark.