China's AI Gambit vs. America's Data Center Backlash: The New Tech Cold War

**Zack Cohen** (0:00)
It seems like China, our authoritarian adversary, has embraced the open-source ethos, probably not for the same reason that America, per se, would embrace that ethos.

**Zack Shapiro** (0:12)
China just doesn't have access to the very best training runs, and so it's not at the frontier, and so the game theoretic optimal strategy is to do open-source in response to that, and my guess is that that will flip over time.

**Zack Cohen** (0:29)
We're back with another week of the Bitcoin Policy Hour. Ken, are we going to have to rebrand?

**Ken Egan** (0:35)
I think we agree that it's, in addition to being the world's number one Bitcoin Policy Podcast, it is also now the world's number one Freedom Tech Policy Podcast.

**Zack Cohen** (0:44)
Yeah, it sounds like we'll have to talk to our friends on the Bitcoin Magazine side, but potentially a rebrand is in order, so I know, I know. Big talk. Okay, as usual, Zack Cohen here with the Bitcoin Policy Institute, joined by Ken Egan, our head of Government Affairs, and Zack Shapiro, our head of Policy. And we've got another loaded docket. We're recording on Monday, July 20th. So we've had the weekend to marinate. I was doing some tweeting this weekend. I've been active on the interwebs, so I've got some stuff I want to dig into.
But yeah, I think the biggest news of last week was the Moonshot Kimi K3 release. And for folks who haven't been keeping up, basically, Kimi K3 is to date.
And I'm not sure about the comparison with the other sort of large frontier level open source model that was announced maybe a week and a half ago. But to date, probably the single most powerful open weight model coming out of a Chinese lab. And the talk on Twitter was, you know, hey, we thought the open source, open weight models were at least a year behind, maybe even two years behind. And on some of the benchmarks that they're testing the models on, Kimi K3 actually outperformed both Fable and ChatGPT 5.6 Sol. So Zack, curious on your initial thoughts here, I think there's a lot to unpack sort of beyond just the sort of shock of the news itself, but curious, you know, your thoughts on this coming out and also the reaction that it seemed to have garnered on Twitter.

**Zack Shapiro** (2:29)
Yeah, I think a lot of the reaction on Twitter is related to the theme we discussed last time about sort of model sovereignty and the question of like, is open source better for enterprise than being reliant on a specific frontier lab to buy from them directly.
Kimi is not there yet, right? I don't think it's like really seriously competing with like Able or GPT 5.6 in terms of like actual frontier capabilities. And so I don't think we have hit that moment. But there's a question about how quickly will these models get there. There's a other related question about the way that this model was created and distillation and how powerful distillation can be and how capable a model can you have sort of run on local hardware, which may be as helpful for data sovereignty? Like that's one question. But then there's also, if you can get these things small enough, you can have models embedded in basically every kind of firmware. You can have a very powerful local AI in your car, on your phone, on your laptop. And you're not reliant on Internet access and the cloud.
You could just have sort of intelligence diffuse directly into everything. And that is a potentially very interesting future, but with its own sort of policy question. And then there is the direct policy angle of like, is the government going to ban Chinese models? So happy to take in whatever direction. But I think those are sort of the three main angles I see.

**Zack Cohen** (4:12)
Yeah, I think we'll probably end up touching on all those in some way or another.
But it seems like the government angle is probably the most interesting one to start on right now. Ken, you flagged a tweet for us actually in our Slack channel. This was from Dean W. Ball. And it sounds like there was some follow up to that, which I didn't get a chance to look at. But let me just read this real quick, and then we can unpack it together. The tweet says, I would guess that the Trump administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open weight Chinese models. You don't need to ban open source, one of the dumber motifs of AI policy discussion. You just need to direct every agency to issue soft law that creates FUD, a Federal Reserve Advisory Bulletin, found that there may be backdoors in Chinese AI models.