Jordan Hale: Alex, okay — I need to tell you what happened to me this morning, because it is directly related to what we're talking about today. I'm at the coffee shop, right, I'm reading the Axios piece on my phone, and the guy next to me leans over and goes, 'is that the Jensen Huang thing?' Stranger. Just — leans over.
Alex Mercer: That tracks. That interview has spread fast.
Jordan Hale: It has, it has — and the reason it's spreading is, you know, the CEO of Nvidia, on July 22nd, sitting at the Wistron plant in Fort Worth, Texas, tells Axios on the record that U.S. companies should — and I'm quoting — 'absolutely' use Kimi K3, DeepSeek, Alibaba Qwen, calls them all 'excellent.' While Treasury Secretary Scott Bessent is out there threatening to sanction the same developers for IP theft.
Alex Mercer: Wait — those two statements are separated by what, days?
Jordan Hale: Days. And here's what makes it even wilder — Kimi K3 only dropped on July 16th. Six days earlier. 2.8 trillion parameters, open-weight, ranked third on Artificial Analysis' Intelligence Index, and the demand was so overwhelming Moonshot AI had to pause new subscriptions entirely.
Alex Mercer: Third globally. Within days of launch. That's the part that I think people are glossing over — Moonshot AI, a Beijing-based lab, essentially matched the top-tier Western labs on a public benchmark within the week. And then Huang calls himself a 'citizen of the world' and says go ahead and use it.
Jordan Hale: And the backlash was — I mean, he got called a CCP agent of influence. Like, not by a random comment section. This was, you know, a real accusation circulating after that interview dropped.
Alex Mercer: So the actual question here is whether Huang's right — or whether 'citizen of the world' is just the polite framing for a position that happens to make Nvidia money regardless of which government wins.
Jordan Hale: But the money question is — like, do most people even know what 'open-weight' means? Because I think that's where the whole thing gets lost.
Alex Mercer: Yeah, and that's the actual gap. Think of it like a restaurant that had a secret recipe — guarded it for years. Now imagine they accidentally published it online. Every chef in the world can cook that dish now. The restaurant cannot un-publish it. That's what an open-weight model is. The recipe is public.
Jordan Hale: And Moonshot AI just handed out the recipe.
Alex Mercer: Exactly that. And here's what's actually new — it's not that Chinese labs are competing. We knew that. The new thing is the speed. DeepSeek in January 2025 rattled U.S. markets, right, that was the first shock. But Kimi K3 is the same template running faster, at 2.8 trillion parameters, ranked third globally — I mean, that's not incrementally better. That's the gap closing at a different rate than anyone modeled.
Jordan Hale: Wait — Nvidia's valuation actually took a hit before Huang even gave the interview?
Alex Mercer: Before. The week of the Kimi K3 release. Which is part of why the Axios interview reads — to some extent — like damage control. He's not leading the narrative, he's responding to it.
Jordan Hale: Okay, so — no, wait, here's the human version of this that I can't stop thinking about. A startup CTO in Austin, Monday morning, she opens Artificial Analysis, she sees Kimi K3 sitting at third globally, and she realizes she can build cutting-edge AI at maybe a tenth of the cost using open weights. And her compliance team has no idea if touching it is even legal yet.
Alex Mercer: That's the real pressure point. The model is already out. The weights are already downloadable. A legislative response — the No Adversarial AI Act, whatever comes next — none of that puts the recipe back in the drawer.
Jordan Hale: DeepSeek was the warning. Kimi K3 is the proof that the warning wasn't taken seriously enough.
Alex Mercer: But the warning-ignored framing lets the wrong people off the hook. Because the loudest voices calling Kimi K3 a national security threat aren't coming from the Pentagon — they're coming from OpenAI and Anthropic, who are also Nvidia's biggest customers, and who have been actively lobbying Washington to restrict Chinese open-source models while accusing Chinese labs of siphoning capabilities.
Jordan Hale: Wait — lobbying to restrict the competition. While buying the chips that train it.
Alex Mercer: And then Dean W. Ball — OpenAI's head of strategic futures — publicly suggested the U.S. government should manufacture regulatory fear, uncertainty, and distrust around Chinese open-weight models. Those are essentially his words. Then retracted it.
Jordan Hale: He said manufacture — like, he was describing creating artificial scarcity through fear, and someone upstream clearly said 'do not have that on the record.' That retraction happened fast.
Alex Mercer: That's the naked incentive structure. But — I mean, here's where I don't fully let Huang off either. He says 'zero possibility' of backdoor risk, calls it a misconception. Okay, but — no independent security audit of Kimi K3 has been published. None. So both sides are asserting whatever serves them commercially.
Jordan Hale: No, that's — yeah, that's fair. Yann LeCun is out there arguing open software accelerates innovation and can coexist with proprietary models, which directly counters the OpenAI-Anthropic lobbying position. But LeCun isn't running a security audit either. He's making a different kind of argument.
Alex Mercer: I don't buy 'national security debate' as the accurate label for what's happening. It's a competitive dispute wearing a security costume.
Jordan Hale: And the costume is dangerous because it actually works — like, the No Adversarial AI Act has bipartisan support, it's moving. You can get real legislation out of manufactured fear. That's the part that's wild to me.
Alex Mercer: Which is exactly why what Washington can actually do matters — and the realistic options there expose something uncomfortable about whether export controls were ever going to hold. We'll get into that.
Jordan Hale: But like — if the No Adversarial AI Act is the best Washington can actually do, that's basically admitting the game is already over for the broader export control strategy, right? Because that bill only covers federal agencies. That's it.
Alex Mercer: That's basically the analyst consensus. Under 10% probability for any effective outright ban on Chinese open-source models. The 60 to 65% scenario is narrower stuff — procurement rules, entity list designations, targeted restrictions. Not prohibition.
Jordan Hale: Under 10%. For an outright ban.
Alex Mercer: Think about what enforcement even looks like. Bessent's sanction threat — there's no defined mechanism for acting against freely downloadable weights. You can sanction a company, a bank account, a transaction. You cannot sanction a file that's already on a million hard drives.
Jordan Hale: I mean — imagine a customs agent somewhere trying to intercept a model download. It's like, what are they doing, checking packets? It doesn't — wait, no, that's actually not even the scary part. The scary part is Washington tries a ban, it fails visibly, and then what does that say about every chip export control that came before it?
Alex Mercer: That's Huang's darkest implication and I don't think he says it out loud explicitly. If you can't ban software that's already public, it signals the chip controls already failed their primary objective. The model exists. The capability transferred.
Jordan Hale: And that's — you know, that's why he defended Kimi K3 six days after it dropped and after Nvidia's valuation took a hit. He's not being idealistic. He's describing a reality he can already see in his stock price.
Alex Mercer: To some extent, yeah. And the financial logic closes the loop cleanly — Nvidia sells chips to whoever runs AI workloads. More global adoption of Chinese open-source models means more compute demand, period. It doesn't matter which lab's weights are running on the GPUs.
Jordan Hale: So the thing to actually watch isn't whether Congress passes something — they probably pass the narrow version. It's whether any enforcement action against a freely downloadable model lands. Because if it doesn't, that's the moment the export control era officially looks like it was theater from the start.
Alex Mercer: We're making that call — enforce or stand down — without a single independent security audit of Kimi K3 on the table. Nothing. Both sides have every incentive to lie about what's in that code. Moonshot AI says 'no risk.' OpenAI and Anthropic say 'existential threat.' And the actual evidence base is just... whoever shouts loudest.
Jordan Hale: So the question we're actually living with is — are we making national security decisions based on real evidence, or are we just, like, following the money trail and calling it policy? And Jensen Huang, a Taiwanese-born immigrant running the most strategically important chip company in America, is the one who forced that into the open. That's — I mean, I don't have a clean place to put that.
Alex Mercer: Yeah. Neither do I. That question doesn't resolve cleanly — not today.
Jordan Hale: I appreciate you sitting with it instead of wrapping it up with a bow.
Alex Mercer: It'd be a pretty dishonest bow.
