The US Treasury Department is busy hunting for "watermarks" of American intellectual property within Chinese neural networks, but Jensen Huang is convinced that the real danger is technological blindness, not borrowed code. While Scott Bessent threatens Beijing with sanctions for data theft on Fox Business, the Nvidia CEO is publicly defending the right of American businesses to use Chinese open-source models. The conflict of interest is glaring: while regulators attempt to build a digital cordon, the primary beneficiary of the AI gold rush understands that any blockade will tank his market cap faster than industrial espionage ever could. The situation borders on the absurd as officials attempt to find traces of distillation in models with trillions of weights, while the market has already voted with its dollar for efficiency over a codebase's country of origin.
In this saga, the chasm between political rhetoric and engineering reality is staggering. Washington is trying to sell a narrative that Chinese AI is a mere derivative product that can be halted by banning the import of model weights. The data suggests otherwise. The new Kimi K3 model from Beijing-based Moonshot AI, boasting 2.8 trillion parameters, hasn't just caught up to Western rivals—it has outperformed GPT-5.6 Sol and Claude Fable 5 in applied frontend coding. This is no longer a cargo cult; it is technological parity achieved under the weight of sanctions. When investors see a Chinese product that is 40% cheaper with comparable quality, they begin asking uncomfortable questions about the wisdom of multibillion-dollar investments in closed American labs.
The Economics of Open Source vs. Fear
Huang’s logic is cold, clinical, and commercially flawless. For him, there is no such thing as "bad" AI if that AI requires compute to run. The release of Kimi K3 triggered a sell-off in chipmaker stocks as Wall Street feared the devaluation of existing data centers. Why pay for closed proprietary models when free, open-weights models are coming out of China? Huang believes the market is miscalculating again, just as it did in January 2025 following the release of DeepSeek R1. His strategy is built on the premise that the availability of models only widens the funnel for hardware consumption.
"Free AI should be great news for hardware. Free AI should be great news for chips. Free AI should be great news for data centers,"
Huang told Axios, effectively admitting that Nvidia benefits from any expansion of neural networks, even those built by a geopolitical adversary. He counters Treasury's national security arguments with technical pragmatism: open models are inherently safer than closed ones because they can be audited in isolated sandboxes. From the Nvidia CEO's perspective, a single dominant model represents a single point of failure for the entire ecosystem, whereas a diversity of solutions creates resilience.
Distillation as a Foundation or Theft
Washington's primary grievance is the "siphoning" of American capabilities through distillation—training models on the outputs of more advanced systems. Scott Bessent calls this theft. However, the industry is far from unified on this point. Even OpenAI’s Dean Ball, who previously warned of "AI communism," admitted that Kimi K3’s performance cannot be explained by simple copying. Huang goes further, asserting that training on existing knowledge sources is the very foundation of intelligence. In his view, punishment should target specific privacy violations rather than the use of training methodologies.
Furthermore, the Nvidia chief is calling on Anthropic to open access to their restricted cyber-model, Claude Mythos, arguing that gatekeeping such developments harms the United States. The irony is that Nvidia already has access to Mythos through the Project Glasswing consortium. Huang is essentially proposing to strip Anthropic of its exclusivity, a move that looks like a final attempt to blur the lines between proprietary software and the public commons.
Nvidia is architecting a world where software costs zero and infrastructure costs trillions. In this paradigm, Chinese state-of-the-art open source is Huang’s best ally and a regulator's worst nightmare. Businesses now face a choice: pay a premium for a "clean" American product or implement high-performance Asian tools under the threat of sanctions. Attempting to ban the diffusion of technology in the internet age is like fighting the rain—there is a lot of noise, but everyone is going to get wet regardless.