Huang: US-China AI Gap Is Dialogue, Not Chips

Jensen Huang, the chief executive of Nvidia, has a diagnosis for the US-China technology standoff that most policymakers are not discussing: the two countries are not talking to each other. Speaking on the Dwarkesh Podcast in mid-April, Huang laid out a case that the real fracture in the AI competition between Washington and Beijing has nothing to do with silicon or semiconductors. It is, he argued, the absence of direct scientific dialogue—the kind of researcher-to-researcher exchange that once existed between the superpowers during less adversarial times.

The conversation arose amid a broader debate about whether American companies should be allowed to sell advanced AI chips to China. Huang's counterargument was blunt: export controls alone cannot slow China's progress, and the attempt to do so may actually accelerate the very outcome the US is trying to prevent. China already possesses the ingredients it needs to build world-class AI systems. The country has abundant energy, a massive domestic chip manufacturing base, and roughly half of the world's AI researchers. "When you have an abundance of energy, it makes up for chips," Huang said. Restricting chip sales, in his view, misses the actual constraint.

The numbers tell a story that supports his concern. Before sanctions took hold, Nvidia commanded 95 percent of China's domestic AI chip market. As of April 2026, that share had fallen to around 55 percent. Meanwhile, China's own domestic AI chip market share had climbed to 41 percent from nearly zero. Huawei, the Chinese technology giant, shipped roughly 812,000 AI chips in 2025 alone—a record. The export controls, rather than freezing Chinese capability, appear to be accelerating Chinese self-sufficiency.

Huang's deeper worry extends beyond the next generation of cyberattacks or exploits. He is concerned about the architecture of global AI itself. If Chinese AI models—many of them open source—become optimized for Huawei hardware rather than Nvidia's CUDA software stack, the American technological standard loses its gravitational pull over the world's AI development. Two separate ecosystems would emerge: one open-source and running on Chinese chips, the other closed and running on American ones. "It would be extremely foolish to create two ecosystems," Huang said. "I think that would be a horrible outcome for the United States." The risk is not military in the traditional sense. It is the slow erosion of American influence over how the world builds and standardizes artificial intelligence.

His position puts him at odds with voices like Dario Amodei, the chief executive of Anthropic, who has argued that selling AI chips to China is akin to "selling nuclear weapons to North Korea." When pressed on that comparison during the podcast, Huang was unsparing: "Comparing AI to anything that you just mentioned is lunacy." He does not argue for giving China an unrestricted pass. Rather, he proposes that the US stay ahead through superior technology, ensure American compute capacity is abundant and accessible, and pursue the kind of international research agreements that might establish shared guardrails on the most dangerous AI capabilities—agreements that would require the two countries to actually sit down and negotiate.

What Huang is describing is a return to an older model of great-power competition: one where scientific exchange and mutual understanding coexist with strategic rivalry. The absence of that dialogue, he argues, is the real vulnerability. Export controls are a blunt instrument that may feel decisive but cannot address the underlying problem. Two countries building AI in isolation from each other, each optimizing for its own hardware and standards, each unable to influence the other's safety practices—that is the outcome Huang fears most. Whether policymakers in Washington will hear that argument, or whether the logic of containment will prevail, remains an open question.