Table of Contents
Nvidia H200 exports to China are now back on the table, with Donald Trump announcing that the United States will let Nvidia ship its powerful H200 artificial intelligence chips to approved Chinese customers, so long as Washington gets a 25 percent cut of the revenue. It sounds like a business-school case study in “win win.” In practice, it is a test of how much American democracy is willing to rent out its technological edge to an authoritarian rival.
Trump’s decision reverses the Biden-era posture that treated advanced AI chips as strategic infrastructure and kept the H200 out of China’s reach. According to the Financial Times, the move would allow Nvidia’s H200 AI processors to be exported to China under a revenue-sharing deal, with the same framework envisioned for AMD and Intel as well, while critics in Congress warn the H200 is “vastly more capable than anything China can make” and a core asset in the AI race between democratic and authoritarian systems Financial Times.
This is not just an export-control tweak. It is a choice about the architecture of power in the 21st century.
Nvidia H200 Exports To China And The New AI Bargain
The Nvidia H200 exports to China deal is built on a simple premise. America can sell the second most powerful AI chips on earth to its primary geopolitical rival, so long as the U.S. Treasury takes a slice.
In Trump’s telling, this achieves three things at once:
- Keeps Nvidia and other U.S. chipmakers dominant in a huge market.
- Throws money at American taxpayers.
- Avoids the messy work of building a principled export regime rooted in rule of law instead of presidential whim.
The H200 is not Nvidia’s cutting-edge Blackwell, yet it is still a supercomputer-grade engine for training large language models, surveillance systems, and military decision tools. When U.S. officials previously blocked these exports, it was not because they hated Nvidia’s margins. It was because they understood compute as a strategic resource.
Trump’s move reframes that. AI compute is still strategic, but now it is strategic revenue.
From a progressive, institutional lens, the most unsettling part is not just the technology transfer risk. It is the structure of the bargain. The federal government is positioning itself as a silent partner in private export deals to an authoritarian state, with legal mechanisms for collecting this 25 percent still hazy. That blurs the line between public purpose and private deal-making in ways that should make constitutional lawyers and anti-corruption watchdogs nervous.
How This Fits Into Nvidia’s Expanding Power
To understand why Nvidia is at the center of this geopolitical drama, you have to see the company as something closer to a public utility of the AI era. Its chips power everything from recommendation algorithms to frontier models and cloud-scale training clusters.
Nvidia is already entangled with U.S. tech giants at home. The company’s deep integration with Amazon’s cloud infrastructure and broader AI stack shows how central its hardware has become to the commercial internet and corporate AI deployments, as explored in this analysis of the Amazon and Nvidia AI integration.
Now add Beijing to that dependency map.
H200 exports, even with vetting by the U.S. Commerce Department, hand Chinese companies a vast amount of compute. That has at least three implications:
- Economic leverage: Chinese buyers will be locked into Nvidia’s ecosystem, which U.S. officials can argue gives Washington visibility and leverage over Chinese AI progress.
- Security leakage: Every shipment risks reverse engineering and domestic copycats. Chinese firms like Huawei already admit they cannot yet match H200 performance, which is precisely why access matters.
- Industrial policy by accident: Instead of a coherent U.S. AI industrial policy, we get ad hoc deals. Export a powerful chip here, design a weaker variant there, pressure allies to adopt controls, then blow a hole in those controls for a 25 percent cut.
For Nvidia shareholders, this is a dream. For democratic policymakers, it should be a warning siren.
Democratic Norms Versus Transactional Geopolitics
Nvidia H200 exports to China sit at the intersection of two competing stories about American power.
In the first story, the United States is a democracy that uses its technological advantages to defend open societies, constrain authoritarian expansion, and uphold a rules-based order. Export controls on advanced chips are imperfect, but they fit this story. They are tools to prevent AI from supercharging surveillance states and military buildups.
In the second story, the United States is a dealmaker-in-chief. National security is something you can monetize. If a transaction creates jobs at home and a revenue stream for Washington, then the security risk is “managed” by definition, often without clear congressional oversight.
Trump’s arrangement leans hard into the second story. It also sidelines democratic institutions in at least three ways:
- Congressional input reduced to noise: Lawmakers in both parties have raised national security concerns and even floated legislation to halt H200 exports for 30 months. Yet the policy signal out of the White House is, effectively, “we already made a deal.”
- Opaque legal frameworks: The FT has reported that officials and companies have struggled to find a clean legal mechanism to route these revenue shares from Nvidia and AMD to the government. That means U.S. policy is out ahead of its own law, instead of the other way around.
- Presidential discretion as system design: Instead of transparent criteria for what AI hardware can go where, we get personalized diplomacy. Trump talks to Xi, consults with favored CEOs, and then announces a new line between “too dangerous to export” and “fine, as long as we get paid.”
Progressives should see this not as a niche export-control story, but as a template. If this logic holds, any future technology with dual-use potential can be normalized through revenue-sharing deals. Hypersonic components. Quantum accelerators. Bioengineering platforms. Just add a sovereign cut.
What This Means For The Global AI Race
From Beijing’s perspective, Nvidia H200 exports to China are a gift, even if Chinese regulators do hedge by limiting access domestically. Access to H200s helps Chinese labs and firms narrow the gap with U.S. AI players, especially in the near term.
From Washington’s perspective, the calculus is messier.
There is a real argument, made by some in the Trump orbit, that making the world dependent on American chips creates long-term leverage. If Chinese AI runs on Nvidia, and Nvidia is subject to U.S. law, then U.S. regulators retain a kill switch. In this frame, controlled exports are better than forcing China to build a fully indigenous stack.
The counterargument is more sobering. Advanced chips are not just another export. They are a force multiplier across surveillance, cyber operations, autonomous weapons, and disinformation. Even if Washington can theoretically cut off supply later, the damage of several years of accelerated Chinese AI development is already baked in.
And then there is the symbolic layer. When the United States reduces a foundational strategic advantage to a revenue-sharing line item, it signals to every other government that values are negotiable. It makes it harder to argue that Western export controls are about democracy instead of market share.
Where Policy Should Go From Here
Nvidia H200 exports to China are now a reality barring a substantial congressional revolt. The question, then, is how to contain the damage and use this moment to push for a more principled AI strategy.
At minimum, a serious response would include:
- Statutory guardrails that define when and how the executive branch can barter export permissions in exchange for direct government revenue.
- A public AI export doctrine that sets clear red lines around military, surveillance, and repression use cases, with enforcement mechanisms that survive electoral swings.
- Stronger multilateral coordination so that U.S. chip policy aligns with European and Asian democracies, instead of being dictated by a single administration’s relationship with Beijing.
- Transparency around enforcement for “approved” customers and the conditions that can revoke that status.
The deeper project is harder. Democracies need to treat compute as infrastructure, not just product. That means connecting export decisions to broader strategies around AI governance, labor markets, and civil rights at home. It means admitting that some profits are not worth the downstream costs.
Trump’s deal with Nvidia is not the end of the story. It is the opening chapter in a long negotiation over who controls the means of intelligence in a world where authoritarian states see AI as a tool for control, not liberation.
The choice in front of the U.S. is stark. It can build a model where democratic institutions constrain technology in service of human rights. Or it can keep improvising deals that rent out its strategic edge, one chip at a time.