A stunned semiconductor industry is grappling with an unprecedented U.S. government demand: Nvidia and AMD must hand over 15% of their China AI chip revenues as a condition of export licenses. The arrangement, disclosed in mid-August, allows the two chipmakers to resume shipments of specific downgraded accelerators—Nvidia’s H20 and AMD’s MI308—to the world’s second-largest economy after months of frozen sales. It marks the first time Washington has directly monetized national-security export controls, turning a licensing process into a revenue-sharing instrument. The move has instantly triggered a firestorm of legal, constitutional, and strategic debates that will likely unfold in courts, Congress, and boardrooms across the globe.

Background: How Export Controls Brought Sales to a Halt

Since 2022, the U.S. has systematically tightened export controls on advanced AI accelerators to slow China’s progress in high-performance computing. Nvidia and AMD engineered special versions of their data-center GPUs to comply with the rules—the Hopper-based H20 and Instinct MI308, respectively—and sold them aggressively to Chinese cloud providers and AI labs. But in April 2025, the Commerce Department abruptly required export licenses for those very products, throwing the market into chaos. Nvidia and AMD were forced to book billions in inventory write-downs as shipments ceased. The sudden freeze left both companies scrambling for policy clarity, and by August the White House had brokered a controversial breakthrough: licenses would be granted, but only with a 15% revenue cut flowing to the U.S. Treasury.

What the Deal Covers: H20, MI308, and the 15% Mechanism

The deal specifically applies to Nvidia’s H20—a compute-resource-limited Hopper module designed to stay below previous performance thresholds—and AMD’s MI308, an Instinct accelerator similarly tailored for export compliance. Neither chip delivers the full capability of the companies’ flagship processors, such as Nvidia’s Blackwell series, but they remain potent tools for large-scale AI training and inference.

The 15% condition is embedded in the export licenses themselves. It is not a per-unit tariff or a statutory tax; rather, it is a negotiated requirement that the companies remit a share of the revenue from each licensed sale to the U.S. government. The payments are tied directly to China transactions of the specified SKUs, creating an ongoing fiscal stream that could amount to billions of dollars annually. Reporting indicates that both Nvidia and AMD agreed to the terms as the price of re-entering a market that represents a substantial slice of their overall revenues.

Financial Fallout: From Multibillion-Dollar Write-Downs to a Revenue Lifeline

The April licensing shock inflicted severe financial pain. Nvidia disclosed inventory and related charges of between $4.5 billion and $5.5 billion tied to the H20 program, while AMD warned of potential charges up to $800 million and an annual revenue hit of around $1.5 billion. These staggering figures made the choice for both companies simple: sell into China with a 15% haircut, or leave stranded inventory to rot and cede market share indefinitely.

China remains a critical market. In 2025, Nvidia’s Chinese sales were estimated at roughly $17 billion, accounting for a low-to-mid-teens percentage of total company revenue. AMD’s China exposure sat around $6 billion, or about a quarter of its top line. Even a partial recovery of those sales, net of the government’s cut, eases the immediate earnings pressure and prevents a permanent loss of customers. Wall Street analysts largely viewed the deal as a net positive for the chipmakers, given the alternative of a complete China blackout.

Constitutional Minefield: Is It an Export Tax?

Legal scholars and former officials immediately trained their sights on the U.S. Constitution’s explicit ban on export taxes. “The arrangement walks right up to the line—if not over it,” said one administrative law expert. The White House insists the 15% payment is a licensing condition, not a tax, but the practical effect is a government levy on overseas sales. That could run afoul of Article I, Section 9, which states, “No Tax or Duty shall be laid on Articles exported from any State.” If courts view the requirement as a functional export tax, the entire program could be struck down.

Beyond the constitutional question, administrative law concerns abound. The Export Control Reform Act gives Commerce authority to limit exports for national-security reasons, but using that authority to extract revenue is novel. Critics argue the department is exceeding its statutory mandate and acting arbitrarily. Labels matter: is the payment a fee, a fine, a tax, or a quid pro quo? Each carries different legal implications, and the wrong characterization could doom the policy.

National-Security Paradox: Containment vs. Commerce

Export controls are designed to deny adversaries advanced technology. This deal flips the script: it permits controlled sales while generating income for the U.S. government. Proponents claim it’s a savvy move—allowing the flow of downgraded chips keeps Chinese AI developers locked into American software ecosystems, such as Nvidia’s CUDA, and provides Washington with leverage. The U.S. can maintain influence over the AI supply chain while siphoning away a slice of China’s computing budget.

Detractors see a dangerous erosion of deterrence. Even intentionally limited hardware accelerates China’s learning curve, and the deal signals that export restrictions can be bought for a price. It may also provoke Beijing to redouble its efforts to create domestic alternatives. Within days of the announcement, Chinese state guidance reportedly urged government-affiliated enterprises to avoid U.S. chips in sensitive projects, a move that could shrink the addressable market for Nvidia and AMD.

Winners and Losers

For Nvidia and AMD, the calculus is straightforward. They reclaim access to a high-margin, high-demand market and can mitigate the financial carnage of the earlier export freeze. Inelastic Chinese demand for AI compute gives them pricing power, and many analysts see the deal as earnings-accretive despite the revenue share.

The U.S. Treasury gains a novel revenue stream that could reach multiple billions annually if volumes recover. But the broader ecosystem risks fragmentation. Smaller chipmakers and software vendors may find themselves shut out if pay-to-export becomes normalized, as they lack the bargaining power of the duopoly. National-security hawks worry that the deal dilutes the blunt effectiveness of controls, and allies who have cooperated on multilateral restrictions may see the U.S. as monetizing a common strategic interest.

Political and Diplomatic Shockwaves

Congress is already gearing up for oversight hearings. Both Republicans and Democrats have flagged constitutional and security concerns, and subpoenas may follow. China’s response has been swift and predictable: state media and procurement guidance portray the U.S. chips as potentially compromised, urging domestic substitution. Beijing is consolidating its national champions—Huawei, Cambricon, and others—and using the U.S. deal as a political cudgel to accelerate indigenous development.

Allies are watching closely. The Netherlands, Japan, and South Korea, which have aligned their export policies with Washington’s, may question why the U.S. is now extracting cash from the very controls they joined. Multilateral cohesion, already fragile, could fracture if partners perceive the arrangement as a unilateral cash grab rather than a genuine security measure.

A lawsuit seems inevitable. Potential plaintiffs include civil-liberties groups, state attorneys general, or even the companies themselves if they later seek to escape the burden. The argument will likely hinge on the constitutional export-tax ban and on administrative overreach. The government will invoke national-security deference, but the high-stakes constitutional question may force courts to engage with the merits rather than dismiss the case.

If plaintiffs secure a preliminary injunction, the licensing program could stall, reintroducing uncertainty for chipmakers. The Supreme Court may ultimately need to decide whether conditioning an export license on a revenue payment is fundamentally a tax. Such a ruling would redefine the boundaries of executive power over trade and national security.

Technical Realities: Downgraded but Still Potent

The H20 and MI308 are not top-of-the-line chips. Nvidia and AMD have implemented technical caps—reduced interconnect bandwidth, memory limits, and firmware restrictions—that make the accelerators less suitable for cutting-edge military or high-performance computing workloads. However, they remain powerful for mainstream AI tasks, and Chinese developers already have deep experience with the software stacks. That familiarity entrenches American technology even as it feeds the competition. Nvidia’s CUDA ecosystem, in particular, creates a moat that rivals find hard to cross.

Critical Analysis: Strengths, Weaknesses, and Likely Outcomes

From a realpolitik perspective, the deal recovers massive market access for U.S. firms, converting a blunt national-security instrument into a dual-use tool of revenue and influence. It keeps Chinese AI labs dependent on U.S. software and may slow an all-out shift to indigenous platforms in the very short term. For Nvidia and AMD, it’s a pragmatic rescue from a self-inflicted inventory nightmare.

But the risks are profound. The constitutional exposure is real and could unravel the arrangement entirely. Monetizing licenses undercuts the credibility of export controls as a security measure, potentially emboldening both adversaries and domestic critics. The diplomatic fallout could isolate the U.S. from allied export-control regimes. And Beijing’s swift push for self-reliance suggests the long-term strategic cost may outweigh the near-term revenue gain.

The most plausible near-term outcome is a partial restoration of China sales, accompanied by legal challenges and congressional theater. Over the medium term, expect a messy hybrid: continued controlled shipments of downgraded hardware, tougher diversion enforcement, and accelerated U.S. and Chinese industrial policies that reshape the semiconductor landscape.

What to Watch

Agency rulemaking and administrative records will reveal the legal reasoning behind the license conditions. Congressional hearings—likely starting within weeks—will test the political viability of the deal. China’s procurement guidance to state-owned enterprises will determine actual demand volumes. And enforcement mechanisms against diversion will determine whether the U.S. can collect its share without hemorrhaging technology. The deal is a grand experiment, and the world is watching whether it becomes a new model for great-power competition or a cautionary tale of overreach.