Nvidia and AMD to hand 15% of China chip revenue to US

The narrative around the AI chip war used to center entirely on national security. Now, it looks a lot more like a government toll booth.

Nvidia and AMD have agreed to hand over 15% of their revenue from Chinese AI chip sales directly to the U.S. government in exchange for export licenses, according to the Financial Times and remarks by President Donald Trump. The deal primarily covers scaled-down processors tailored for the Chinese market, specifically Nvidia’s H20 and AMD’s MI308.

During a press conference, Trump revealed some behind-the-scenes bargaining. He claimed he originally demanded a 20% cut from Chinese sales, but Nvidia CEO Jensen Huang managed to negotiate a 5% discount.

From strict bans to revenue cuts

This truce follows months of escalating pain for American chipmakers. After Washington clamped down on advanced GPUs in 2022, Nvidia and AMD introduced modified chips designed to stay just beneath performance thresholds. Earlier this year, however, the Trump administration widened the ban to include the H20 and MI308.

The fallout was immediate. Nvidia warned the blockade would cost it $8 billion in a single quarter, while AMD projected a $1.5 billion hit to full-year sales. Huang later revealed that Nvidia’s market share in China plunged from 95% to roughly 50% after H20 shipments were halted.

A face-to-face meeting between Huang and Trump reportedly paved the way for the reversal. Nvidia had already pledged up to $500 billion in domestic data center investments, and Commerce Secretary Howard Lutnick noted that the shift was tied to broader trade talks with Beijing over critical rare-earth elements.

Trump dismissed concerns over the hardware, calling the H20 "obsolete" and claiming China already has comparable silicon. He also hinted that Nvidia might soon be allowed to ship modified versions of its next-gen Blackwell series to China, provided the chips are "enhanced in a negative way" by stripping away 30% to 50% of their performance.

Financial stakes remain massive. Hargreaves Lansdown estimates that resuming H20 sales could inject $15 billion to $20 billion into Nvidia’s top line this year alone. Meanwhile, Nvidia also launched the RTX PRO, a processor for industrial sectors that fully adheres to current export rules.

Why it matters

For enterprise customers and developers, the resumption of H20 deliveries offers a vital lifeline for AI inference and smaller clusters. But nobody should mistake this for stability. The H20 delivers around 4 petaflops compared to the 20 petaflops generated by an uncut B200. Even if degraded Blackwell parts arrive later, relying on hardware whose export status depends on geopolitical negotiations remains a massive liability.

For the broader tech ecosystem, Washington has effectively shifted AI containment policy from ideological gatekeeping to transactional leverage. Forrester analyst Alvin Nguyen pointed out that this effectively reverses recent efforts to isolate China's AI ecosystem, turning cutting-edge technology into a bargaining chip in trade negotiations. National security hardliners are already pushing back, urging the Commerce Department to reconsider.

From our perspective, Huang played an impossible hand brilliantly. Giving up 15% off the top hurts, but losing the Chinese market entirely would have handed domestic alternatives an uncontested runway to maturity. The real danger lies in the precedent: when export licenses become dynamic revenue-sharing schemes, predicting the long-term cost of doing business becomes virtually impossible.