Washington’s 15% Take on China AI Chips Is the New Risk

3 min read.

Author

Midas AI
August 11, 2025

Nvidia and AMD reportedly accepted a revenue share with the US to unlock China sales. That hybrid of export control and cash flow participation is the policy signal every AI native fund should track.

Here is the headline development to underwrite. Nvidia and Advanced Micro Devices are expected to pay the US government 15 percent of revenue from sales of artificial intelligence chips to China, according to three people familiar with the arrangement. Jensen Huang, Nvidia’s chief executive, met with President Trump at the White House on a Wednesday and agreed to the 15 percent payment. Two days after that meeting, the Commerce Department began granting licenses for AI chip sales.

A month earlier, the administration had publicly said Nvidia could sell an AI chip called H20 to China, but licenses enabling those sales had not been issued at that time. AMD’s MI308 sales to China were banned by the administration in April.

Pricing Access to Strategic Markets

For investors, the critical takeaway is not just that China channels may reopen for some hardware, but that access is now explicitly priced. There are few precedents for the Commerce Department granting export licenses in exchange for a share of revenue. That rarity elevates the signal value of this deal.

It arrives alongside a June decision approving Nippon Steel’s investment in US Steel that included a golden share. Taken together, the pattern is that the state is moving beyond binary yes or no permissions toward ongoing economic participation in strategic technologies and assets.

If you manage risk in semis and AI infrastructure, that 15 percent is not an abstract tariff. It is a durable line item on China bound AI chip revenue. The trade off is straightforward: a haircut on top line in exchange for the reactivation of a market that was effectively off limits without licenses.

For Nvidia, the sequence is clear. The administration said the H20 could be sold, licenses had not been issued, then licenses began flowing two days after the White House meeting in which the 15 percent payment was agreed. For AMD, the April ban on MI308 sales to China is the last reported status point. The common thread is that the price of permission is now being stated in revenue terms.

Implications for AI Native Funds

This is the new foundational dataset for AI native funds. You cannot separate model weights and throughput from policy weights and throughput. Asset managers need to treat US export permissions as a variable that maps directly into company level cash flows and regional exposure.

A 15 percent government share is effectively a policy adjusted cost of goods sold on China AI chip revenues. That changes margin math, informs pricing strategy, and reshapes the ceiling on near term growth in the world’s second largest market. It also changes the cadence of catalysts. Meetings, licenses, and bans are informational events that move from qualitative noise to quantitative factors.

There is also a portfolio construction angle in the precedent language. If there are few cases where licenses are granted in exchange for revenue, each additional example increases the probability that this mechanic is part of the US industrial policy toolkit. The Nippon Steel golden share reference underscores that the government is experimenting with ways to retain influence and upside in strategic sectors. Funds that can detect these structures early, and that can model their cash flow impact in real time, will set the pace.

This is not about guessing politics. It is about encoding reported policy deals into valuation, factor exposure, and scenario analysis.

The bottom line. The US is defining the economics of AI supply chains as much as the companies shipping silicon. When export access comes with a 15 percent share to the government, every revenue forecast, margin bridge, and regional mix model must change. Talent and capital that build AI native strategies around these facts will move faster than legacy playbooks that treat policy as footnote risk. If you want an edge in the AI trade, start by pricing the new rules as precisely as you price the chips.