NVIDIA and AMD agree to Trump's "15% AI chip tax," in exchange for importing H20 and MI308 to China, with gross profits going directly to Washington.

Nvidia and AMD agree to pay 15% of sales from China AI chips to the U.S. government in exchange for export licenses, setting a precedent for a "chip tax" that impacts the global supply chain and national security balance (Background: In the AI talent competition where cash is flowing, Sam Altman says tech giants are betting on the chosen ones to achieve AGI breakthroughs) (Background: Grok 4 is now available for free use, Elon Musk's xAI faces off against GPT-5) In the context of escalating U.S.-China tech tensions, Nvidia ( and AMD ) have suddenly agreed today (11) to pay 15% of sales from their AI chips H20 and MI308 in the Chinese market to the U.S. government in exchange for long-restricted export licenses. This transaction, referred to by the market as the "chip tax," has drawn Washington, Silicon Valley, and Beijing into a new equation: Can fiscal revenue, technology control, and market survival coexist? Overview of the transaction: The 15% "chip tax" sets a precedent. Over the past two years, the U.S. has ramped up export controls on high-end chips to China, forcing Nvidia and AMD to release downgraded versions H20 and MI308 specifically for the Chinese market. Even so, the two chips faced delays in the approval process this year, causing shipping stagnation. The latest agreement finally unlocks inventory: the two companies will directly transfer 15% of their gross revenue from each Chinese order to Washington, amounting to an estimated $1 billion per month. This model differs from traditional tariffs and one-time fines; instead, it embeds "taxes paid only upon sale" into their long-term operations, creating an unprecedented template. Government, enterprises, and finance: A triangular table of interests. For Washington, this money provides immediate relief for the continuously expanding fiscal deficit, and under the premise of not completely cutting off supply, it maintains a technical throttle on China's AI industry. The U.S. Department of Commerce began issuing new licenses last week, indicating the White House's acceptance of the compromise of "money for market." Nvidia CEO Jen-Hsun Huang has warned that a complete blockade would only accelerate China's self-research efforts; now he has temporarily retained about 13% of revenue from China, while AMD avoids losing up to 24% of its Chinese business. Morgan Stanley estimates that after the lifting of restrictions, AMD's related revenue could rebound to between $3 billion and $5 billion by 2025, while Bernstein expects Nvidia's H20 sales in China could reach $23 billion next year. Although profits are reduced by 15%, "partial profits" are better than "zero sales," and the two companies have chosen to minimize damages. However, there are concerns within Washington. Jacob Feldgoise, a researcher at the Center for Security and Emerging Technology in D.C., warned: "This expedient may undermine the national security basis of U.S. export controls and could lead to a decline in allies' confidence in U.S. policy." In other words, packaging national security with cash flow could shake the alliance built on security logic. The Chinese market and local competitors: Passive protection or accelerated catch-up? From Beijing's perspective, this "tax" not only increases procurement costs but also highlights the risks of external dependence. A CCTV-related account, "Yuyuantan Tian," questioned the safety and performance issues of H20, interpreted as pressure in negotiations. In reality, local alternatives have rapidly emerged: the Huawei Ascend series is estimated to currently account for 20% to 30% of domestic AI demand. When Nvidia and AMD are forced to hand over 15% of their revenue, it effectively creates a "passive protective umbrella" for Chinese manufacturers, narrowing the price gap and encouraging the use of domestic chips. In the long term, whether U.S. companies will continue to invest in customized "downgraded versions" or simply shift their R&D focus to other markets will be a key observation point. A new template for the tech cold war: Export controls linked to fiscal revenue. The agreement between Nvidia and AMD ties export restrictions, geopolitical issues, and direct fiscal revenue together in the same contract, shaping a new normal of "incomplete depeg." If this model is replicated in sensitive areas such as quantum computing and biotechnology, the global supply chain will face more diversion and repositioning. For companies, R&D routes are pulled by policy dialectics; for governments, how to switch weights between security, industry, and finance will test governance wisdom. In the short term, Washington gains cash, Nvidia and AMD maintain their passage, and Beijing gains time to cultivate self-sufficiency; it appears to be a three-way win, but due to trust gaps and technical competition, long-term variables remain numerous. The "chip tax" is just the beginning. At the crossroads of intense global power and capital movement, the next chip may soon fall into someone else's hands.