Nvidia is preparing to resume selling Groq chips in China, per sources cited by Reuters, after a two-year pause imposed by Trump-era restrictions. CEO Jensen Huang disclosed “purchase orders” received and manufacturing restarted, with U.S. and Chinese regulatory clearance secured. The deal includes a 25% sales cut to the U.S. Treasury and third-party hardware testing — a bureaucratic tollbooth for Washington to profit from a client list it once tried to ban.
This decision fits a pattern of U.S. export policy behaving more like a toll road than a firewall. Since 2021, Washington has alternated between banning Chinese access to advanced semiconductors, then grandfathering exceptions with profit-sharing clauses. Nvidia’s earlier H200 sales required a “license and a 25% tax-like payment” from China, per the company’s February 2026 earnings call. The result isn’t a defense of national security but a revenue model for governments on both sides of a trade war.
Sources disagree on the scale. Reuters notes “Nvidia is preparing Groq chips,” while CNBC’s Huang quotes claim only a “small number of H200 products” have been approved and no revenue recognized yet. Bloomberg’s focus on OpenClaw — an AI project touted as the “next ChatGPT” — reveals how Nvidia’s comments ripple through markets, boosting Chinese equities despite unclear ties to its chip sales strategy.
The deeper story is about the erosion of technological advantage through bureaucratic self-sabotage. By letting China buy chips only after extracting concessions, the U.S. government guarantees both sides will innovate around the barriers. The H20 chip, a weakened version of H200, was already deployed by Chinese firms; now, OpenClaw’s AI ascent may owe as much to stolen code as to American silicon. Meanwhile, Nvidia’s $5.5 billion charge from prior bans shows how export controls backfire on their enablers.
Absent from this coverage are the end users: Chinese data centers and research labs that will now accelerate AI development using subsidized U.S. hardware. Nor does any source ask how many of these chips will end up in tools threatening U.S. military assets. The bigger blind spot is the Chinese state’s own hypocrisy — while Xi purges generals for corruption (per Al Jazeera), it funnels hundreds of billions in subsidies toward its semiconductors sector.
If regulators maintain the status quo, investors and strategists should watch March 2027, when Nvidia’s Blackwell chips launch. Will Washington let China purchase them without a 25% stake in its data centers? Or will the U.S. pivot to banning AI model training outside American soil, as some policymakers urge? The next phase of this saga will decide whether America’s tech edge is a fortress or a tollbooth.