-2.jpg "Brooke Becher")
Just three months after halting AI chip exports to China, the Trump administration has quietly reversed its policy, assuring Nvidia that it can resume shipments of its H20 processors, which are specially designed to meet U.S. export limits for China. Nvidia confirmed the move in a blog post, noting it can now restart shipments under the guidance of the Commerce Department.
Rival AMD also won clearance to restart MI308 sales in China, prompting both companies’ shares to climb over four percent in early trading. The flip decision comes after intense lobbying by Nvidia CEO Jensen Huang, including recent meetings at the White House and in Beijing.
This means China — home to roughly half of the world’s AI development companies — now has access to vital compute capacity it didn’t have before. Under the revised export rules, so-called “green‑zone” chips like the H20 can flow freely, while more advanced GPUs remain restricted. The move restores billions in potential revenue for American chip makers and reignites the race for AI supremacy between the United States and China. It’s a delicate balancing act between safeguarding U.S. national security and maintaining global competitiveness in the semiconductor arena. And with Nvidia now cleared to pursue approvals for its next‑generation Blackwell‑based GPUs, that competition is only set to intensify.
Why Did the U.S. Restrict Chip Sales to China in the First Place?
This all started in October 2022 under President Joe Biden, when the Commerce Department rolled out export controls aimed at protecting U.S. national security. The rules cut off China’s access to high‑end AI chips, such as Nvidia’s 2020‑era A100, with the intent to prevent the Chinese military and surveillance agencies from accessing cutting-edge processing power.
A year later, those rules were broadened to cover any processors built on the same architectures, forcing U.S. vendors to either apply for special licenses or deliberately throttle performance before shipping. Noting dual priorities, then-Commerce Secretary Gina Raimondo made clear that American firms “can, will and should sell AI chips” like the H20, while keeping top‑tier products reserved for the U.S. and its allies.
This materialized in the “AI Diffusion Rule.” Introduced in January 2025, the set of regulations established global performance thresholds that blocked sales of flagship GPUs like the H100 and H200 to China, while carving out a “green zone” for lower‑powered models. Nvidia engineered its H20 to fall squarely within that green zone — avoiding extra export paperwork — and, in response, Chinese companies began optimizing midrange offerings such as the H800 and ramping up their own semiconductor research, fueling Beijing’s push toward domestic self‑sufficiency in AI hardware by 2027.
Then in April 2025, the Trump administration slammed the door again, banning even chips deemed “compliant” as it ramped up tariffs and tech restrictions on Beijing. Three months later, it backtracked and said licenses for the H20 (and AMD’s MI308) would be approved. The U-turn underscores how export controls now move in lockstep with tariff brinkmanship — both tools in a the ongoing AI power struggle.
What Chips Can Nvidia Sell to China Now?
With the ban lifted, Nvidia can now resume China-bound shipping of its H20 inference GPU, which features 96 GB of HBM3 memory and a 4.0 TB per second memory bandwidth. Previously, this chip computation power has been deliberately capped to fit within U.S. export‑control limits.
Nvidia can also offer its new RTX PRO series, launched in March. It’s built on the Blackwell architecture, which boosts FP32 compute, expands memory capacity and improves interconnect speeds for enterprise AI workloads.
Reopening China sales is a major commercial win for Nvidia, as China represents roughly 13 percent of its revenue and hosts about half of all global AI developers.
What Does This Mean for China’s AI Industry?
China’s AI sector lags behind the United States by roughly three to six months, according to White House AI Czar David Sacks, a gap that only widened when export controls choked off access to advanced GPUs. Still, homegrown companies like DeepSeek, Baidu and Alibaba have rolled out surprisingly competitive models of their own by squeezing performance out of midrange chips through clever, open-source software techniques.
“Chinese labs are stretching the hardware they already have,” Jay Dawani, CEO of AI software provider Lemurian Labs, told Built In. He points to DeepSeek’s R1 model as a prime example: Unable to access to Nvidia’s H100-class GPUs, the company redesigned its model architecture and optimized training efficiency to achieve “near-frontier” performance with millions of less-capable H800 chips — well below the U.S. export-control threshold. DeepSeek’s approach even drew scrutiny from U.S. authorities, who are investigating allegations that the company smuggled high-end semiconductors through Southeast Asian shell companies in order to train its flagship model.
“DeepSeek shows, Chinese researchers are learning to get more out of less,” Dawani said. “Once those techniques mature, they travel quickly across the ecosystem and reduce the strategic value of raw compute supremacy.”
All told, China’s AI industry remains “deeply dependent” on American GPUs, especially those made by Nvidia, according to Dawani. He estimates that around 75 percent of the chips powering AI model training in Chinese data centers still run on Nvidia’s CUDA platform, and that the company has shipped more than a million export-compliant H20 chips to China since late 2024 — outpacing projected shipments of Huawei’s new Ascend chip by a five-to-one margin. In short: Chinese companies can “keep the lights on” with domestic chips, Dawani continued, but they can only sustain “frontier-level” progress with continued access to Nvidia hardware.
Now, with green‑zone GPUs available once again, AI development in China will likely go into overdrive, as companies move beyond stopgap workarounds and resume full‑scale model training and inference. On his recent trip to Beijing, Nvidia’s CEO Jensen Huang was lauded by Chinese firms like DeepSeek, Alibaba, Tencent and Huawei for helping to get the H20 ban lifted, thus restoring their AI research pipelines.
To keep up with surging demand, Huang pledged to ramp up H20 shipments in the coming months, essentially making up for lost time. Meanwhile, major players like Baidu and Tencent are still doubling down on domestic chip alternatives, diversifying their hardware portfolios to fortify self‑reliance and hedge against any future export restrictions.
What Does This Mean for America’s AI Industry?
The decision to lift the chip export ban is a clear financial win for U.S. chipmakers like Nvidia and AMD. But giving China easier access to critical hardware could also cost America its competitive edge in the AI industry.
Regardless, Sacks called the move “pragmatic,” arguing that while “we’re not selling our latest greatest chips to China, we can deprive Huawei of having this giant market share.” Indeed, Commerce Secretary Howard Lutnick told CNBC that China is only getting Nvidia’s “fourth best chip,” and that it is not as good as the chips American companies are using. “The Chinese are more than capable of building their own,” he said. “You want to keep one step ahead of what they can build, so they keep buying our chips.”
Behind the scenes, Lutnick reportedly wove the chip‑export reversal into broader negotiations involving access to rare‑earth metals, easing restrictions in return for China’s commitment to resume shipments of minerals essential to U.S. manufacturing.
For Nvidia, CEO Jensen Huang dubbed the decision a “turning point,” pledging to “accelerate the recovery” of $4.5 billion in unsold inventory and rebuild ties with China-based customers. AMD chimed in with its own good news that MI308 license applications are progressing, restoring competition in a market that had gone cold.
Not everyone shares the optimism, though. Critics like Rep. John Moolenaar warn that reopening access to Nvidia chips could “strengthen China’s military capabilities, suppress citizens and threaten U.S. innovation.” Even so, many executives see the policy shift as a way to free up capital for next‑generation architectures, with America still at the top.
“America remains ahead, but its margin of safety now depends less on simple chip counts and more on sustaining leadership in software tooling, foundational research and next-generation silicon,” Dawani said. If the U.S. sticks to “coherent” export rules — restricting only truly frontier GPUs while allowing mid-tier parts — the country will likely maintain its advantage, he said. But if policies falter, or China’s efficiency gains outpace U.S. innovation, America’s lead could shrink fast.
Ultimately, the real test will be whether U.S. firms can leverage renewed China sales to fund breakthroughs beyond today’s green‑zone GPUs. In that sense, the chip ban reversal isn’t just a business decision — it’s a strategic chess move that bets on America’s ability to sustain long‑term leadership in the global AI race.
Frequently Asked Questions
Why can’t Nvidia sell chips to China?
The United States government prevents Nvidia from selling some of its more advanced AI chips in order to prevent the Chinese military and surveillance agencies from accessing cutting-edge processing power. These restrictions were initially imposed in 2022 by President Joe Biden and were then expanded by President Donald Trump in 2025. Trump has since reversed some of the restrictions.
What chips can the U.S. sell to China now?
As of July 2025, Nvidia can sell its H20 inference chip, as well as its RTX PRO series, to China. AMD can also sell its MI308 chip.
