Nvidia announced Monday that it’s filing applications to restart sales of its H20 artificial intelligence chips to China, capping a spasmodic few months that saw the Trump administration impose restrictions, then quickly reverse course after a high-profile dinner meeting.
The company expects to receive U.S. government licenses soon and begin deliveries shortly after, according to a blog post. Nvidia is also introducing a new “RTX Pro” chip designed specifically for the Chinese market, calling it “fully compliant” with regulations and ideal for digital manufacturing applications like smart factories and logistics.
The H20 chip sits at the center of a broader U.S.-China tech standoff. While not Nvidia’s most advanced AI processor, the H20 is the most powerful chip the company can legally sell to China under existing export controls. It’s specifically designed for “inference” tasks — running existing AI models for day-to-day applications — rather than training new AI systems from scratch.
Chinese tech giants including ByteDance, Alibaba, and Tencent had been aggressively stockpiling these chips in the first three months of this year in anticipation of stricter export controls. The chip’s appeal lies partly in its superior memory bandwidth compared to Chinese alternatives, along with Nvidia’s widely-adopted software ecosystem that makes the hardware easier to deploy.
The regulatory back-and-forth began in April when the Trump administration restricted H20 sales, potentially costing Nvidia $15 billion to $16 billion in revenue, judging by how much Chinese firms reportedly splashed out for them in the first quarter alone. The move targeted chips exceeding specific performance thresholds, including total memory bandwidth of 1,400 gigabytes per second or input/output bandwidth of 1,100 GB per second.
But the restrictions were fairly short-lived. Soon after CEO Jensen Huang attended a $1 million-per-head dinner at Trump’s Mar-a-Lago resort in early April, the administration paused the ban. According to NPR, the White House changed its mind after Nvidia promised new U.S. data center investments. Within a week of NPR’s report being published, Nvidia announced plans to build AI servers in the U.S. worth as much as $500 billion over the next four years, with help from partners such as TSMC.
The flip-flopping has drawn criticism from U.S. lawmakers who argue it undermines the country’s efforts to limit China’s AI capabilities and who point to DeepSeek to underscore why it matters. The Chinese startup took the AI world by storm earlier this year by building an impressive model using Nvidia’s H800 chips, which are slightly more powerful predecessors to the H20. (The U.S. banned the sale of those H800 chips back in October 2023, but Chinese suppliers have managed to figure out workarounds.)
In a statement sent to TechCrunch, Nvidia spokesman Hector Marinez said Huang has been meeting with officials in Washington and Beijing this month and “emphasizing the benefits that AI will bring to business and society worldwide.”
In the meantime, the whole episode underscores the ongoing balancing act that U.S. policymakers are attempting, with concerns about national security running up against powerful commercial interests. Given what we’ve already seen in 2025, we can probably expect more reversals of its kind, too.
