Nvidia CEO Jensen Huang speaks to journalists as he arrives for a press conference at a hotel in Beijing on July 16, 2025.
Adek Berry | Afp | Getty Images
BEIJING — Nvidia CEO Jensen Huang was all smiles and compliments as he made his third trip to China in just about half a year.
As the leader and co-founder of the world’s first, newly-minted $4 trillion market cap company, Huang had particular reasons to be happy when he met the press on Wednesday: Nvidia expected it would be able to resume sales of its less advanced H20 artificial intelligence chips to China after a three-month pause.
“Many of my competitors are my friends,” he noted.
Huang said his understanding was that allowing Nvidia chips into China was part of an exchange with the U.S. for Beijing to release critically needed rare earths. CNBC has reached out to the White House for comment.
Wearing his iconic black leather jacket, Huang walked into the sunny courtyard of the Mandarin Oriental hotel about 15 minutes earlier than scheduled and took multiple questions in the nearly 90-degree Fahrenheit weather.
“Only in China can we do this out in the sun!” he said.
Then he realized the press conference was supposed to be held inside an air-conditioned room.
“What are we doing out here? Why didn’t somebody say so?” he said.
He was swarmed by local reporters asking for signatures of books and T-shirts. “Who needs an autograph? I’ll do it while I’m listening.”
Here are the highlights of what he said over 90 minutes:
Whom he met
Huang said he had a “wonderful meeting” with Chinese Vice Premier He Lifeng, and clarified that the discussions did not include China’s restrictions on battery technology or rare earths.
Earlier in the week, he met with Xiaomi founder and CEO Lei Jun, whom he labelled as “a brilliant business person.” He said the two discussed artificial intelligence for large language models, autonomous driving and robotics.
Xiaomi uses Nvidia’s automotive chips in its electric cars.
Huang said he told U.S. President Donald Trump about his planned voyage to China during a meeting with the White House leader last week to celebrate Nvidia’s $4 trillion market cap.
“[Trump] said, ‘Have a great trip,'” Huang said.
Export controls
Nvidia on Tuesday said it expected to resume its H20 chip shipments to China soon following assurances from the U.S. government. The company was forced to halt such sales in April due to new U.S. requirements at the time.
“In terms of the H20 ban and the lifting of the ban, it was completely in control of the U.S. government and China government. The discussion has nothing to do with me,” Huang said, rejecting the idea that he had played a part in changing Trump’s mind.
“It’s my job to inform the president about what I know very well, which is the technology industry, artificial Intelligence, the developments of AI around the world,” he said.
Huang emphasized Nvidia complies with the final policy decision and that tariffs are just something the company has to “adapt to.”
What’s next for Nvidia in China
U.S. chip restrictions nearly halved Nvidia’s market share in China, Huang said in May. Due to the U.S. export controls on China, the company said it missed out on $2.5 billion in sales during the April quarter and will likely take another $8 billion hit in the July quarter, pegging its sales at $45 billion over the period.
The U.S. effectively banned Nvidia from selling its most advanced chips to China back in 2022.
“I hope to get more advanced chips into China than the H20,” Huang said in response to a CNBC question, “and the reason for that is because technology is always moving on. It’s not like wood.”
He stressed that, years from now, there will be better and better technology available, adding, “I think it’s sensible that whatever we’re allowed to sell in China will continue to get better and better over time as well.”
But Huang would not give a definitive answer about how many orders Nvidia had received, or when the company would restart local sales of its chips — which he acknowledged were not the company’s best, but which could still train AI models.
He said the U.S. government was still processing the licenses for Nvidia to sell the chips to China, and that the company would need to restart its supply chain — a process he indicated could possibly take nine months.
Huawei
Huang also discussed the outlook for competing Chinese tech giant Huawei, which has been impacted by U.S. sanctions that precede the export controls on Nvidia.
“Anyone who discounts Huawei and anyone who discounts China’s manufacturing capability is deeply naïve,” Huang said, pointing also to how Huawei has “excellent chip design” and their own connected cloud system.
“They can go to market all by themselves.”
Underpinning Huawei’s AI model capabilities is an entire tech system that doesn’t rely on any of Nvidia’s chips or tools. Instead, Huawei has developed its own Ascend chips, which works with the company’s “CANN” system that acts as an alternative to Nvidia’s CUDA. It has also built an AI-specific cloud computing system called CloudMatrix that launched last year.
Asked about indications that Huawei’s AI chip systems are still challenging for many developers to switch over to, Huang said, “That’s just a matter of time.”
He said “the important thing to realize I’ve been doing this for 30 years, they’ve been doing it for a few, and so the fact they’re already on the dance floor tells you something about how formidable they are.”
China’s AI
Huang rained down praise on Chinese AI models, as he had during a speech Wednesday morning at the opening ceremony of the high-profile supply chain expo in Beijing.
“The Chinese models, DeepSeek, Qwen, Kimi, are excellent,” he said, referring to the breakthrough from a Chinese startup, Alibaba’s model and another one from an Alibaba-backed startup Moonshot.
“I think over time it will be increasingly less important which one of the models are the smartest,” he said. “It’s going to be which one of the models are the most useful.”
China-developed DeepSeek shocked global investors in January with the release of an AI model that undercut OpenAI on development and operating costs. It’s not clear how DeepSeek managed to develop the model under broad U.S. chip restrictions on China, but the startup’s parent, High-Flyer, reportedly stockpiled Nvidia chips.
One aspect that Huang said he particularly appreciated about Chinese AI models was that they are open source, making them available for people to download for free and use on their own computers.
He said many companies in many countries downloaded DeepSeek R1 — “99%” of people — to use it locally for healthcare, robotics, imaging and other applications.
As Huang was about to end the press conference, a reporter asked whether he would come back to China again this year.
“I hope so. You have to invite me.”