Nvidia gets nod from Washington to resume sales of H20 China chip

Digest opened free editor

Rola Khaleda, FT editor, chooses her favorite stories in this weekly newsletter.

NVIDIA said the Trump administration has resisted restrictions on exporting a major artificial intelligence product specifically designed for the Chinese market, saying it hoped to resume the H20 chip “soon.”

In a batch in favor of the leading artificial intelligence maker, NVIDIA said on Tuesday that it had received guarantees from the United States government that it would approve licenses for H20. It also announced the new Amnesty International chip of China, which he said was “completely compatible with export rules.

NVIDIA has reported a $ 4.5 billion fees in April quarterly for the excess stock after Washington prevented H20 sales to China in that month. The chip is designed to comply with a previous version of the export control elements on the semiconductors.

The assurances come after the intense pressure by the American chip maker, warning that America has risked the loss of its leadership in the artificial intelligence of Chinese companies, including Huawei, if it cuts critical technology exports.

“It is really important for American companies to be able to compete and serve the market here in China,” Ginsen Huang, the chief company in Nafidia, told reporters on a visit to Beijing on Tuesday.

Huang in the Chinese capital to meet government officials and clients to update them on how his company plans to handle the local market. He is looking for talks with Prime Minister Lee Qiang, who will be the Supreme Chinese official who has met so far, as the CEO is a growing role for the diplomat in the US -Chinese technology competition.

NVIDIA has also announced plans to issue a new graphics processing unit (GPU) for the Chinese market compatible with US export controls. Last week, FT NVIDIA mentioned an updated chip release designed for China, based on the Blackwell RTX Pro 6000.

These chips are especially useful for industrial customers who need to provide simulation pictures. They have also become a common option for developers to run small AL models, given the difficulties of buying the most advanced company chips designed for heavy business operations.

The new GPU of RTX Pro lacks the most advanced technology in NVIDIA, such as high -frequency memory and NVLINK, which improves the connection to transfer the fastest data, FT said earlier.

Huang said the new slide is “ideal for Twin Digital Twin for smart factories and logistics.”

The H20 slice is less powerful than graphics processing units than the best, but it is still witnessing a strong demand in China, especially after the launch of the Deepsek R1 model in January caused an increase in the use of artificial intelligence in the country.

The Chinese artificial intelligence players, including Bytedance, alibaba and TENCENT, were senior buyers at H20 earlier this year before the ban was entered into force.

Chinese customers were busy calling NVIDIA on Tuesday to understand how to apply for the license, according to a person who is aware of the discussions. Despite Washington’s assurances, there is still great concern about uncertainty, including the duration of H20 orders.

Meanwhile, there is pressure from Beijing to increase the purchases of local artificial intelligence chips from NVIDIA, including Huawei and smaller companies like Cambricon and Biren.

But NVIDIA is still the preferred option for the burdens of artificial intelligence, since the strong software ecosystem makes it easy to train and operate models.

The White House and the US Department of Commerce did not immediately respond to suspension requests.