Costco's Kirkland brand became a massive financial success. Here's how
Subtitles
  • Off
  • English

Here are the chips that Nvidia can sell to China

The company made specially-designed chips for the Chinese market to comply with U.S. export controls

We may earn a commission from links on this page.
Start Slideshow
Start Slideshow
Jensen Huang sitting in a chair, speaking, with both of us hands up, in front of a purple backdrop
Nvidia CEO Jensen Huang at the Bipartisan Policy Center on September 27, 2024 in Washington, D.C.
Photo: Chip Somodevilla (Getty Images)

After the U.S. placed chip-export controls on China in October 2022 to curb the country’s technological advances, U.S.-based chipmaker Nvidia (NVDA) designed less-powerful versions of its artificial intelligence chips to continue its presence in the Chinese market while complying with the rules.

Advertisement

Nvidia’s most advanced A100 and H100 chips were blocked under the restrictions, leading it to develop the reduced-capability A800 and H800 chips. Chinese AI startup DeepSeek used the H800s to train and develop its AI models that rattled Silicon Valley and Wall Street earlier this year.

However, under updated chip-export controls introduced in October 2023, those graphics processing units, or GPUs, were also banned, prompting the company to develop three more compliant chips.

Now, concerns are mounting over whether Nvidia will be able to sell these specially-designed chips for much longer, as trade tensions between the U.S. and China rise.

China and Hong Kong are the chipmaker’s fourth-largest market, representing about 13% of its annual revenue — or $17.1 billion — for fiscal year 2025, according to Nvidia’s Form 10-K.

Earlier this week, the Trump administration announced new U.S. trade restrictions against dozens of Chinese tech firms, citing national security and foreign policy concerns — a continuation of the years long effort to maintain the U.S. lead in chips and AI.

Here are three chips Nvidia can sell to China.

Advertisement
Previous Slide
Next Slide

The H20

Jensen Huang standing in front of a black backdrop that has a drawing on it and the words Generative AI
Nvidia CEO Jensen Huang during his keynote address at the Nvidia GTC on March 18, 2024 in San Jose, California.
Photo: Justin Sullivan (Getty Images)

To comply with the updated chip-export controls, Nvidia developed the H20, which is a modified version of its more powerful H100 GPUs that are used by major firms such as Google (GOOGL) and Meta (META). The H20 has become its most prominent chip in the Chinese market.

Advertisement

The H20 has a reduced core count making it lower performing but still able to deliver AI capabilities.

This week, H3C — one of China’s largest server makers — warned that the country could face potential shortages of the H20, as demand has grown significantly since DeepSeek’s boom.

Advertisement
Previous Slide
Next Slide

The L20

Jensen Huang holding up a chip during his keynote speech
Nvidia CEO Jensen Huang during his keynote address at the Nvidia GTC on March 18, 2024 in San Jose, California.
Photo: Justin Sullivan (Getty Images)

The L20 is based on Nvidia’s Ada Lovelace architecture, and was built on the 5 nanometer process. The L20 is a modified version of the AD102 — Nvidia’s flagship GPU. The AD102 can support gaming and AI.

Advertisement
Previous Slide
Next Slide

The L2

jensen huang standing at the bottom of the photo in front of a big black backdrop and the nvidia logo above him
Nvidia CEO Jensen Huang during his keynote address at the Nvidia GTC on March 18, 2024 in San Jose, California.
Photo: Justin Sullivan (Getty Images)

The L2 is based on the Ada Lovelace architecture, and is a modified version of the AD104 GPU.

Advertisement