Nvidia China AI Market Share Hits Zero: Jensen Huang’s Quiet Confession That Shook Silicon Valley

In interviews, Jensen Huang typically doesn’t appear worn out. Everyone is a little more enthusiastic about transistors when he leaves the room wearing the leather jacket, talking quickly, and making jokes about the size of objects with his hands. However, there was a change in tone when he stated, “in China, we have now dropped to zero,” on the Special Competitive Studies Project podcast last week. Not exactly rage. It was more akin to the silence of a man watching a market he constructed disappear out the back door while Washington quarreled over who had left it open.

About 95% of China’s AI accelerator market was controlled by Nvidia two years ago. That figure represented the entire ballgame, not just a rounding error. Similar to how American highways run on asphalt, Nvidia silicon powered China’s largest AI labs, cloud computing companies, and research institutes. Then came the export restrictions, which have been tightening since 2022 and are stacked on top of one another. Even the H20, a chip Nvidia had created especially to remain inside Washington’s borders, required a license to ship by April 2025. In actuality, the license procedure killed the industry.

The speed at which the collapse occurred is remarkable. Nvidia’s China share could fall from 66% in 2024 to about 8%, according to a prediction made earlier this year by the research firm Bernstein. That was deemed aggressive by analysts. According to Huang’s admission, the slide completely defied that prediction. Although shares are trading close to $198, less than 10% below the all-time high, investors appear to think the stock can absorb the loss, but the underlying picture is more unsettling. In short, foreclosure from China enabled rivals to create more expansive ecosystems to compete with Nvidia globally, according to the 10-K filing.

It’s difficult to ignore who entered the vacant room. Moore Threads, MetaX, Huawei, and Cambricon. Almost every analyst note about Asia now includes names that three years ago meant very little to American investors. According to reports, Huawei alone is aiming for $12 billion in chip sales by 2026. The real American advantage is still the so-called CUDA moat, or Nvidia’s software ecosystem, which rivals can’t just copy. Their hardware isn’t yet up to par with Nvidia’s top tier. However, moats are only effective when the castle is still occupied. The moat turns into scenery when the army is removed.

The reporting contains an odd detail that keeps coming up. When Nvidia’s B300 servers make it to Chinese consumers via the remaining channels, they sell for about a million dollars apiece. nearly twice as much as in the United States. Due in part to a high-profile prosecution of a Supermicro co-founder earlier this year, the grey market that once kept prices closer to parity has shrunk. Therefore, Chinese companies that are paying twice as much for chips that they aren’t supposed to have are still able to obtain them, albeit in smaller quantities, more slowly, and without a warranty from the manufacturer. In the meantime, Nvidia receives no revenue.

CEOs seldom issue such a warning in front of legislators as Huang did. He claimed that caving in to a market the size of China “does not make a lot of strategic sense” and that the policy “has already largely backfired.” He cited China’s deep talent pool, inexpensive energy, and astounding number of AI researchers, which he described as one of their national treasures. The remarks had a different impact than they would have from, say, a Washington think tank because they came from someone whose business profits greatly from American export policy in other ways.

As this develops, there’s a sense that the US may have created one question while providing an answer to another. The initial concern was that Chinese military AI would be accelerated by sophisticated American chips. As a result, an American company is witnessing its global stack lose ground in the largest single market in the world, while the Chinese hardware industry is being forced into self-sufficiency faster than almost anyone anticipated. It’s still unclear if that trade was worthwhile. It appears that the B300s in Shenzhen are not waiting for a response.

On May 20, Nvidia releases its earnings. By any standard corporate measure, the management’s goal of 77% revenue growth is ridiculous and implies that the rest of the world is more than just picking up the slack. However, the China column on all subsequent spreadsheets now shows zero, and zero has a tendency to remain zero for longer than anyone anticipated.