top of page
Nvidia headquarters in Santa Clara, California, photographed on 4 August 2018.

Nvidia Stops Licensed AI-Chip Shipments to China

2 hours ago
3 min read

File photograph · 4 August 2018, Nvidia headquarters in Santa Clara, California. Coolcaesar / Wikimedia Commons, CC BY-SA 4.0. Cropped for the web hero and adapted for the social card.

Nvidia said on November 17 that it had stopped shipping licensed Hopper data-center chips to China and excluded China data-center compute sales from its next-quarter forecast. The disclosure stalls an export channel negotiated through Washington’s controls and Beijing’s import restrictions after it generated $2.4 billion in Nvidia’s latest quarter.

The company’s filing with the U.S. Securities and Exchange Commission recorded $2.4 billion in China shipments during the quarter ending October 25. Nvidia reported $115.8 billion in total revenue for the same period, making those licensed China sales about 2.1% of quarterly revenue. Its official outlook said it was “not assuming any Data Center compute revenue from China” in the following quarter.

Chief financial officer Colette Kress went further during the November 17 earnings call. Reuters reported that she said Nvidia was “no longer shipping” licensed Hopper data-center products to China. The company did not say why shipments stopped or whether the halt would be permanent, so the announcement does not establish that Nvidia has abandoned the Chinese market.

The channel had already been narrow. In January, the U.S. Commerce Department shifted H200 and comparable chips from a presumption of denial to case-by-case licensing. Exporters had to show that China-bound production would not reduce supply for U.S. customers, screen buyers, and submit chips for independent testing in the United States. Nvidia’s subsequent filing said the route also subjected qualifying chips to a 25% U.S. tariff after their import for inspection.

Beijing’s approval remained a second gate. Nvidia’s August quarterly filing said U.S. authorities had licensed small H200 shipments to selected customers, but Chinese restrictions prevented the company from selling everything covered by those licenses. Nvidia recorded a $400 million charge tied to H200 inventory and purchase obligations, and said that the approved shipments were only a fraction of what the program allowed.

That two-government structure matters more than the quarterly sales figure alone. Washington tried to preserve leverage by licensing a controlled slice of advanced computing power rather than opening the market. Beijing, meanwhile, restricted large-scale imports while building domestic semiconductor capacity and reducing dependence on U.S. technology. Nvidia could obtain a U.S. license and still lack a functioning route to Chinese customers.

The latest halt therefore exposes a limit of export-control bargaining: a permission negotiated by one state does not create a market when the other state withholds access. It also gives Chinese chipmakers more room to expand their software, developer and customer ecosystems, even where their hardware still trails Nvidia’s most advanced products.

The immediate financial effect is bounded but real. The $2.4 billion in licensed sales was a small share of Nvidia’s record quarter, but it was still substantial in absolute terms. Removing it from the next forecast shows that Nvidia is treating the channel as unavailable for now, not merely delayed by routine licensing paperwork.

What remains unresolved is the cause. Nvidia did not attribute the halt to a new U.S. order, a formal Chinese ban, customer demand, compliance concerns or inventory logistics. China’s Foreign Ministry reiterated its previously stated position on U.S. chip exports without announcing a new measure. Until either government or the company supplies more evidence, the defensible conclusion is narrower: shipments have stopped, next-quarter China data-center compute revenue is forecast at zero, and the duration is unknown.

Pamphlets previously reported that Washington and Beijing opened a joint trade-board process covering tariffs and agricultural goods. The Nvidia decision shows how much harder technology remains: even a channel approved after political negotiations can collapse when industrial policy, security controls and competing national development strategies all govern the same shipment.

Comments


bottom of page