News

Nvidia Stopped Shipping Licensed Hopper Data Center Chips to China After a $2.4 Billion Quarter

Colette Kress told analysts licensed Hopper data-center products to China were $2.4 billion in the quarter ended October 25 and that Nvidia is no longer shipping them. Q4 guidance of $130 billion assumes no China data-center compute.

Nvidia just closed the China data-center chip business it spent a year getting Washington to allow. On the fiscal third-quarter call, interim chief executive Colette Kress told analysts that licensed Hopper data-center products, the family that includes the H200, brought in $2.4 billion from China in the quarter ended October 25, and that the company is no longer shipping them. The line is not in the printed earnings release. It is in her prepared remarks: guidance has no licensed Hopper data-center products in it.

The rest of the quarter was a record. Nvidia reported $115.8 billion of revenue, up 20 percent from the prior quarter and 103 percent from a year earlier, with data-center revenue of $107.6 billion. GAAP earnings were $2.70 a diluted share. For the current quarter, which includes an extra week, the company guided to $130.0 billion plus or minus 3 percent, and it said outright that it is not assuming any data-center compute revenue from China in that outlook.

Two point four billion dollars is about 2 percent of the data-center line. It is also the entire licensed China compute story, booked in one quarter and then taken out of the model.

A license Nvidia is no longer using

The H200 was the exception after export controls shut more advanced parts out of China, with Commerce reviewing licenses one buyer at a time. The commercial result showed up this quarter as $2.4 billion. Kress did not say why shipments stopped, or whether the halt is meant to be permanent.

That silence sits on top of a mess Nvidia already had in public. Earlier this month Techpresso covered the empty Beijing office of New Matrix, a firm Commerce had cleared to handle H200 servers, and the deserted Maginfra office tied to a company on the entity list. None of that is a stated reason for Kress's line. It is the backdrop investors had when she delivered it.

In a CNBC interview after the call, Kress said China is still about 10 percent of shipments, and that those remaining sales are gaming, automotive and workstation. The cut is the licensed Hopper data-center product, not the whole country P&L.

Demand that does not need China

Nvidia can drop that revenue because everything else is still short of chips. Data-center revenue rose 21 percent quarter on quarter. Vera Rubin, in its first quarter of production shipments, is the ramp Kress is selling to investors, and the fourth-quarter guide of about $130 billion assumes none of the China compute that just printed. Gross margin is expected to step down to 71 to 72 percent, which the company has been flagging as mix, not as a China hole.

The $150 billion extra buyback the board approved in September, and the $28.1 billion returned to shareholders in the quarter, are the other signal: this is a company funding itself from hyperscaler demand, not from a special license to Beijing.

What changed is the permission structure. For one quarter, licensed Hopper chips to China were a real number. They are now a number Nvidia is willing to print in the rear-view mirror and omit from the next one.

Some offers on this page may be paid placements or contain affiliate links.

Finpresso: daily AI & finance brief

Free daily newsletter, read in 5 minutes.

Subscribe free