US imposes export controls on advanced computing chips to China

Background

Before the October 2022 rules, the United States had already acted on specific companies. On 26 August 2022 Nvidia disclosed that the government had imposed a licence requirement on exports of its A100 chip and its forthcoming H100 to China, including Hong Kong, and to Russia, and said about $400 million of potential sales to China in that quarter could be affected^[2]. The Bureau of Industry and Security (BIS), the part of the US Commerce Department that administers export controls, later noted that it had already placed company-specific restrictions on advanced chips for artificial intelligence^[1].

What the two rules did

On 7 October 2022 BIS announced two rules restricting China's ability to obtain advanced computing chips, to develop and maintain supercomputers, and to manufacture advanced semiconductors^[1]. BIS said those capabilities were used by the PRC for advanced military systems, faster military decision-making and autonomous military systems, and for human-rights abuses. Assistant Secretary Thea Rozman Kendler noted that China sought to become a world leader in AI by 2030^[1].

Export controls work through licences: an item placed on the control list cannot be shipped without government permission, which may be granted or refused. The October rules added certain advanced computing chips, and computers containing them, to the Commerce Control List, required licences for items going to supercomputer or chip-making end uses in China, and extended US jurisdiction over some foreign-made advanced computing items^[1].

The manufacturing thresholds

A second front was chip production itself. The rules restricted equipment for Chinese fabs — the factories that pattern circuits onto silicon wafers — making advanced logic chips, defined as 16/14 nm FinFET or below, DRAM of 18 nm half-pitch or less, and NAND flash with 128 or more layers^[1]. They also limited US persons from supporting such fabs without a licence^[1].

Phased entry into force

The rules took effect in phases: chip-manufacturing items on 7 October, the US-persons restrictions on 12 October, and the advanced computing and supercomputer controls on 21 October 2022^[1].

How chipmakers responded

Nvidia responded with China versions of its chips that stayed under the thresholds: the A800, and then the H800, reported in March 2023, whose chip-to-chip data rate was reportedly cut to about half of the H100's^[3].

The tightened rule of 2023

A tightened rule announced on 17 October 2023 extended licensing to chips such as the A800, H800, L40, L40S and RTX 4090, and to more countries. Nvidia reported that it took effect immediately for its main data-center products from 23 October 2023^[3].

Why it mattered

The measures combined several kinds of control in one package: restrictions on particular products, restrictions tied to the end use of an item, limits on named manufacturing equipment, and limits on what people — not just goods — could do. Because the rules were written around numeric thresholds such as lithography node, memory half-pitch and layer count, chipmakers could compare their products against those lines and design variants that fell on the permitted side, which is what the A800 and H800 did^[3]. When the thresholds were revised the following year, those variants were pulled back under licence requirements^[3].