Washington tightens the chip controls again

The Bureau of Industry and Security issued a second package of export-control rules, a year after the October 2022 controls had first restricted advanced chip sales to China. BIS said the update was necessary to “maintain the effectiveness of these controls, close loopholes, and ensure they remain durable” — a direct response to how the industry had adapted to the original rules.

The 2022 controls had set performance thresholds meant to catch the most capable AI training chips while leaving room for lower-end parts. Nvidia responded by designing the A800 and H800, chips tuned to sit just under those thresholds while retaining most of the training performance of the restricted A100 and H100. The October 2023 rules closed that gap directly: they dropped interconnect bandwidth as a control parameter — the specification the workaround chips had been designed around — and added a new “performance density” threshold intended to prevent chipmakers from combining many smaller, uncontrolled chips to match a restricted part’s aggregate capability.

The rules also broadened the controls’ geographic reach. Licensing requirements, previously focused on China, were extended to cover shipments to companies headquartered in any of the 21 additional countries under US arms embargoes, closing a route by which restricted chips could reach China via third countries. BIS added foundry due-diligence requirements to identify restricted chip designs before manufacture, and named 13 additional Chinese entities to its Entity List. The rules took effect a month later, on 17 November 2023.

The update confirmed the pattern the 2022 rules had begun: controls, corporate engineering around them, and a further round of controls. Nvidia again produced China-specific parts calibrated to the new thresholds, and Washington tightened the rules again in successive rounds through 2024 and 2025, including a third major round in December 2024.