The Council on Foreign Relations has published an assessment arguing that Huawei cannot catch Nvidia in AI accelerators, and that U.S. export controls on advanced chips should therefore remain in place. The analysis frames China's AI chip shortage as structural rather than temporary: the deficit stems from constrained access to leading-edge fabrication, not from design capability alone.

At the center of the argument is Huawei's Ascend accelerator family, the flagship of China's domestic AI compute effort. Huawei has positioned Ascend as the local alternative to Nvidia's data center GPUs, which Washington has restricted from export to China. The CFR piece concludes that this substitution effort is failing to close the gap in both performance and volume.

The core bottleneck, according to the analysis, is manufacturing. Huawei's chip design arm HiSilicon relies on domestic foundry capacity — effectively SMIC — that cannot match the leading-edge process nodes available to Nvidia through TSMC. Nvidia's current accelerators are fabricated on TSMC's most advanced nodes and packaged with cutting-edge memory and interconnect supply chains. Huawei's Ascent line, by contrast, must be produced on trailing processes using workaround engineering, and U.S. export controls also restrict China's access to advanced chipmaking equipment, high-bandwidth memory, and the tooling needed to scale domestic fabs.

That gap compounds at the system level. Training large AI models requires not just individual accelerators but large clusters of tightly networked chips. Even where Chinese firms can produce individual silicon, the CFR assessment suggests they struggle to deliver the volume and consistency of supply that hyperscale AI training demands — a deficit that cannot be fixed quickly with money alone, since equipment and process constraints bind over multi-year cycles.

The second half of the argument is prescriptive. The CFR analysis states that U.S. export controls should remain, on the logic that controls are working: they are slowing China's ability to build frontier AI compute and preserving a decisive lead for U.S. ecosystem players led by Nvidia. The piece positions continued restrictions as a lever that raises the cost and timeline of China's AI ambitions rather than a policy that has been rendered moot by indigenous progress.

The argument cuts against a competing view in the policy debate — that export controls have already failed because Huawei and other Chinese designers have shipped alternatives regardless. The CFR piece instead reads those domestic efforts as evidence of forced substitution at lower capability, with Huawei unable to match Nvidia's performance or output rather than simply replacing it.

For the semiconductor industry, the stakes are straightforward. Nvidia's dominant share of AI accelerators rests on access to TSMC's leading edge, advanced memory suppliers, and a toolchain that U.S. policy actively shields from Chinese replication. If the CFR's read is right, that position holds as long as equipment and component controls stay in force. If Huawei's Ascend roadmap or SMIC's process advances outpace the analysis, the commercial calculus for hyperscalers and Chinese cloud providers shifts accordingly — a question the next round of Ascend shipments and SMIC capacity figures will help answer.