How New US AI Chip Export Ban Could Reshape China's AI Landscape

US officials appear alarmed and have begun banning the export of advanced artificial‑intelligence chips to China.

The latest round of restrictions seeks to virtually cut off China’s market for high‑end GPUs and accelerators, extending beyond data‑center chips to domestic industries.

These rules could affect most of Nvidia’s GPU families, including the H800 and A800 kits, a setback for Chinese internet giants that planned to spend $4 billion on such chips in 2024. Intel and AMD, which also sell chips in China, face similar challenges.

The new regulations close loopholes left by last year’s ban on Nvidia’s H100, which forced China to use slower versions (H800/A800) that reduce interconnect speed.

US officials told the media that the measures also prohibit the use of these chips and could impact Intel, AMD, and companies that export semiconductor‑manufacturing equipment. The stated goal is to prevent China from obtaining advanced semiconductors that could accelerate AI breakthroughs, especially for military applications.

According to Commerce Secretary Gina Raimondo, the policy is designed to control access to computing power, which should significantly slow China’s development of next‑generation AI models, particularly those with modern warfare uses.

US officials claim the restrictions are not intended to harm China’s economic growth.

Nvidia had already created the trimmed H800 and A800 versions to comply with earlier trade rules. With the high‑end H100 already blocked, the new policy further hampers China’s AI market.

Nvidia expects only a modest short‑term financial impact, stating it complies with all applicable regulations and continues to support thousands of applications across industries. The company noted that the restrictions cover models A100, A800, H100, H800, L40, L40S, RTX 4090 and related systems such as DGX and HGX. After the announcement, Nvidia’s shares fell about 6% in early New York trading, while AMD and Intel each dropped roughly 3%.

Intel and AMD are also likely to be affected. AMD’s top‑tier MI250X was already subject to last year’s export limits, and its MI210 falls below the 600 GB/s bandwidth threshold. AMD estimates the MI210’s TPP score at 5,792 with a power density of 8, suggesting it will not be sold in China once the rules take effect later this fall.

AMD has announced a special accelerator similar to Nvidia’s A800/H800 for the Chinese market, and Intel’s China‑specific Gaudi2 HL225B claims performance surpassing Nvidia’s A100 on certain AI workloads. Intel says it is reviewing the regulations and assessing potential impacts.

Consumer‑grade GPUs are largely exempt, as the rules explicitly target chips designed for data‑center use. However, many consumer cards share the same silicon as data‑center variants.

An exception in the BIS document applies to cards with a TPP (tiles per pipe) of 4,800 or higher. This is why Nvidia indicated its RTX 4090 may no longer be sold in China; its estimated TPP is about 5,285, making it the only consumer‑class GPU currently subject to export control.

By contrast, AMD’s flagship consumer GPU, the RX 7900 XTX, scores a TPP of 3,904, below the threshold.

The rule contains a potential loophole, as Chinese entities have reportedly used shell companies to acquire modified Nvidia GPUs and Intel processors for nuclear‑weapon simulation and other dual‑use applications. The new regulations do include provisions to make indirect imports more difficult.

As vendors launch more powerful cards, the number of GPUs subject to the restrictions could rise. For example, AMD’s 7900 XTX offers roughly 2.5× the FP32 performance of its predecessor, suggesting future high‑end desktop GPUs may soon breach the TPP threshold unless the US adjusts the limits.

Industry analysts at TrendForce predict the restrictions could shrink China’s demand for Nvidia’s high‑end AI servers from 5‑6% of global demand to 3‑4%. Major Chinese cloud providers such as ByteDance, Baidu, Alibaba Cloud, and Tencent are expected to begin stockpiling GPUs before the rules take effect. Nvidia may also try to allocate its scarce H800 inventory to Chinese customers.

In the longer term, TrendForce expects Chinese firms to accelerate domestic chip development, citing Alibaba’s Pingtouge ASIC efforts and Huawei’s Ascend platform. Analysts also note that Chinese companies might shift AI development to rented resources abroad.

While export restrictions make it harder for China to obtain US‑origin AI chips, they do not address the fact that AI accelerators are widely deployed in public clouds, which remain accessible worldwide. The current US policy does not resolve this cloud‑access issue.