AI Trends

US BIS Restricts Chinese Firms' Remote Access to Advanced Overseas Computing Power: How Are Chinese AI Companies Affected?

East MoneyUpdated Nov 18 10:23

AI summary

On November 16, the US Department of Commerce's Bureau of Industry and Security (BIS) upgraded AI chip controls, restricting Chinese companies from remotely accessing advanced computing power through overseas data centers: when a data center provides more than 12 million TPP of computing power to companies whose headquarters or ultimate parent company is located in China (including Macau) or other such places, the chip exporter must apply for a license, with case-by-case review; data centers that have already provided large-scale computing power services must obtain licenses retroactively within 90 days of the rules taking effect. Morgan Stanley estimates that 12 million TPP is roughly equivalent to 700 to 800 H100 chips, which may slow Chinese AI companies' overseas training and model iteration in the short term. Nomura Securities believes the restrictions may expand the advantage of major internet companies over independent model companies and accelerate the adoption of domestic AI chips, with Alibaba's self-developed T-Head chip seen as a buffer.

Why read this

SelectedAI score 76

The new rules bring overseas remote computing power under license review. The article provides TPP conversion figures and institutions' assessments of domestic chip substitution, offering a clear view of the regulatory boundaries and the impact on the industrial chain.

Topics

Tags

Body · AI translation

Article body

On November 16, the U.S. Department of Commerce's Bureau of Industry and Security (BIS) escalated its AI chip control measures, restricting Chinese companies from remotely accessing advanced computing power through overseas data centers.

The new rules state that if a data center provides computing power exceeding 12 million TPP (Total Processing Performance) to a company whose headquarters or ultimate parent company is located in China (including Macau) or other such places, then chip exporters that are aware of this situation must apply for a license before supplying advanced AI chips to that data center, and licenses will be reviewed on a case-by-case basis.

According to Morgan Stanley's estimates, 12 million TPP of computing power roughly equates to 700 to 800 H100 chips. For large-model training that often requires clusters of thousands or even tens of thousands of GPUs, 12 million TPP of computing power is negligible.

BIS also requires that data centers that had already provided large-scale computing power services to relevant companies before the rules took effect must, if they wish to continue providing services, apply retroactively for licenses within 90 days from the date the rules take effect.

In response, China's Ministry of Commerce said in a statement on the evening of November 17 that the U.S. side has generalized the concept of national security, abused export controls, and arbitrarily interfered in normal trade cooperation between Chinese companies and companies of other countries. The U.S. approach runs counter to the important consensus reached by the heads of state of the two countries, undermines market rules and the international economic and trade order, and disrupts the stability of global semiconductor industrial and supply chains. China expresses strong dissatisfaction and firm opposition to this.

The Ministry of Commerce stated that China hopes the U.S. side can work with China in the same direction and return to the correct track of resolving differences through consultation and cooperation. If the U.S. side insists on going its own way, China will take countermeasures and firmly safeguard national sovereignty, security, and development interests.

Since 2022, the United States has continued to tighten chip exports to China, blocking Chinese companies from obtaining advanced AI chips. Although it is difficult to directly purchase overseas high-end chips, Chinese AI companies can still remotely obtain computing power from overseas data centers through leasing.

Now that the new BIS rules have restricted channels for remote access to overseas computing power, the market is closely watching the subsequent impact of this restrictive move on Chinese AI companies.

Morgan Stanley analysts believe that advanced chips are crucial for training large models with trillions of parameters, so the new BIS rules may in the short term have a negative impact on Chinese AI companies' overseas training and slow down model iteration.

Affected by this news, on November 17, the Hong Kong-listed shares of leading Chinese large-model developers Alibaba, Tencent, Zhipu, and MiniMax all fell.

However, in the view of some analysts, internet giants are more resilient to risk than independent large-model companies. Nomura analysts wrote in a research report that stricter restrictions may create opportunities for internet giants, allowing them to expand their competitive advantage over independent AI model companies in foundation model research and development—the latter may face greater challenges in obtaining advanced AI computing power to train next-generation models, while Alibaba and ByteDance may be in a more favorable position thanks to their existing GPU resources and stronger financial strength.

For Alibaba, its self-developed T-Head AI chips also provide a certain buffer. Nomura analysts believe that restrictions on remote access to overseas computing power will further accelerate the adoption of domestic AI chips, benefiting companies with stronger channels for obtaining computing power resources and self-developed chip capabilities, and Alibaba has certain advantages in this regard. Morgan Stanley analysts also said that T-Head's stable chip supply channel has become a key competitive advantage for Alibaba in mitigating the impact of the new BIS rules.

Data from market research firm IDC shows that in the first half of 2026, T-Head's data center AI chip shipments were 133,000 units, ranking second among domestic players, behind only Huawei Ascend. According to a Morgan Stanley research report, Alibaba plans to increase T-Head's share of computing power contribution from more than 10% to 40%-50% in the medium to long term.

The new BIS rules are forcing more computing power demand to shift to domestic AI chips. In the view of Citi analysts, Chinese wafer foundries, semiconductor equipment manufacturers, and AI chip suppliers will benefit. On November 17, shares of several companies in the domestic AI chip industrial chain, including SMIC, Hua Hong Semiconductor, and Biren Technology, rose.

(Source: Southern Metropolis Daily) Middle of the article Embedded advertisement

Source

East Money