ChinAI #377：美国数据中心霸权——中心—卫星结构

Greetings from a world where…

this could be the Iowa football’s year

…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).

Feature Translation: In the US-China AI race, the deciding factor isn’t compute, it’s another thing

Context: Among U.S. policymakers and Western media outlets, I’ve seen a narrative popping up that China is winning the AI race by spreading its technology through infrastructure deployments abroad. For instance, this Newsweek article observes, “Beijing is only too happy to spread its tech through existing channels like the Belt and Road Initiative infrastructure export program.” It’s always clarifying to read how Chinese experts view the situation on the competition over the global deployment of data centers. The winner of last week’s Around the Horn issue, a Beijing Cultural Review essay published back in 2024 by Yanhui He, researcher at Peking University’s school of journalism, presents a very different perspective: the U.S. maintains hegemony in global data center construction, based on a “center-satellite” structure.

Key Takeaways: Sometimes we forget the baseline reality that Google, Microsoft, Amazon and other U.S. companies own more than half of the globe’s hyperscale data centers.

This Beijing Cultural Review piece reminds us that American “hegemony in computing power” rests not just on chips and algorithms but rather large-scale appropriation and integration of resources such as computing power, electricity, cooling energy, and land. “Regions such as Ireland and the Netherlands have thus become strategic hubs for companies like Amazon and Google, thanks to their extensive internet infrastructure,” He writes.
Based on 2023 data, the U.S. hosted 5,000 data centers, whereas China, Germany, and the UK combined to host less than 2,000. Again, this doesn’t even count all the overseas data centers owned by U.S. companies.

The essay alleges that the U.S. advantage is rooted in a “center-satellite” [中心—卫星] structure that relies on exploiting other countries. We’ve noted some examples from European countries that have robust internet infrastructure. As France and Germany endeavored to build their own sovereign clouds, they also continued to rely on Microsoft and Google cloud infrastructure.

He continues, “This ‘center-satellite’ dependency structure is even more pronounced in developing countries. In Nigeria, 80% of public health data is stored on cloud servers owned by companies such as Google, Amazon, and Microsoft. Local scholars have pointed out that this practice primarily benefits internet companies by boosting their profits and data acquisition, without enhancing Nigeria’s own capacity to store public data. In 2022, the Indian industrial giant Adani Group signed a cloud computing contract with Google to migrate its core applications and data to Google Cloud, while the African bank NedBank announced that Microsoft would manage its customer data.”
Notably, the author frames the U.S.’s global deployment of data centers as exploiting these satellite zones, which has triggered counter-movements against data centers and cross-border data flows. However, U.S. dominance is sticky: “Even for entities located outside the U.S., the cost of routing data to the U.S. for processing remains relatively low. For example, data from numerous European enterprises is processed at U.S.-based data centers before being utilized by local users. The majority of micro and small enterprises oppose Europe's stricter regulation of cross-border data flows, as it would significantly increase their operating costs.”

Where does China fit into this center-satellite structure? The essay positions China’s data center approach as constituting a “special case” system [例外体系]. In contrast to Europe, China’s domestic data center market is dominated by Chinese companies; the only international provider that ranks among the top five is Amazon (with 8.1% share).

In this sector, China’s restrictions on foreign investment have played a mixed role. They’ve protected the domestic market, requiring foreign entities to obtain a license to operate data centers. At the same time, Microsoft and Amazon have entered the market by partnering with Chinese firms that handle local operations.

He also points out that China’s digital expansion is beholden to the U.S.-dominated “center-satellite” order. For example, NetEase Games, Bilibili, and other Chinese companies with overseas operations rely heavily on Amazon: “It commands nearly 80% of the market share for data processing services supporting Chinese companies' international expansion.”

The essay concludes that the current AI race can also be seen as “a competition over the strategic deployment of data centers” where satellite zones bear the costs of data centers on behalf of the dominant power.

FULL TRANSLATION: In the US-China AI race, the deciding factor isn’t compute, it’s another thing

ChinAI Links (Four to Forward)

Must-read: Can China close the high-end AI data gap?

Published in Inferential Gap, this was one of the most informative analyses I’ve read in the past month. The post is motivated by a Forbes report that Chinese AI labs rely on U.S. suppliers for post-training data (e.g., step-by-step demonstrations provided by experts, verifier scores on agent completion tasks). Incorporating original research and calculations, the author provides an overview of China’s post-training data landscape and concludes, “I don’t think the U.S. providers possess a permanent, across-the-board moat when it comes to high-quality post-training data.”

Should-read: Models for International Private Sector Cooperation on Frontier AI Safety

This report, led by Conor McGlynn and Saad Siddiqui, points out the role of private sector cooperation in addressing frontier AI safety risks. I especially like the section that reviews other institutional models in related domains. I was happy to contribute to this one.

Should-read: There Is No A.I. “Race”

For The New Yorker, Chang Che debunks the “What About China” question in debates over whether the U.S. can regulate and/or slow down AI development amidst a technological race with China. His piece also incorporates insights from my book and comments from myself.

Should-read: China races to build data centres in bid for AI supremacy

In a Financial Times report, Eleanor Olcott traveled to Ulanqab, Inner Mongolia to understand China’s data center buildout. See also last week’s Around the Horn for a Chinese-language perspective, written by someone who calls Ulanqab their hometown.

Thank you for reading and engaging.

These are Jeff Ding’s (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.

Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).

Any suggestions or feedback? Let me know at [email protected] or on Twitter at @jjding99