ChinAI #268: AWS China not on Cloud Nine

Greetings from a world where…

we pull down the east wall to repair the west [拆东补西]

…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: AWS China, you should lower your prices

Context: The surface-level question that drives this Leiphone article (link to original Chinese) is: Why are prices for AWS China so high? Amazon Web Services China, which provides cloud computing services through two local operators in Beijing and Ningxia, is grappling with its “pricey” reputation.

• Xiaofei Xu reports, “Some enterprise users told Leiphone that in the China region, for the same 1GB of traffic, AWS costs 0.9 RMB, while Tencent costs 0.8 RMB. This is a considerable cost difference for traffic-based users.”

• Industry insiders close to AWS estimate, “AWS product prices are about twice as expensive as Alibaba on average, and 20 to 30 percent more expensive than Microsoft. And special discounts are also difficult to request.”

The deeper question at the heart of this piece is about the future of competition for China’s cloud computing market, which may increasingly be tied to integration with large AI models. This is a more gossipy piece than ones I prefer to favor, but I think it provides some essential background context as U.S. policymakers consider restricting US cloud companies from selling services to Chinese companies.

Key Takeaways: As tech giants in China have launched a new round of “price wars” on their cloud and large model services, the air is filled with rumors about AWS China’s potential price reductions.

• One agent told Leiphone that discounts can now reach 30%. Another insider Yi Zhang pushed back, arguing that AWS China has tightened its discount policy in recent years: “Not long ago, an internal document was even issued to emphasize the need to strictly control the illegal abuse of vouchers. Therefore, the ‘30% discount rumor; may be a disguised ‘customer acquisition method’ adopted by some salespeople and agents under the pressure of price wars.”

• There’s a lot of messiness with sales agents and partnership arrangements (in which a middle-man provides the customer access to cloud services). Another industry insider Wang Gang reveals, “Some sales agents will make up a name and say that customer A has been using XX cloud before and now intends to move to AWS, but needs us to do a POC (proof-of-concept) test, we support it, and they will apply for 50k (AWS credits). Then tomorrow, they will find another reason to apply for 50k again, and collect it from here and there, but in the end, these vouchers are actually diverted to customer B.”

One important thing to understand is that AWS China is not only competing with Chinese cloud computing providers such as Alibaba and Bytedance for local Chinese customers; it is also competing with US companies such as Microsoft and Google for Chinese companies that have overseas operations [中国出海企业].

• As an example of the first category: AWS China counts Xiaohongshu, China’s Instagram-like social media giant, as its largest local customer. Three years ago, the short video giant Kuaishou was also an AWS customer, but it eventually built its own data center in Inner Mongolia. Similarly, Tesla’s China business uses AWS for its vehicles, but it is also beginning to build its own center.

• As for the second category, AWS used to be the default choice for many large Chinese companies that are expanding globally. Qing Xiao, a cloud account manager, made this point, “Even if you choose a cheaper cloud vendor, how much money can you save? It is better to be safe and choose AWS, which is relatively more stable and less risky despite its high cost.” In the other direction, Yan Lin, who works in the gaming industry, claimed that, as they expand overseas, Chinese game companies are looking to other competitors like Google Cloud because it can bundle together advertising, Android integration, and other services.

Leiphone concludes, “It is not difficult to see that Amazon’s first-mover advantage from its original products is gradually disappearing, and its latecomer disadvantage in emerging AI large model products has caused it to be slow to catch up.”

• It is important to note that applications of large-model products are still emerging, and other factors such as AWS’s advantages in elastic cloud compute could still be decisive. From the article: “AWS has lost the ‘first move’ in the era of big models.” Investor Ya Li said frankly. In his opinion, AWS is neither like Microsoft, which “deeply binds” the big model leader OpenAI and continues to lead in technology; nor like Google, which has its own Gemini series of models and can keep up with OpenAI in technology.

• The article expands on this point: “Many Chinese companies with overseas operations have reported to Leiphone that at present, in terms of AI big models, Microsoft is in the first echelon, Google and others are in the second echelon, and AWS belongs to the third echelon…Some corporate customers have described to Leiphone that at the POC test sites of some large model projects, whenever they meet teams such as Microsoft and Google, AWS often becomes a ‘runner-up’.”

***Note: Names of sources cited in the article are all pseudonyms.

FULL TRANSLATION: AWS China, you should lower your prices

ChinAI Links (Four to Forward)

Must-read: Goldfish Memories

China Media Project translates and analyzes a popular post on the “rapidly collapsing” Chinese internet:

Written by He Jiayan (何加盐), an internet influencer active since 2018, the essay concluded, based on a wide range of searches of various entertainment and cultural figures from the late 1990s through the mid-2000s, that nearly 100 percent of content from major internet portals and private websites from the first decade of China’s internet has now been obliterated. “No one has recognized a serious problem,” wrote He. “The Chinese-language internet is rapidly collapsing, and Chinese-language internet content predating the emergence of the mobile internet has almost entirely disappeared.”

Should-read: Import AI 374 - China's military AI dataset; platonic AI; brainlike convnets

Much thanks to Jack Clark for featuring my recent article on permissive action links and the history of U.S. sharing/non-sharing of nuclear safety and security technologies with other states. Jack captures some of the article’s key points and distills lessons for the AI safety community.

Should-read: Open-Source Assessments of AI Capabilities: The Proliferation of AI Analysis Tools, Replicating Competitor Models, and the Zhousidun Dataset

Another item featured in last week’s ImportAI issue was an article by Berkeley Risk and Security Lab (BRSL) researchers on their discovery of a military-relevant Chinese dataset (608 images of military naval vessels equipped with Aegis combat systems). In their preprint, BSRL “demonstrate that an analysis of the Zhoushidun dataset as a quantitative assessment capability represents a potentially useful predictor of China’s machine-learning arsenal.” Authors: Ritwik Gupta, Leah Walker, Eli Glickman, Raine Koizumi, Sarthak Bhatnagar, and Andrew W. Reddie.

Should-read: The Race for U.S. Technical Talent: Can the DOD and DIB Compete?

By Diana Gehlhaus, James Ryseff, and Jack Corrigan, this CSET report uses LinkedIn data to systematically measure flows of tech workers across defense and nondefense sectors in the U.S. economy. Published back in August 2023, so I’m late to catch-up on this reading, but this has been immensely helpful to my research, as I start a new project on civil-military integration.

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).

Also! Listen to narrations of the ChinAI Newsletter in podcast format here.

Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99