Greetings from a world where…
“I got through that sentence like a subject and a predicate” (what a lyric)
…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: Summer is here, pour some "cold water" on data centers
Context: In late March 2023, WeChat and QQ, two of Tencent’s main social media platforms, experienced an outage. Execs were fired. Key functions like WeChat pay were down for many hours. The root cause? A cooling system failure at a data center located in Nansha District in Guangzhou. This week’s feature translation examines the trend of liquid cooling in data centers, and how it connects to the rise of AI. It comes from Naojiti (脑极体), a tech media platform based in Tianjin. We previously featured an article from this source on how China was using AI in covid control (ChinAI#81).
Key takeaways: The AI revolution has significantly increased demand for liquid cooling, a technology used to make sure data centers don’t overheat.
High-performance applications such as training AI models require more and more powerful chips (GPUs now can power 700 watts per chip), which means hotter chips and higher cooling requirements.
Sugon, Huawei, Lenovo, Alibaba Cloud, Inspur, Nettrix are all relatively early adopters in liquid cooling. Also, Alibaba Cloud’s “Kirin” was the first product to implement chip-level liquid cooling.
The year 2022 has been called the “first year of liquid cooling.” But we’ve seen this story before. The year 2014 was called the “first year of smart homes.” This industry has cooled off: people bought bluetooth-connected appliances and then never updated.
What makes “the first year of liquid cooling” different? The article compares it to 2019, the first year of 5G commercial applications. This momentum was sustained because of the “strong pulling force of policies.” In the case of liquid cooling, China’s dual-carbon goals and East-West compute transfer project (for more on this initiative, see ChinAI#179) serve as the policies that will continue to pull liquid cooling along. In 2022, the Ministry of Industry and Information Technology set certain requirements for power usage effectiveness for new large-scale data centers, which were further strengthened by the East-West compute transfer project.
Limitations
About 80-90% of data centers still use the traditional way of cooling (air cooling). In many cases, this method can still meet the carbon reduction and power usage effectiveness goals. As Deng Xiaoping famously said, “Black cat or white cat, if it can catch mice, it's a good cat.”
Liquid cooling faces the tension between standardization and customization. you need standardization to reduce costs and scale. At the same time, data centers have to meet very different needs. According to the article: “For example, enterprise-level data centers used for AI large-scale model training and small data centers have completely different requirements for liquid-cooled server products.”
More details in FULL TRANSLATION: Summer is here, pour some "cold water" on data centers
ChinAI Links (Four to Forward)
Should-read: The U.S. May Be Overstating China’s Technological Prowess
For ChinaFile, Johanna M. Costigan crafted an informative Q&A about my recent article on China’s diffusion deficit, which argues that assessments of a state’s scientific and technological power should pay more attention to how states spread and embed new advances throughout the entire economy.
Should-read: The Who, Where, and How of Regulating AI
In IEEE Spectrum, Eliza Strickland gives us a tour of AI regulations from around the world. Includes some comments from me about China’s rules on generative AI.
Should-read: Two other articles I considered translating for this week
Special topic | AI Security: AI Security Challenges and Governance Research (link to original Chinese) — in China Information Security, an influential publication, two Peking University researchers put forward an AI governance framework. I especially liked how they distinguished between AI safety (功能安全) and AI security (自身安全), instead of looping everything under the 安全 umbrella.
What are the significant impacts of the new Ministry of Science and Technology regulations on hospitals, scientific research institutions, and enterprises? (link to original Chinese) — Caijing E-Law examines the effects of new rules on human genetic data.
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 chinainewsletter@gmail.com or on Twitter at @jjding99
While it seems as though China is falling behind in sophisticated chip design and LLM architecture, the physical issues (EG cooling) are still a place where they might have a strategic advantage over western firms. I wonder how long that edge might last.