ChinAI #210: Huawei's New Undertaking in New Infrastructure
Greetings from a world where…
video games are becoming central to American culture (I’m all in on The Last of Us show)
…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: Huawei’s New Undertaking
Context: Based on voting from the past Around the Horn issue, many readers were also interested in Huawei’s efforts to build AI computing centers, in line with the Chinese government’s “new infrastructure” initiatives. This article (link to original Chinese), from jiqizhineng [机器之能], leans a bit too much into corporate PR-speak at times, but it provides a detailed look at Huawei’s overall strategy, through the lens of the author’s recent visit to the Nanjing Kunpeng Ascend AI Computing Center.
Key Takeaways: Why is Huawei planning to build more than 20 AI computing centers like the one in Nanjing? First off, demand has grown exponentially in recent years.
According to jiqizhineng, from 2012 to 2018, AI computing power consumption increased by almost 300,000 times. Models like ChatGPT require training infrastructure (e.g., Microsoft’s cluster of 10,000 V100 GPUs).
Currently, the Nanjing center has served more than 100 clients. “Among them, AI-type enterprises and Internet-oriented enterprise users account for about 40%. There are also some traditional industries, such as manufacturing, where the main scenarios are intelligent quality inspection and safe production,” the article reports.
Second, Huawei wants to use the computer center ecosystem to establish a full-stack independent and controllable [自主可控] industrial chain.
This complete chain consists of three layers: 1) chips, servers, and other hardware equipment at the bottom; 2) basic software including database management and operating systems in the middle; and 3) application software (e.g. ERP, CRM, and industrial software) on the top.
American companies dominate many of these basic layers. The article explains: “Take the server business as an example. Chips, memory, and hard disks basically account for 90% of the output value of servers. A major Chinese server manufacturer may have an annual revenue of more than 50 billion, but it has to pay more than 17 billion to Intel alone. If Intel rebates a little more, the company's profit may be a little more; if it doesn't give rebates, then breaking even is not bad.”
In each of these layers, Huawei is trying to develop its own technology. In the hardware layer, Huawei released its Ascend 910 to compete with Nvidia (GPUs) and Google (TPUs). Its HarmonyOS and EulerOS systems, as well as its Gauss database systems, are trying to occupy the middle layer. At the upper layer, Huawei has pushed its Mindspore AI framework (see ChinAI #175 for more details).
“The whole process adopts a domestically-produced technology system. This solves the “chokehold” problem through the entire chain.” Chen Junyi, CEO of the nearby Jiangsu Kunpeng Ascend Ecosystem Innovation Center, said.
Will Huawei succeed?
AI, as an ecosystem-based industry, is very different from Huawei’s main competencies in telecommunications, which is a highly standardized market defined by maximizing product competitiveness. Huawei is trying to adapt by providing open source versions of the software systems mentioned above.
Huawei’s efforts will depend on China’s broader “new infrastructure” push. Clarified by the National Development and Reform Commission in April 2020, independence and controllability has become a national strategy to “gradually replace hardware and software such as chips, operating systems, and terminals with domestic products.”
FULL TRANSLATION: The Power of Seeds
ChinAI Links (Four to Forward)
Should-read: Thread on China vs. U.S. in AI research
Paul Scharre, Vice President and Director of Studies at CNAS, kickstarted a fruitful conversation about Nikkei’s report on China’s AI publications. Helen Toner made an important point about how U.S.-China co-authored papers defy the U.S. vs. China narrative. Per Stanford’s AI Index report, “U.S. and Chinese AI researchers teamed up on far more published articles than collaborators between any other two nations.”
Should-read: Translation Snapshot: Tech-Related Chinese National Strategies
Back in October 2022, Ben Murphy compiled some brief descriptions of CSET translations of Chinese technology strategies. As I’m working on a new research project on China’s indigenous innovation efforts, I’ve found it really helpful to go back through these documents.
Should-read: China’s Corporate Social Credit System and Its Implications
The latest SCCEI China Brief summarizes a working paper, by Lauren Yu-Hsin Lin and Curtis J. Milhaupt, on China’s corporate social credit system based on publicly-available scores from Zhejiang Province. They find: “neither better-governed nor more profitable firms received higher overall scores, but highly-leveraged firms subject to higher default risks were associated with lower total scores.”
Should-watch/listen: Huawei’s Noah’s Ark Lab hosts conversation about ChatGPT (in Chinese)
Huawei convened a cool group of people, including Tsinghua Professor Minlie Huang, to discuss ChatGPT. Topics covered: how far are Chinese language models from ChatGPT, what are the main difficulties, commercialization opportunities, and long-term impacts?
*This would be a great opportunity for people who want to contribute to ChinAI to translate some segments from this conversation. All translation work is compensated!
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@example.com or on Twitter at @jjding99