Greetings from a world where…
“we stayed up so late, possessed by delirium, that we came up with a theory of everything, only we forgot to write it down”
…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: Security AI chips in chaos for 2 years, but no heroes in sight
Context: Currently, there’s obviously a lot of interest in the chips used to train AI models. I’m equally, if not more, curious about the chips used to run AI models on end-side applications. Take, for example, Huawei subsidiary HiSilicon’s 3559A, a security AI chip launched back in 2018. It gives surveillance cameras a boost in computing power (4TOPS to be exact) so that they can do additional data processing at the source, such as performing facial recognition. In August 2020, the U.S. strengthened restrictions that limited HiSilicon’s access to software from U.S. companies to design chips. In this week’s feature translation, a Leiphone article (link to original Chinese) packed full of insightful interviews, we explore what happened next.
Key Takeaways: Before August 2020, HiSilicon dominated the 3-5 billion RMB security chip market, controlling 70% of the market share for Internet Protocol Cameras (IPCs = digital video camera that receives control data and sends image data via an IP network)
HiSilicon took advantage of the government’s Sharp Eyes Program, which aimed to provide 100% security camera coverage of China’s public spaces by 2020.
The year 2018, when HiSilicon launched the 3559A, was the hottest time for AI in China; at the time, the security market had a low penetration rate for intelligent cameras.
Two years of turmoil, three major beneficiaries, zero HiSilicon replacements
U.S. restrictions in 2020 disrupted the security chip market, causing massive price surges for security chips. Hisilicon 3559As went from 500 RMB to over 3,000 RMB, and even mid-end/low-end chips registered huge price spikes.
Two years later, three companies — Novatek[联咏科技], Fullhan[富瀚微], and SigmaStar[星宸科技] — have filled the gap left by HiSilicon for mid-end and low-end security chips.
However, these competitors still have not made a chip that’s competitive with HiSilicon’s 3559A. Jianguo, a pseudonym for an industry insider interviewed by Leiphone, explains: “First of all, to design an AI chip with 4TOPS computing power three or four years ago, the technology itself is a threshold; secondly, achieving 4TOPS computing power requires a relatively advanced process, such as 12nm, in order to achieve a balance between performance and power consumption… and interface IP needs to be updated…the licensing fees for these IPs are all over 10 million RMB, which greatly increases the cost.”
The road ahead for edge AI chips: diversify beyond security applications
According to the article: “People in the industry believe that the growth of the security market is slowing down, and the possibility of becoming the next HiSilicon in the security chip market is very small.”
Here’s Jianguo again: “Now everyone should want to widen the road first, and then see what applications can use the edge AI chips with large computing power.”
Various forms of robots, including sweeping robots and logistic robots are a potential big market for high-performance chips that can run computer vision algorithms. In addition to robots, these chips could be applied in smart cars and smart transportation.
The article concludes with some intrigue: “It is reported that HiSilicon will return to the security market at the end of this year or early next year at the earliest. If the king returns, what kind of problems will the current beneficiaries worry about?”
FULL TRANSLATION: Security AI chips in chaos for 2 years, but no heroes in sight
ChinAI Links (Four to Forward)
Should-read: GovAI Summer Fellowship 2022 Wrap Up
Cool to see all the projects that the recent class of GovAI summer fellows worked on, under the stewardship of GovAI’s research manager Emma Bluemke. Special shoutout to Felipe Calero Forero, who researched the history of Sematech, in an effort to derive lessons for state-sponsored industry cooperation in AI.
Should-read: China’s Xi Stacks Government With Science and Tech Experts
Karen Hao reports for The Wall Street Journal: “Chinese leader Xi Jinping has packed the top ranks of the Communist Party with a new generation of leaders who have experience in aerospace, artificial intelligence and other strategically important areas…
…Chinese officials with technical expertise occupy 81 seats, nearly 40% of the total, in the new Central Committee—the elite body that decides major national policies—according to data compiled by the Washington-based Brookings Institution think tank and shared exclusively with The Wall Street Journal. That compares with less than 18% in the previous Central Committee.”
Should-read: After the CHIPS Act: The Limits of Reshoring and Next Steps for U.S. Semiconductor Policy
Vishnu Kannan and Jacob Feldgoise, in a paper for Carnegie Endowment for International Peace, provide a sensible and detailed analysis of the implementation challenges for the CHIPS Act, as well as recommendations that go beyond the CHIPS Act to secure a stronger foundation for U.S. semiconductor policy.
Should-attend: Introducing the Emerging Technology Observatory
CSET’s newly launched Emerging Technology Observatory (ETO) offers data and insight on critical emerging technology issues. Join CSET virtually 4-5PM eastern time on December 5, as ETO Analytic Lead Zachary Arnold introduces ETO’s three inaugural tools: the semiconductor-focused Supply Chain Explorer, the Country Activity Tracker for national-level AI indicators, and the Map of Science, our exploration engine for global research in science and technology. Zach will be joined by CSET Senior Advisor Melissa Flagg and CSET consultant John VerWey to discuss the insights that can be drawn from ETO’s unique resources.
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