ChinAI #330: Chinese Universities Top Global CS Rankings

Plus, a deeper dive into the CSRankings methodology

Sep 29, 2025

Greetings from a world where…

football is an infuriating sport

…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: Ending CMU’s Dominance, Tsinghua University Takes the Top Spot in CSRankings! Peking University Leads in AI

Context: Developed by University of Massachusetts Amherst Professor Emery Berger, CSRankings has emerged as an influential ranking system to identify the top universities in computer science, including the subfields of AI, systems, theory , and interdisciplinary areas (e.g., human-computer interaction). When I give presentations on China’s AI development, my slides on Chinese AI talent draw on CSRankings metrics. Thus, I was intrigued to see that Chinese science and tech media are closely tracking the latest CSRankings updates, as reflected in this week’s feature translation (link to original xinzhiyuan [AI_Era] post).

The reason I use CSRankings in my own research is because it avoids two issues that beset other ranking systems: 1) unlike US News and World Report, CS Rankings doesn’t rely on subjective surveys that may be too beholden to past reputations rather than present-day performance; 2) citation-based metrics can be manipulated by universities and “citation cartels”. Instead, CSRankings simply judges departments based on their faculty’s publications at selective computer science conferences.

Key Takeaways: AI_Era highlights that Tsinghua University has taken the top spot in the 2025 CSRankings for the first time, overtaking Carnegie Mellon University (see screenshot below).

Tsinghua University, Shanghai Jiao Tong University (my dad’s alma mater), Zhejiang University, and Peking University ranked 1st, 3rd, 4th, and 5th, which means Chinese universities had four of the top five ranks
From the article: “With Tsinghua University reaching the top spot, China’s global ‘discursive power’ [话语权] in computer science has significantly increased.” In terms of this quiet shift in the global landscape of computer science academia, AI_Era concludes, “Experts predict that if Chinese universities continue to focus on open-source ecosystems, international research collaborations, and cultivating young talent, achieving half of the top 10 spots in the CS Rankings within the next five years is ‘not an unattainable goal.’”

Source: CSRankings (all fields on, 2024 to 2025, institutions from around the world)

If we zoom into the AI subcategory of the CSRankings, however, a different picture emerges.

The AI_Era article claims that Peking University ranks No. 1 globally in the AI category — based on just the year 2025 and all the CSRankings AI conferences in computer vision, natural language processing, and information retrieval, etc. Here’s how they sorted things.
If I were to pick the best indicator of the strongest AI universities around the world, I would instead use the three most selective machine learning conferences (ICLR, ICML, and NeurIPS) and use data from the last five years, as one-year snapshots can be distorting. For my results, see image below.

Based on this methodology, Chinese universities are certainly impressive ,with three of the usual suspects ranking in the top 10 (Peking University, Tsinghua University, and Shanghai Jiao Tong University).
Still, overall, CMU, MIT, and UC Berkeley are at the top, and US universities hold 13 of the top 20 slots. In my opinion, this gives a better picture of the world’s top 20 AI universities than the AI_Era query.

I’ll conclude with a few more musings on the influence of CSRankings, as it’s remarkable that smart Chinese tech reporters are following it so closely. Cynthia Rudin, a Duke University Professor who leads an Interpretable Machine Learning Lab, wrote up a fascinating post that pinpoints some of the flaws in this rankings system. A few details from her post that stuck out to me:

Rudin points out that CSRankings may capture a lot of low-quality papers: “All conferences accept papers of widely varying content, quality, and topic. The review process at these conferences is fast and sloppy, and reviewers often have little experience or are from a different subfield than the paper they are reviewing. Many papers at the conferences are actually not particularly interesting or revolutionary - even reading the titles of the papers would allow you to understand this - and many other papers are just ‘click bait’ papers that generate attention from reviewers on hot topics but have little insight, and no new important methods or applications…CSRankings encourages this fast and sloppy way of doing science to get as many papers as possible into these few venues.”
Rudin’s post also mentions that CSRankings is shaping the computer science academic job market: “I heard recently about someone being hired for a faculty position (thankfully not at Duke!) so the department could count their copious low-ish quality papers for CSRankings. This is too disgusting for me to comment on it further.”

FULL TRANSLATION: Ending CMU’s Dominance, Tsinghua University Takes the Top Spot in CSRankings

ChinAI Links (Four to Forward)

Must-read: Every Great Tech Hub Needs Regulation

I think one of the most common misconceptions about China’s AI policy is that there’s an “all gas, no breaks” regulatory approach. In her substack The Long Game, Johanna Costigan expertly breaks down how Hangzhou wants to both move fast and also regulate. In her analysis of some recent court cases involving generative AI services, Hangzhou is “building a new reputation for itself: an innovator in AI regulation targeting areas such as copyright infringement and unfair competition.”

Should-read: U.S. Loses Appeal for Chinese AI Researchers

Given this week’s translation, the U.S. government should be doing everything possible to support our own higher education institutions and attract talent from around the world, especially China, right? Sadly, as Juor Osawa and Qianer Liu report for The Information, we are doing the exact opposite.

Should-apply: US-China AI Governance PhD Fellowships

The Future of Life Institute supports PhD fellowships (with $40,000 annual stipend) for students interested in working on US-China AI governance research. Deadline to apply this cycle is November 21. See page for more details about research priorities.

Should-read: Correction to ChinAI #328 on Cold Reality for Chinese AI Start-ups

One problem with some of the Chinese-language sources featured in ChinAI is that the fact-checking standards are not as rigorous as those at the top English-language publications. Thanks to Matt Sheehan for doing some of that double-checking for us in the comment section of this previous ChinAI issue: the original Chinese article had wildly overstated the total financing amount of Chinese AI funding rounds in the first half of 2025 (by almost 40 times).

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

ChinAI #329: China's Shifting Cloud Market — A Review of Q1 and Q2 data

Jeffrey Ding

Sep 22, 2025

Greetings from a world where…

for the first time in what feels like 20 years, the offense carried the Hawkeyes to a win

Feature Translation: Mid-Year Review of Major Cloud Providers

Context: This week’s feature translation (link to original Leiphone article) is a really meaty check-in on China’s cloud computing landscape, with a focus on three trends: 1) the DeepSeek-fueled AI boom; 2) the rise of Volcano Engine (Bytedance’s cloud division) and diverging strategies of domestic Chinese giants; 3) adjustments by foreign cloud providers Microsoft and AWS. I’m very grateful to Marianne Lu for contributing this week’s feature translation and analysis. Marianne is a technology and security policy fellow at RAND. She graduated from Stanford University, where her thesis focused on U.S.-China AI competition.

What follows is Marianne’s analysis (lightly edited by me):

Key Takeaways: Cloud companies operating in China saw a performance surge in the first half of 2025, driven by the DeepSeek-fueled AI boom.

The article starts with an anecdote: earlier this year, Zhang Xun, a sales rep for Volcano Engine (Bytedance’s cloud offering), worried about meeting his sales quotas. DeepSeek’s release, however, “dropped a bombshell in the market,” driving AI demand across industries and sending cloud usage soaring. As he put it: “GPUs flew off the shelves in the first half of the year.”
A surprising source of demand came from government agencies and state-owned enterprises, which at times showed more enthusiasm for deploying DeepSeek than private firms.
All major providers rode the DeepSeek boom, but Volcano Engine emerged as the fastest-growing player, with revenues more than doubling year-on-year. Why? ByteDance had stockpiled GPUs that once sat idle but suddenly became scarce and valuable, and then also priced them at steep discounts.

But the AI momentum may not have carried into the second half of the year.

As the third quarter began, sales teams saw business tighten. Finding new customers proved difficult, with one industry professional remarking that beyond a handful of large firms, few notable AI applications existed in China.
Meanwhile, the cloud price war showed no signs of easing. Large clients tend to bypass vendors by building their own infrastructure, and smaller ones view cloud companies’ offerings as interchangeable, often leaving sales teams with no choice but to keep competing on price.
The cooling of cloud hype mirrors the pattern noted in an earlier ChinAI newsletter, where DeepSeek all-in-one machines struggled to diffuse past early adopters. Similarly, while DeepSeek gave cloud sales a short-term lift, sustaining long-term growth looks uncertain.

Both domestic and foreign cloud providers are recalibrating their strategies in China, with AWS fighting to stem client losses to Chinese competitors, and Microsoft reducing its exposure to Chinese government and enterprise customers.

When it comes to domestic companies, Alibaba Cloud and Volcano Engine are chasing growth, with Alibaba in particular doubling down on globalization to tap revenue beyond the crowded domestic market. Tencent Cloud and Huawei Cloud, by contrast, have shifted gears, scaling back investment and prioritizing profitability in a sector that is not central to their bottom line.
As domestic cloud firms pursue diverging strategies, foreign companies are also recalibrating their approach to the market.
AWS has come under sustained pressure from domestic rivals. Over the past few years, domestic cloud providers have chipped away at its client base, poaching first small businesses and then larger clients like Xiaohongshu, Ctrip, and BYD. AWS has fought back with steep discounts, offering prices up to 60% below Alibaba’s rates, but with limited success. As one foreign cloud manager lamented, “When it comes to price wars, who can outdo the domestic cloud providers?”
To stay competitive with Chinese clients going overseas, AWS has also launched the Pan Amazon initiative, bundling cloud with Amazon’s wider ecosystem of logistics, e-commerce, and ads. The program’s results remain to be seen, but AWS retains an edge in its global reach, landing billion-dollar contracts this year with TikTok and SHEIN.
Despite headwinds, Microsoft has seen steady gains in China. OpenAI’s early lead prompted many Chinese startups and tech giants to lean on Microsoft Cloud for model training, and more recently, for a wider range of AI offerings. But to adapt to the market, Microsoft has also reshaped its priorities. First, it cut back on its government and enterprise clients, who were slow to ramp up cloud usage and increasingly constrained by domestic substitution rules. At the same time, it has continued to shift from traditional software to cloud services.
Jeff’s comment: to me, this was a jaw-dropping stat from the article: “In fact, over the past two years, the Chinese market has consistently been the primary consumer of Microsoft Cloud’s AI services. In the second half of last year, roughly one-quarter of Microsoft Cloud’s global OpenAI-related business revenue came from the Chinese market.”

Looking forward, the article predicts that the second half of the year will see a new round of competition, “with the battle set to break out in full force across both domestic and international markets.”

Analysis: One of the things that struck me about the piece was how foreign cloud firms are navigating the Chinese market. Yes, geopolitics plays a role––Microsoft easing away from government clients is an example––but much of it comes down to the same difficult reality domestic players face: finding and keeping customers in an increasingly cutthroat market. The article leans into this with almost novelistic “battle” imagery, bits of gossip, and lively anecdotes, like Tencent Cloud employees now having to track not only Alibaba’s moves but also Volcano Engine’s, doubling their surveillance load. Translating it was, dare I say, even fun.

FULL TRANSLATION: Mid-Year Review of Major Cloud Providers: Soaring GPU Revenues, Diverging Strategies Among Domestic Players, and Ongoing Adjustments by Foreign Cloud Providers

ChinAI Links (Four to Forward)

Must-read: Topology of “China AI”

In the Concurrent newsletter, Afra maps out a topology of China’s AI ecosystem. It’s a conversation with H, an AI engineer at one of China’s largest tech companies who leads an open-source project:

A few weeks ago, I met him in Shenzhen during a self-organized "China AI/tech tour" with several US tech writers.¹ After peppering him with questions—the most obvious one being "What's Silicon Valley's biggest misunderstanding about China's AI?"—I discovered someone so thoughtful and genuinely bilingual.
His daily routine reveals the paradox: wake up, doomscroll X, debate on LessWrong, drive his NIO to work, then use English more heavily than when he studied in California. He loves watching Westworld and treats Gödel, Escher, Bach as AI scripture. Apart from geography, H could be any engineer at Anthropic or Google. He represents the global open-source AI community: following identical cultural patterns, reading the same viral posts, grappling with the same technical problems, laughing at the same memes. He just happens to work in China.

Should-read: Selective Restraint: How China Regulates Facial Recognition in 2025

Sabina Nong, ChinAI reader and recent Stanford MIP graduate, took up my challenge from a few issues back to investigate cases when China restricts digital surveillance. It’s really cool to see where Sabina’s curiosity led her: an in-depth analysis of China’s recent regulations on facial recognition technology and video surveillance — including their limitations and how they fit in a broader logic of authoritarian governance.

Should-apply: US-China AI Governance PhD Fellowships

Sponsored by the Future of Life Institute, The Vitalik Buterin PhD Fellowship in US-China AI governance collaboration is for PhD students who plan to work on US-China AI governance research, or for existing PhD students who would not otherwise have funding to work on US-China AI AI governance research. The deadline is November 21, 2025.

Should-read: ChinAI #305: Computing Power Shifts in the AI Inference Era

This week’s issue focused on the cloud giants in China. If you’re interested in how smaller cloud computing vendors are trying to compete on the edge, check out this previous issue of ChinAI.

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.

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

ChinAI #328: The Cold Reality for Chinese AI Start-ups

Jeffrey Ding

Sep 15, 2025

Greetings from a world where…

Key & Peele sketches never get old

Feature Translation: The Realities of AI Start-ups in 2025

Context: A branch of TMTpost (a Chinese tech, media, and telecommunication outlet) interviewed twelve developers, five AI community mods, and seven investors to get their take on the current state of the AI startups ecosystem in China. Much thanks to Lily Li for contributing this week’s feature translation and commentary. A recent graduate of University of Toronto with a PhD in theoretical computer science, she is making the transition to the AI safety field through projects in technical and governance research.

What follows is Lily’s analysis (lightly edited by me):

Key Takeaways: For software-based AI startups, the general consensus is that Chinese consumer and businesses are unwilling to pay for AI products and services. The article postulates that:

Chinese consumers are unwilling to pay because they have grown accustomed to browsing the web from their phones and complex/standalone features do not integrate well into their viewing habits.
- The article quotes one developer saying, “The general consensus is: launch products overseas. Develop it in secret and make money on the down low. Launching products domestically is a loss-making endeavor. If it's truly unfeasible, relocate everyone to Singapore or elsewhere.” Chinese video tool Kling AI, for instance, generates 70% of its $100 million USD in annual revenue from overseas.
Chinese businesses are unwilling to pay because labor is relatively cheap and most industries are not fully digitized and thus cannot take advantage productivity software. From the article: “The ARR (Annual Recurring Revenue) of leading AI companies is five to one hundred times higher in North America and the average annual procurement budget for North American businesses is nearly 10 times higher.” (see image below)

The interviewees were far more optimistic about hardware-based Chinese AI startups. China already has many tech hardware companies with robust supply chains and access to vast pools of talent. Taking advantage of these strengths may allow these startup to overtake their competition.

The secondary markets have reflected this sentiment as opportunities to invest in hardware projects become a hot commodity in the past six months.

AI startup have a high barrier of entry and VCs find themselves competing for the same list of about 200 top talent with the technical background and powerful network necessary to launch a successful AI product. While some investors court “second-tier developers” with competitions, developer communities, boot camps, and free compute credits, the path for these developers is necessarily an arduous one.

Large Chinese tech companies have been caught flat-footed when it comes to AI. Their models are much weaker than those of frontier US companies and they are still mostly focused on their usual revenue metrics (traffic on their existing platforms) rather than making investments to AI infrastructure and developing the China AI ecosystem. The result is the loss of AI talent to opportunities abroad and their ecosystem is generally unattractive to domestic developers.

The article cites this blunt assessment from a media outlet, “The US is using trillions of dollars of 'asset-heavy' investments to turn AI into the next generation’s water and electricity, while China is pursuing AI with an 'asset-light' mindset, still focused on monetizing views.”
Another section: “Several leading developers remarked that if they were to start a new project, they most likely would not choose the AI ecosystem of a major domestic company, nor would capital from market-driven funds be their top choice. Instead, they prefer Open AI’s incubator because it offers both financial support and a robust AI ecosystem.”

FULL TRANSLATION: The Realities of AI Start-ups in 2025

ChinAI Links (Four to Forward)

Must-read: China's Big AI Diffusion Plan is Here. Will it Work?

Matt Sheehan launched his Substack newsletter with a meticulous breakdown of China’s AI Plus initiative. He extensively builds out both the bearish and bullish case against AI Plus and then concludes with his own judgement. As just one example of how Matt does his homework on these issues, check out his comments on last week’s ChinAI full translation: Matt takes the extra step to poll multiple contacts to verify whether the 90% target applies to the whole economy or six priority sectors.

Should-read: Taiwan’s “silicon shield” could be weakening

I try to keep up-to-date on Johanna Costigan’s work, and I’m late to this MIT Tech Review article she published last month on Taiwanese semiconductor champion TSMC. It’s a detailed analysis of how this firm is navigating China’s ramped up pressure and U.S. demands for friendshoring.

Should-read: Working with US CAISI and UK AISI to build more secure AI systems

This was a really interesting OpenAI update on their collaborations with the U.S. and UK AI security institutes, which gave these institutes access to non-public prototypes of their systems. As they write, “These collaborations represent some of the deepest public-private collaborations on evaluating real-world frontier AI systems for security and against misuse, and we hope they serve as promising models for the field.”

Should-read: Silicon Valley enabled brutal mass detention and surveillance in China, internal documents show

This Associated Press investigation, based on independent reporting by Yael Grauer, finds that U.S. tech companies have supplied technologies that directly support Chinese surveillance systems. The evidence is sourced from tens of thousands of leaked emails from a Chinese surveillance company as well as thousands of procurement records.

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.

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

ChinAI #327: Critiquing China's AI Plus Plan

A Strategic Reading of China’s Three-step AI Strategy for the Next Decade

Jeffrey Ding

Sep 08, 2025

Greetings from a world where…

another season of a one-dimensional Iowa Hawkeyes offense beckons

Feature Translation: The Nation Sets the Tone for “AI Plus”

Context: Last week’s Around the Horn vote produced a runaway winner: a strategic reading of China’s AI Plus initiative (link to original Chinese), from Synced (jiqizhixin). My readers love a plan. Last month, China’s State Council released implementation guidelines for its AI Plus initiative, which was initially proposed in a 2024 work report. The jiqizhixin report deems these implementation guidelines as more targeted and systematic than China’s New Generation AI Plan from 2017.

And you all were right to choose this one: it’s a useful breakdown, and more importantly, it serves as a good launching pad for me to give some grouchy comments and hot takes.

Key Takeaways: The plan positions AI as China’s key growth engine for the future (“the core of new quality productive forces”, according to jiqizhixin’s reading). In the past, these productivity forces have included emerging industries such as nuclear fusion or quantum computing. Here, AI stands above the rest. These implementation guidelines lay out three-step blueprint:

By 2027: achieve over 70% penetration rate of intelligent terminals and agents in six key sectors.
By 2030: achieve over 90% penetration rate of intelligent terminals and agents across the entire economy.
By 2035: AI will usher in an intelligent economy and society. The jiqizhixin report interprets this as: “AI will be as universally adopted as electricity and the internet, becoming the ‘base infrastructure’ of society.”

Let’s go through some of the planks that stood out to me. First, the six key sectors targeted for a 70% AI adoption rate are: science and technology, industry, consumption, social well-being, governance, and global cooperation.

This is such a weird mishmash of industries to highlight as early adopters. Global cooperation is: a) not really a sector, b) will see much slower AI diffusion as compared to, say, finance, and c) does not have an outsized impact on economic productivity. Industrial devevelopment, the second sector listed, basically includes the entire economy, as it includes industry, agriculture, and services. The one that makes sense to me is AI for science, as a promising application sector where AI could “transform the R&D chain, significantly shortening the distance between the laboratory and the market.”

The goals and timelines are unbelievably optimistic, especially 90 percent adoption rate across the entire economy by 2030! As the article points out, “The 90% figure means that nearly every industry, every organization, and even the majority of individual users will be using smart devices or agents, moving AI from application-specific landing towards an economy-wide driver.”

People often ask me if I think that Chinese policymakers are reading my book. It makes me laugh for a number of reasons. First of all, I doubt many people have actually read the whole book. So, no, Xi Jinping and his advisors are not poring through my duration model of how fast 76 different countries reached a certain level of computerization. Even my wife skipped that section.
Second, if Chinese leaders were taking my book seriously, they would have picked a very different benchmark. In the section titled, “The Decisive Years in the US-China AI Competition,” here’s what I wrote, “Thus, regardless of which arrival date is used, if AI, like previous GPTs, requires a prolonged period of gestation, substantial productivity payoffs should not materialize until the 2040s and 2050s.”
Ding, Jeffrey. Technology and the Rise of Great Powers: How Diffusion Shapes Economic Competition. Princeton University Press, 2024.
- Stay with me here as we run the numbers. How long does it usually take for a general-purpose technology (GPT), like AI, to make its mark? The first step in this exercise is to establish an “arrival date” for the GPT. My book uses Jovanovic and Rousseau’s measure1 of when a GPT attains 1% adoption in the median sector. According to this approach, electricity arrived as a GPT in 1894, and the computer arrived as a GPT in 1971. Based on these arrival dates, it took over three decades for both electrification and computerization to reach 70 percent of households in the U.S. (the fastest adopting country). *Note: According to Jovanovic and Rousseau, it took even longer than three decades for the median manufacturing sector to achieve 70 percent adoption of electricity (as a share of total horsepower)
- In Technology and the Rise of Great Powers, I estimate AI’s arrival as a potential GPT to sometime in the mid-to-late 2010s. This is based on a 2018 census survey metric, with 2.75 percent of firms in the median sector reporting the use of AI technologies. If AI follows the path of electricity and the computer, we will not even get close to 70 percent adoption across the economy — let alone 90 percent! — until after 2040.
In sum, to my friends who are so afraid of information hazards that they forget how to produce information in the first place: Don’t worry! I have not written a playbook for Chinese leaders to adopt. They probably aren’t reading it; and if they are, they’re certainly not reading it closely.

Okay, I should say something nice about these AI Plus implementation guidelines, which should be complimented for taking diffusion seriously. I did appreciate the sections on open-source communities: “Establish evaluation and incentive mechanisms for open-source contributions, and encourage universities to recognize such contributions in student-credit and faculty-assessment systems” (translation from the superb Geopolitechs newsletter).

Two things can be true at the same time about China’s open-source AI ecosystem. First, Chinese labs are dominant in producing top open-source AI models. It was notable to see jiqizhixin cite Interconnects, Design Arena, and Hugging Face, where Chinese models top leaderboards. Second, these student-credit and faculty-assessment planks are designed to address low activity and weak participation levels in China’s overall open-source software community.

One last small snack for thought. This interpretation of the AI Plus implementation guidelines mentioned one Chinese AI regulation that I had overlooked: Measures for the Security Management of Facial Recognition Technology Applications, effective June 2025.

From the jiqizhixin article: “The core of these measures is to strictly regulate the use of facial recognition…Data must be stored locally and retained for a limited period, and important applications must be registered. Facial recognition must not be mandatory as the sole method for identity verification, and alternative methods must be provided. Data collection in public places must be reasonable and legal, and the deployment of equipment in private spaces is prohibited.”
My question is why was there no English-language analysis of these regulations (including from ChinAI)? I’ve always thought this would be fascinating project for an enterprising researcher: when and why do authoritarian countries curtail the expansion of digital surveillance? Another example: we had all the coverage about the dystopian ramifications of China’s Covid-era health QR codes. Did anyone cover the rollback?

Full Translation: The Nation Sets the Tone for “AI Plus”: a Strategic Reading of China’s Three-step AI Strategy for the Next Decade

ChinAI Links (Four to Forward)

Must-read: The AI Plus initiative – China’s blueprint for AI diffusion

For Trivium China, Kendra Schaefer and Tom Nunlist did their own in-depth reading of China’s AI Plus implementation guidelines. This note was originally published in their Trivium Tech Daily newsletter.

Should-read: China Has a Different Vision for AI. It Might Be Smarter.

For The Wall Street Journal, Josh Chin and Raffaele Huang report on China’s approach to AI development, which differs from a Manhattan Project for AGI. Includes some comments from me.

Today’s issue was dense, so let’s close with two fun recommendations

Should-read: Assassin’s Apprentice

Just finished this fantasy book by Robin Hobb. I’m not a big pet person, but I kinda get it after reading this one.

Should-watch: Eastern Gate

I’ve watched 5 out of 6 episode of this Polish spy thriller on HBO. Turn on the original Polish audio and use the subtitles!

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.

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

Jovanovic, Boyan, and Peter Rousseau. “General Purpose Technologies.” In Handbook of Economic Growth, 1st ed., edited by Philippe Anghion and Steven Durlauf, vol. 1. Elsevier, 2005. https://econpapers.repec.org/paper/nbrnberwo/11093.htm.

ChinAI #326: Around the Horn (21st episode)

Jeffrey Ding

Sep 01, 2025

Greetings from a world where…

summer has come and passed (hopefully)

Around the Horn (21st episode)

In many ways, Around the Horn issues reflect the whole purpose of ChinAI — if you want to learn more about China, read the stuff that smart Chinese people are reading and writing.

This is our 21st trip around the horn. Here’s how it works (see ChinAI #316 for the previous edition):

I give short previews of ten articles that caught my interest as I was scanning through sources (all published within the past week or so). The title for each preview links to the original article in Chinese.
Readers vote on next week’s feature translation by replying to the email and/or commenting on the post with the number of your preferred article. *Votes from readers who are paid subscribers to ChinAI count a little more.
The main idea is that any of these 10 links would have made for a great feature translation this week — like True Detective season 1 and 3, there are no skips!

1) Can WeCom Cure Companies’ “AI anxiety”?

Summary: Like their American counterparts, Chinese business managers have found that AI tools seem to not generate the promised productivity boosts. Can WeCom — Tencent’s communication and office tool, which connects to WeChat — solve these anxieties?

Source: 雷峰网 (Leiphone) — media portal that covers China’s science and tech landscape, with a focus on AI-related happenings.

2) The Suzhou Woman at Nvidia’s Throat

Summary: A profile of Luo Weiwei, a former NASA scientist, who returned to China to start Innoscience. Her company is the only Chinese supplier of Nvidia’s 800 VDC power architecture.

Source: 量子位 (QbitAI) — news portal that regularly covers AI issues, similar to Leiphone. They also publish longer reports like this one.

3) The Truth of AI Entrepreneurship in 2025

Summary: Based on twelve face-to-face interviews with developers, this piece looks at the unglamorous shortcomings of China’s AI start-up scene. One developer’s AI agent product had gained tens of thousands of users but had fewer than 10 paying subscribers.

Source: 钛媒体创投家 via IT桔子 (IT Juzi) — good source on financing and venture capital in technology fields.

4) Hot Topic: A Mysterious “极" character bug appears in DeepSeek V3.1 . Is the model malfunctioning?

Summary: On August 21, DeepSeek open-sourced its latest base model V3.1. After a week of testing it out, users found a puzzling bug: some of the outputs were randomly replaced with the Chinese character "极" (meaning: extreme). What are the possible causes?

Source: 机器之心 (jiqizhixin) — media portal similar to Leiphone in coverage.

5) Data Labeling Industry Development Research Report

Summary: This report summarizes the current status of China’s data labeling industry. It then analyzes development trends before identifying challenges such as establishing clear standards and training labelers.

Source: 中国信通院 (CAICT) — The China Academy of Information and Communications Technology is a think tank under China’s Ministry of Industry and Information Technology.

6) Building High-Quality Datasets for AI Guidebook

Summary: Another CAICT report! This one looks at the current outlook for building high-quality datasets, with details on key policies and example applications. The section on “establishing an AI data engineering system” [建立AI数据工程体系] intrigued me.

Source: 中国信通院 (CAICT)

7) When the Stars of High-Performance Computing Shine

Summary: A deep dive into the history of China’s high-performance computing industry, traced through the career of one of its key figures — Zheng Weimin. Now, HPC engineers play a prominent role in training large language models.

Source: AI科技评论(aitechtalk) — focuses on in-depth reports on developments in the AI industry and academia.

8) When Scenic Spots use Facial Recognition for Photos: Does it free up tourists or make it easier for others to spy

Summary: The Nandu Personal Information Protection Research Center team conducts some field tests to investigate: As facial recognition diffuses to scenic spot photography services, what risks lie hidden behind this convenience?

Source: 隐私护卫队 — a portal connected to the Nandu Personal Information Protection Research Center.

9) Strategic reading of China’s “AI+” Strategy

Summary: This article provides an in-depth breakdown of the State Council’s recent Action Plan on AI development. The plan outlines a 2035 goal for universal adoption of AI, on the same level as electricity and the internet.

Source: 机器之心 (jiqizhixin) — media portal similar to Leiphone in coverage.

10) Under 40°C heat, a block of ice drifts through the city

Summary: China has experienced an unprecedented heatwave. This longform piece tackles this story from a fascinating angle: an ice factory that produces 100-kg blocks of ice — necessities for vegetable markets, constructions sites, coal mines, and other locations.

Source: 极昼工作室 (Perpetual Light Studio), a reporting unit under Sohu that does longform human-interest stories.

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.

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

Loading more posts…