ChinAI #348: China's Compute Year in Review - frenzy, growing pains, and key milestones

Greetings from a world where…

watching best-on-best hockey is euphoric

…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: 2025 China Computing Power Industry Chronicles - Frenzy, Growing Pains, and Value Reversion

Context: For Leiphone, Yilun Liu published a cool year-in-review piece that highlights key moments and trends in China’s computing power industry. This piece recaps over 10 articles Leiphone published in 2025 on the computing power industry.

Key Takeaways: The all-in-one machines that were so hyped at the beginning of 2025 became “things of the past” [明日黄花] after only four months.

• Recall last year, around this time, the flurry of pieces about how China was deploying and diffusing DeepSeek across hospitals, local governments, and the military! Yet, few people were asking: How? The answer was an “all-in-one machine” (see snippet below).

ChinAI #321

• In this week’s feature translation, Leiphone recaps their article “The Truth About DeepSeek All-in-One Machine Deployment”. Here are some of the issues with all-in-one machines as diffusion engines: Some of these units only deployed distilled DeepSeek models; it was difficult to update the software without hardware upgrades; most vendors focused on quick sales and many buyers just wanted the easy PR-win; sustainable adaptation required a lot of organizational and skill investments. “Many companies, after consulting with the company, find they lack the technical capabilities to maintain the system and give up,” the piece reports.

• Liu concludes

Four months ago, all-in-one machines were the darlings of the market, their names plastered everywhere at industry conferences, technology exhibitions, and airport advertisements; four months later, they've become a thing of the past. All-in-one machines occupy a place in many manufacturers' product catalogs, yet actual sales are scarce, even among cloud vendors. The boom has faded, and the market has finally sobered up: the true moat of the all-in-one machine business has never been some technological premium, but rather the customer relationships and channel networks accumulated through hardware expertise.

The frenzy of compute leasing has also faded, as the maturing industry is slowly rooting out fraudulent schemes.

• I covered this briefly with respect to commission rebates:

  ChinAI #339: China's Cloud Ecosystem Enters its (much-needed) Governance Phase

• Liu links to previous Leiphone reporting on “The ‘Hidden Ills’ of Computing Power Leasing.” Here’s one scheme: Cloud providers split computing power projects amongst themselves; one provider might only take on 20% of the business, but they claim 100% of the revenue on their financial statements, just to artificially inflate their stock price. Zhou Yu, a general manager of a financial leasing company, said, “It is quite common for listed companies to hype themselves up as a computing power concept stock, to drive up their stock prices. Many companies never intended to actually develop computing power business from the beginning; they were just using it as an excuse to double their market value.”

Successive IPOs for the “Four Little Dragons” of Chinese GPUs seek to challenge Nvidia in AI inference chips. These four upstarts are: Moore Threads, Muxi, and Illuvatar CoreX [天数智芯].

• These new players are seeking to compete not with Nvidia’s most advanced chips (used to train frontier models) but rather Nvidia’s 4090, which is an AI inference chip used to deploy models.

• On the potential for these Chinese inference chips achieving large-scale commercialization, the article cites the project manager of a compute power leasing company, “Our company will definitely implement domestic computing power projects; we are currently in in-depth discussions with a listed domestic chip company.”

FULL TRANSLATION: 2025 China Computing Power Industry Chronicles - Frenzy, Growing Pains, and Value Reversion

ChinAI Links (Four to Forward)

Must-read: Assetizing, Trading, Franchising: China’s Strategy for Building a National Data Economy

Ran Guo’s Asia Society Policy Institute paper analyzes three important Chinese policy experiments: data assetization, state-owned data exchanges, and public data franchises. Here’s one great section on the first policy:

China’s recent policy recognizing data as a corporate asset, which took effect in January 2024, emerged against this backdrop as an effort to reinvigorate the data market and promote more effective data utilization…Another comprehensive survey found that 199 public firms (2% of all publicly listed firms) reported data assets in 2024, with a total valuation of $309 million. The figures are very low relative to the massive size of China’s digital economy. Moreover, private corporations have demonstrated greater skepticism toward data assetization than state-owned enterprises…As Zhao Zhigang, Director of the Center for Finance and Accounting Research at the Chinese Academy of Fiscal Sciences, succinctly summarized, “Listed companies are cautious, central SOEs are slow to act, local SOEs are more proactive, LGFVs are the most active, while leading internet companies remain hesitant due to data regulatory risks.” This asymmetric pattern has highlighted a deeper problem with data asset capitalization: successful tech businesses such as Baidu, Alibaba, and Tencent that make effective use of data resources often find the government’s administrative red tape too burdensome to navigate.

What I like about this piece is that most research on China’s data governance has focused on cross-border transfers — a topic that is obviously important but should not be all-consuming. Guo’s paper focuses on domestic-facing data policies that could serve as a template for emerging digital powers.

Should-read: Does the UAE have an Advantage in Building Data Centers?

This GovAI publication, by Amelia Michael, has all the ingredients of an excellent report: an important and clearly scoped research question, rigorous methodology, and a finding that challenges conventional wisdom:

“This report’s estimates indicate that the US maintains structural advantages in data center construction, including (perhaps counterintuitively) cheaper energy, a more hospitable natural environment, and a robust domestic data center industry, with its primary disadvantages being higher building construction costs and permitting delays.”

Should-read: Inference Scaling and AI Governance

Two GovAI report recommendations for the price of one in this week’s issue! Toby Ord examines the implications of inference scaling, specifically a scenario where scaling affects computational resources used when serving a request:

Rapid scaling of inference-at-deployment would somewhat lower the importance of open-weight models (and of securing the weights of closed models), reduce the impact of the first human-level models, change the business model for frontier AI, reduce the need for power-intensive data centres, and potentially undermine AI governance measures that rely on training-compute thresholds.

Should-read: Recreating the Smells of History

In Knowable Magazine, Kaja Šeruga tracks how museums, historians, and scientists are trying to integrate scent into their exhibits. The European olfactory heritage project Odeuropa “has created an AI-driven database of more than 2.5 million historical smell references, mined from 43,000 images and 167,000 historical texts published in seven European languages.”

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