Entropy Increase vs. Entropy Decrease: Reading China's AI Strategy Through Yu Xiaohui's Qiushi Essay
> This post, suggested by a friend, analyzes China's AI policy thinking through the lens of thermodynamic entropy.
On May 16, 2026, *Qiushi* journal (Issue 10, 2026) published a long-form article by Yu Xiaohui, president of the China Academy of Information and Communications Technology (CAICT), titled *"Accurately Grasping the Frontiers of AI Development and the Competitive Landscape."* As a senior figure in China's AI policy establishment, Yu's article offers a window into how Chinese decision-makers view this technology revolution at mid-year.
This post interprets the article through the physics metaphor of entropy increase (disorder) and entropy decrease (ordered structure), contrasting the US and Chinese AI development models.
Key points
1. The US model as "entropy increase": when compute devours order
- Bloomberg reported in May 2026 that surging data center power demand in the Lake Tahoe region has caused blackouts for local residents.
- Symptoms cited: unchecked compute expansion outpacing grid capacity, competition for scarce electricity between data centers and residents, and absence of national-level coordination. Data Center Watch reported a striking number of US data center projects shelved in Q2 2025 due to power shortages.
- Full-stack synergy: "AI competitiveness is formed through the overall synergy of 'algorithms–frameworks–chips–systems'; ecosystem stickiness does not depend on a single product." This echoes the NVIDIA CUDA ecosystem lock-in. China's response: systematically push coordinated integration across the full stack, countering the fragmentation of each chip vendor maintaining its own software stack.
- Efficiency over scale: mainstream model inference token output costs fell ~99% over three years. DeepSeek's engineering-driven open-source approach matched top-tier models at low cost; Huawei Ascend achieved deep adaptation with DeepSeek-V4; DeepSeek's open-source downloads rank among the world's highest.
- AI as public infrastructure, not a rent-seeking asset: Yu proposes concentrating 10,000-GPU-class clusters in hub nodes, coordinating compute with power and network planning, raising the share of green and nuclear energy, building compute interconnection networks (extending the "East Data West Computing" project), and enabling SMEs to access high-quality compute easily — alongside strengthening open-source mechanisms ("technology open-source + ecosystem co-building").
- Quality inspection: vision large models sharply cut miss rates vs. manual sampling, enabling end-to-end online inspection.
- Process optimization: models output optimal process parameters, lifting yields beyond what human experience achieves.
- Equipment maintenance: predictive maintenance models significantly reduce unplanned downtime.
- Common thread: AI converts engineers' decades of tacit, fragmented experience into reusable, iterable explicit knowledge — another form of "entropy decrease."
- US "disorder" bred radical innovation, rich application exploration, and distributed energy development.
- China's "order" risks suppressing grassroots innovation, path dependency, and efficiency/rent-seeking challenges in public investment.
- Yu himself is candid: "strong engineering deployment capability does not equal strong original innovation capability"; China's models still lag world-class ones in complex reasoning and tool use, and original frontier innovation remains in a catch-up phase.
2. Yu's three core judgments on AI competition
3. Industrial intelligence: from "showrooms" to scale
AI-empowered manufacturing transitions described in the article:4. Global governance: inclusive sharing vs. technological blockade
Yu advocates "inclusive, AI-for-good" principles: sharing AI achievements with Global South countries, promoting open-source general models, local expansion with inclusive empowerment, implementing the AI international cooperation initiative, and building China–BRICS and China–ASEAN AI cooperation centers. Gallup/YouGov surveys showing polarized US public sentiment toward AI are cited as evidence that monopoly-driven technology erodes social trust.5. The dialectic of entropy: disorder incubates order
The author cautions that neither model is purely good or bad:Conclusion
Yu's essay is less a policy declaration than a roadmap for steering AI development from disorderly expansion toward ordered evolution. The real question is not "who wins," but "whose model is more likely to make this technology a net benefit for society as a whole." The implied answer: with inclusiveness as the goal, synergy as the means, and governance as the guarantee, AI should shift from an instrument of the few to a precondition for the many.
References
1. Yu Xiaohui (2026). *Accurately Grasping the Frontiers of AI Development and the Competitive Landscape*. Qiushi, Issue 10, 2026. https://www.qstheory.cn/20260515/a9132f0ae2814f3c88eb44300722d4f8/c.html 2. Bloomberg (2026). Lake Tahoe power crisis report (May 13, 2026). 3. Data Center Watch (2025). Q2 2025 data center project suspension statistics. 4. Grand View Research (2025). *Generative AI In Content Creation Market Size Report, 2030*. https://www.grandviewresearch.com/industry-analysis/generative-ai-content-creation-market-report 5. Gallup / YouGov (2026). US public attitudes toward AI surveys.