China’s October 16, 2026 decision to designate AI model weights and autonomous agent technologies as strategic assets has opened a new chapter in the global technology wars. The digital barrier rising against US chip embargoes raises the question: is a Great Wall being built around AI?
Let me say upfront what I'll say at the end: information and technology cannot be locked behind walls. History has proven this to us time and again. Neither the ancient empires of the past nor today's superpowers have been able to stop the borderless flow of knowledge. Tomorrow's digital barriers won't manage it either.
 |
| Digital transformation of the Great Wall: AI-powered preservation and smart tourism |
The AI Great Wall: Model Weights and Agent Technologies Now Subject to Export Controls
China's Ministry of Commerce (MOFCOM) and Ministry of Science and Technology published a draft of the "Catalogue of Technologies Prohibited or Restricted from Export" on October 16, 2026. With this move, Beijing officially designated AI's most powerful "brains" (model weights) and self-operating autonomous assistants (agent technologies) as strategic assets. Strict restrictions were placed on the outflow of advanced AI technologies abroad; digital intelligence was placed under the shield of national security and economic sovereignty.
This move is a direct response from Beijing on the software and algorithm front to US hardware embargoes. In the global technology race, it's no longer just chips — software code and digital independence have also become national security boundaries.
What Does the Decision Mean?
Advanced models are being placed under state protection: The core technology of models with at least 2 trillion parameters is entering the scope of national security.
There is a purpose to protect open-source power: The aim is to prevent models like Alibaba's (Qwen) or DeepSeek's from being copied by rivals and used against them.
Autonomous agents are being kept within borders: Capabilities that can write software and conduct cyber operations on their own are being brought under control.
It's not a total ban — it's a licensing requirement: Use via cloud (API) is not blocked; downloading the model's "brain" is not permitted.
The US Embargo Process
China's move is a response on the software front to US hardware restrictions. The process unfolded as follows:
May 2019: The first major infrastructure and software restrictions targeting Chinese technology manufacturers.
October 7, 2022: Exports of advanced AI chips and manufacturing equipment were banned.
October 17, 2023: Loopholes were closed; Nvidia's alternative chips like the A800 and H800 were also blocked.
October 2024 and beyond: Restrictions were expanded; some advanced chips were also subject to licensing requirements for third countries.
These bans affected global chip companies differently. Nvidia and AMD continued to grow thanks to Western AI investments despite losses in China. ASML, whose advanced lithography system sales to China were blocked, experienced periodic selling pressure due to revenue loss in the critical market and weakening expectations.
 |
| US chip embargo: The hardware power behind the AI tech war |
Strategic Asset Decision: What Does Mutual Blockade Produce?
The Chinese side defends this move on two main grounds:
First, economic protectionism. It is seen as a shield preventing the free copying of billions of dollars' worth of national AI labor and strategic data. I agree with this assessment; the protectionist reflex was inevitable. However, I don't think these walls will be long-lasting.
Second, national security. Models with 2 trillion parameters and the cyber/military capacity of autonomous agents require critical code to remain within the country. After the US moves, China's counter-move was expected; it came. Mutual moves will continue in the short and medium term, but military-technological blockades won't be permanent in the long run either.
The US and Western side read this as an aggressive digital barrier. The difficulty of accessing ready-made model weights will push developers toward alternative infrastructures. Some Washington-based assessments count the blow to the open-source world as a step that slows global innovation. China harmed collective R&D with its latest decision; but it didn't open this path alone. The US's previous decisions are also part of the same spiral.
The Limits of Information and the Risk of Uncontrollability
Mutual embargoes harm the open-source ecosystem, which draws its strength from free sharing. Restricting models from leaving borders narrows independent developers' optimization space; it makes academic use harder and shifts the market toward closed, paid systems.
At the heart of this struggle is not just security, but also economic share and global competition. As digital walls rise, it becomes less visible which country is developing what cyber or military capabilities. As transparency decreases, oversight weakens too. Code hidden behind walls can carry greater risks without collective oversight.
Model Weights: The Rare Earth Elements of the Digital World
The global race is no longer just about chips and computing power. Model weights, source code, and autonomous agent systems are also being brought under control, like rare earth elements or advanced semiconductors.
The draft proposes adding 40 new technologies to the catalogue; 23 of them are in the AI field. Public comments will be collected until November 15. However, the temporary pre-registration requirement has been in effect since October 16. While the debate continues, actual control has already begun.
China classifies digital assets to be transferred into three tiers: export-impossible, license-required, and free circulation. The main items caught in the oversight radar are model weights, training checkpoints, algorithm code, architectural schematics, autonomous agents, high-powered computing clusters, and domestic chip design information.
Which Models Are in the Crosshairs?
A model can fall under restriction if it meets any of these three thresholds:
Extremely high computing power spent during training
At least 2 trillion parameters
Ranking in the global top 10 in performance tests
Projects trained on China's own national data pools are subject to an even stricter red line. These thresholds largely overlap with the US's own security criteria. Beijing is turning Washington's technical language into an asymmetric defense tool.
In practice, DeepSeek's newer versions, Alibaba's Qwen series, and models like Moonshot (Kimi) and Tencent (Hunyuan) are among the natural subjects of this framework.
The Most Critical Distinction: API Free, Weights Restricted
The draft's most strategic point is here:
API, application, or web access is not counted as export
Installing the model on a foreign server or transferring weights is counted as export
Technologies previously published as open source are out of scope
With this formula, China keeps control of model files without completely losing its global user base. Using DeepSeek or Qwen via cloud is possible; taking the weights and running them on local infrastructure requires a license. This is a search for balance that both protects the technology from being copied and sustains revenue and data flow through the API. In the long run, the winners will be those who can control their own infrastructure and models; the losers may be part of the open-source ecosystem based on free circulation.
US–China Symmetry
While the US restricts advanced chips and production capacity, China is bringing model weights and agent technologies under oversight. In the same period, US restrictions on some Chinese-origin AI products are escalating tensions.
The emerging picture is a transition to digital bipolarity. In the short term, the hardware advantage appears to be with the US, but China's algorithmic capacity — producing efficient models with limited hardware — could challenge the balance. This is not a permanent peace; it is a spiral in which the parties target each other's weak points.
Long-Term Consequences
Model weights are becoming as strategic as chips.
The practical space of open source is narrowing.
Third countries face licensing and political choice pressure when building their own sovereign clouds.
Competition is shifting from "who builds the better model?" to "who controls their model better?"
In the coming years, the global ecosystem may move toward a more fragmented picture. Actors wanting to process their own data locally may be caught between remaining dependent on Western-centric systems and requesting a license from China. AI risks becoming a geopolitical tool whose valve can be turned down. But for how long?
The Concealment Efforts of Technologies That Changed History
History has repeatedly shown that information cannot be locked behind walls.
Ancient China–Silk: The production secret was protected for centuries by the death penalty. The monopoly collapsed in 552 AD when silkworm eggs were smuggled out.
1440–Printing Press: Gutenberg's circle kept the technique like a trade secret; it spread across Europe through war and migration.
1789–Steam Engine: England tried to block plans and engineer emigration; the knowledge crossed continents anyway.
1796–Smallpox Vaccine: Jenner chose to share the formula rather than commercialize it; the knowledge crossed borders.
1876–Telephone / 1880s–Alternating Current: Despite patent and propaganda wars, the technology became a global standard.
1943–Penicillin: Kept as a military secret during the war; afterward it became humanity's shared remedy.
1969–Internet: Born as a closed military network; over time it evolved into an open, global infrastructure.
2026–Large AI Models: Against US chip restrictions, China placed model weights under protection. But the spread of openly weighted models is again pulling AI toward a decentralized development space.
AI Will Be Humanity's Shared Heritage
Restricting information is a futile effort in the long run. History shows us that neither China, which tried to hide the secret of silk in its palaces, nor England, which wanted to keep the steam engine on its island, could stop the fluid nature of knowledge. Silkworm eggs were smuggled inside bamboo canes; plans were etched into memories and crossed continents. Alternating current lit up the world, and the internet, born as a military network, became the shared arena of all humanity.
Today's digital barriers or chip blockades drawn around AI model weights may serve as short-term geopolitical shields, but in the long run they will yield to the spreading power of knowledge. It is not the protectionist approaches that hide technology behind walls that win — it is those who continuously produce, develop, and open it to the world.
Moreover, the world is not just the US and China. Other countries are already beginning to build their own AI capabilities to break free from dependence in the face of these restrictions.
There is not a single factor behind these decisions, and it is clear that over time these restrictions will loosen and disappear. No technology can remain secret forever and eventually spreads to the whole world. For this reason, I believe these decisions will be short-term and will ultimately invalidate themselves. AI, like the printing press, electricity, or the internet, will sooner or later fit into no one's monopoly and will become humanity's shared heritage.
Information can no longer be hidden the way China concealed its silk production secret for 3,000 years. We are no longer in the era of the printing press, nor of steam engines, nor of patented telephones. Everything will happen at the speed of AI.
Sources:
China's Ministry of Commerce (MOFCOM) & Ministry of Science and Technology: Catalogue of Technologies Prohibited or Restricted from Export (Draft Text Open for Comment), Announcement No. 48 (October 16, 2026)
Global Media and Analysis Reports: Reuters, Bloomberg, The Washington Post, Global Times