On July 23, 2025, the U.S.-China battle for technological supremacy entered a new phase. The Trump administration has launched an ambitious ” U.S. AI Action Plan “. This plan will be a powerful engine to further accelerate U.S. giants such as OpenAI, Google, and Nvidia. But can it really “curb” China’s onslaught?
In this paper, we will read the full text of this AI Action Plan and analyze China’s surprising strategy in response to it. In conclusion, this plan will “accelerate” the U.S. AI industry to unprecedented levels. However, it has fundamental limitations and blind spots in “containing” China’s rise. This is because while the U.S. strategy bets on overwhelming hardware “scale,” China has already begun to achieve remarkable results on the asymmetrical battlefield of algorithmic “efficiency.”
This is not a mere technological competition. It is the beginning of a grand battle for supremacy in the age of AI, where two different philosophies, the “scale” of the United States and the “optimization” of China, collide.
Chapter 1: U.S. Full-Acceleration Strategy: The Goal of the “AI Action Plan
The essence of this plan is not to attack China directly. Rather, it is a domestic industrial policy designed to unleash the dynamism of the private sector, which is the greatest strength of the United States, and to push it out so fast that no other country can keep up. The strategy consists of three main pillars.
1. unlocking innovation: promoting deregulation and openness
The first pillar of the plan is the “elimination of bureaucratic formalism and cumbersome regulations” that have stymied AI development . It includes forcing federal agencies to repeal rules that impede AI and removing guardrails introduced during the Biden administration, DEI (diversity, equity, and inclusion), and climate-related requirements from the CHIPS Act.
At the same time, the plan strongly encourages an open source and open-weighted AI model . This is to prevent innovation from being monopolized by a few giant tech companies and to allow small businesses and academia to freely participate in innovation.
More notably, the provision counters the liberal bias that conservatives decry as “Woke AI” . It mandates that the Large Language Models (LLMs) procured by the federal government be “objective” and protect “free speech.” This is in direct response to the domestic culture wars over AI values.
2. building the heart of AI: concentrated investment in infrastructure and energy
The second pillar is to dramatically accelerate the construction of the physical infrastructure that will power AI: data centers and semiconductor factories. The plan names and condemns “radical climate dogma” and makes clear its commitment to relaxing environmental regulations that could be an obstacle to construction.
The plan also confronts head-on the reality of power shortages brought on by the explosive growth of AI. By upgrading the U.S. power grid and accelerating the deployment of new power sources such as enhanced geothermal and next-generation nuclear energy, the plan seeks to meet the enormous energy demands of AI supercomputing.
3. making u.s. ai a global standard: “full stack export” concept
The third pillar is to deploy U.S. technological leadership around the world. The centerpiece is an ambitious initiative to provide allies with a “secure full-stack AI export package” that includes hardware, models, software, and standards . This is a strategic move to bring countries around the world into the U.S. technology ecosystem and prevent dependence on China.
However, this approach of pushing “U.S. dominance” to the fore risks creating friction with allies. If the message of “buy U.S. products” is too strong rather than cooperation, it may in fact lead to the defection of allies.
Chapter 2: China’s Asymmetric Strategy: The Wisdom of the Dragon to Turn Constraints into “Innovation
While U.S. plans pursue “scale” and “speed,” China is playing a very different game. Faced with the severe constraints of U.S. export restrictions, China is using them against it to build its own robust AI ecosystem.
1. algorithmic betting: DeepSeek shows that “efficiency” is a weapon
China has been cut off from access to cutting-edge Nvidia GPUs due to U.S. export restrictions. However, companies like DeepSeek have used this adversity as a springboard to find a way to innovate their algorithms and architecture . Their strategy is not to invest computational resources by brute force, but to pursue “efficiency” to extract higher performance with fewer resources.
DeepSeek’s model delivers performance comparable to top U.S. models at a fraction of the announced cost. The secret lies in a sparse architecture called Mixture of Experts (MoE) . This is a technique that activates only a small fraction of the model’s parameters depending on the task, dramatically reducing computational complexity. Combined with a proprietary technology that improves memory efficiency, this enables high performance even on older chips subject to export restrictions.
This success proves that advances in AI do not necessarily depend on massive computing power. This is a powerful challenge from China to the U.S. belief in “scaling laws” .
2. building “good enough” homegrown technology: the reality of Huawei and SMIC
On the hardware side, China is also steadily moving toward self-sufficiency: Huawei’s Ascend series chips (910B, 910C) are the most promising domestic alternatives to Nvidia GPUs . While they are inferior to Nvidia in terms of individual chip performance, they can be optimized as a whole system to provide surprisingly competitive performance for certain tasks .
SMIC, a major Chinese semiconductor company, manufactures these chips. Despite being cut off from access to state-of-the-art manufacturing equipment (EUV), SMIC has managed to produce 7nm process chips using older equipment . This comes at a significant cost in terms of low yields and high costs , but it is still significant that they have established a system that can supply “good enough” AI chips domestically.
China is accelerating a national effort to become self-sufficient not only in chips, but in the entire semiconductor ecosystem, including design tools (EDA) , high-bandwidth memory (HBM) essential for AI , and the open source RISC-V architecture .
Behind this is the Chinese government’s strong industrial policy of “self-reliance and self-reinforcement” . A massive state fund called the “Great Fund” supports R&D and a “domestic substitution” policy creates a protected market for domestically produced technologies .
Chapter 3: The Limits of the Cudgel: Are U.S. Export Controls Really Working?
A pillar of the U.S. strategy toward China was the “cudgel” of export controls. These restrictions have certainly been effective in preventing China from mass producing cutting-edge 3nm and 5nm chips . SMIC’s 7nm chip production is severely constrained in its scale by US regulations .
This tactical victory, however, created an unintended strategic counterproduct. It was the U.S. regulations that forced Chinese companies to switch to domestic suppliers and, as a result, became the biggest catalyst in dramatically accelerating China’s efforts to become self-sufficient.
In addition, the policy has caused billions of dollars in economic losses to Nvidia and other U.S. companies and there are loopholes in the regulations, such as smuggling and through third countries .
The most important change is that export controls have changed the dynamics of competition itself. The main battleground of competition shifted from “access” to hardware to “efficiency” in how efficiently it is used. The U.S. now faces stronger rivals who have been trained by adversity and have mastered the art of asymmetric warfare.
Final evaluation: Accelerators cannot be brakes.
So let’s return to the core question. Can the “U.S. AI Action Plan” curb China’s drive to catch up?
In conclusion, this plan will be immensely effective in accelerating the U.S. AI industry, but will do little to curb Chinese progress.
Why will the acceleration be successful? The plan directly addresses eliminating the biggest bottlenecks for U.S. AI companies: data centers, semiconductor plants, and energy. This will ensure that Nvidia and OpenAI have the computing power they need to develop next-generation models and further widen the gap with China . Deregulation will allow companies like Google and Meta to innovate more quickly .
Why do they fail to restrain? The biggest blind spot in the plan is its adherence to the “symmetrical competition” of winning on hardware scale. This completely overlooks the “asymmetric battlefield” of algorithmic efficiency, which China has already axed . The plan will help the U.S. build a bigger hammer, but it will not work against an enemy that has learned how to fight without a hammer.
The plan is an “accelerator” for the US, not a “brake” for China.
Where are the decisive battlefields? The future of the U.S.-China AI competition will depend on which of the two opposing hypotheses is correct.
- The U.S. bet: that the “scaling rule” will continue. Progress in AI is determined by computational power and quantity of data. If this hypothesis is correct, the U.S., with its superior quantity, will win.
- The Chinese bet: that “algorithmic shortcuts” exist. Smart algorithms can override hardware differences and enable leaps in performance. If this hypothesis is correct, China can remain competitive under U.S. regulations.
In reality, the future will likely be a hybrid of both. While hardware scale will continue to be important in cutting-edge research, algorithmic efficiency may be the winner in the broad economic dissemination of AI.
Final Outlook: “Barbellization” of the Global AI Market This AI Action Plan is guaranteed to make the U.S. race run faster, but will do little to slow down China, which is running in the next lane. As a result, the global AI market could be polarized by the ” barbell effect.”
At one pole, the United States dominates the high-end, computationally intensive frontier (AGI research, supermodels). At the other pole, China dominates the low-cost, efficient, and widely deployable “good enough” AI applications market.
And the economic value of the latter may be greater than the former in the long run. The U.S. AI Action Plan may be so confident of its own victory that it is unintentionally encouraging a future in which China is victorious on an entirely different battlefield.
