● China, AI, Shock, Race
Immediate Analysis of the U.S.-China AI Rivalry: How Far Has China Advanced Along the U.S. AI and Semiconductor Value Chain?
The key issue is not simply that “China’s AI has advanced significantly.”
The more important point is that U.S. restrictions have accelerated China’s AI self-sufficiency and domestic semiconductor localization.
This trend connects AI semiconductors, data center investment, power infrastructure, physical AI, robotics, and the outlook for the KOSPI in one framework.
For Korean semiconductor companies such as Samsung Electronics and SK hynix, the critical task is to distinguish whether China’s progress represents a competitive threat or a source of new demand.
1. The “Sanctions Paradox” Created by U.S. AI Restrictions
The United States has imposed strong restrictions on China’s access to advanced semiconductors.
Examples include limits on Nvidia’s high-end AI chip exports, controls on advanced equipment, and restrictions on semiconductor supply chains.
The original objective was to slow China’s AI development.
In practice, however, the outcome has been different.
- China has increased investment in domestic AI model development.
- China is advancing semiconductor localization and equipment self-sufficiency as a national strategy.
- Chinese technology firms and research institutions are optimizing innovation under constrained computing resources.
- Models such as DeepSeek and Kimi K3 are creating pricing pressure across the U.S.-centered AI market.
This is the so-called sanctions paradox frequently cited in global economic analysis.
Instead of slowing progress, restrictions have accelerated internal innovation.
2. China’s AI Model Benchmarks: DeepSeek R1 and Kimi K3
DeepSeek R1 is one of the most widely cited examples of China’s AI progress.
It is estimated to have been trained at roughly one-tenth the cost and time of leading U.S. models.
In coding and certain reasoning tasks, it has been assessed as broadly competitive with U.S. models.
Kimi K3 is another notable case.
It has been described as an open-weight model with an exceptionally large parameter scale.
Based on source reports, approximately 280 billion parameters were mentioned, reinforcing China’s visibility in open AI competition.
The key point is not that China has fully overtaken the United States.
Rather, the performance gap between U.S. and Chinese AI models is narrowing rapidly.
Market commentary has even characterized the gap as roughly three months of technology lead.
3. Five Criteria for Assessing the U.S.-China AI Rivalry
AI competition cannot be assessed by chatbot performance alone.
True AI leadership depends on data, models, computing power, power infrastructure, and research capacity.
Viewed through these five criteria, China’s progress becomes more visible.
① Computing Power: The U.S. Still Holds a Clear Lead
In AI, computing power is the foundation of competitiveness.
Training and operating advanced models requires large-scale GPUs, data centers, and cloud infrastructure.
By data center count alone, the United States remains dominant.
- Total data centers worldwide: approximately 12,255
- Data centers in the United States: approximately 4,767
- Data centers in China: approximately 376
Even if China ranks second globally, the gap with the United States remains substantial.
In other words, the United States currently maintains a clear advantage in AI infrastructure and computing capacity.
On this basis alone, China is unlikely to catch up in the near term.
② Power Infrastructure: A Surprisingly Strong Card for China
However, the next major bottleneck for data center expansion is not only GPUs.
It is power infrastructure.
AI data centers require very large amounts of electricity.
Even if more data centers are built, expansion is limited if power supply is insufficient.
U.S. power generation capacity is stable, but growth is relatively slow.
By contrast, China’s power generation capacity has expanded rapidly, and some analyses suggest it now exceeds the U.S. by more than two times.
The gap is also widening.
This is a critical point.
The United States leads in data center count, but may face power bottlenecks as it expands further.
China, by contrast, has greater room in power supply, giving it a stronger foundation for long-term data center expansion.
China also has an advantage in low-end semiconductors and power infrastructure.
Grid systems, renewable energy, batteries, transmission networks, and power equipment do not always require leading-edge AI chips.
China’s strength in general-purpose and industrial semiconductors can therefore support large-scale deployment.
This implies that China’s AI semiconductor strategy is not only about directly competing with Nvidia.
China is building domestic control from the lower and adjacent layers of AI infrastructure first.
③ AI Models: U.S. Leadership Remains, but China Has a Strong Cost Advantage
In model performance, the United States still leads.
OpenAI, Google, Anthropic, and Meta continue to set the benchmark for the global AI market.
China’s main advantage lies in cost.
Chinese firms have been actively using distillation techniques.
Distillation transfers knowledge from larger models to smaller ones by leveraging training outputs and reasoning patterns.
The advantages are clear.
- Training costs can be reduced materially.
- Inference costs can be lowered.
- Efficient deployment is possible even in GPU-constrained environments.
- For enterprise users, the price-performance ratio can be attractive.
Enterprises do not focus only on maximum performance.
They also evaluate cost-effectiveness.
Profit is revenue minus cost.
As a result, Chinese AI models can influence the market even if they do not deliver absolute top-tier performance.
This also affects the profitability of AI semiconductor and cloud infrastructure companies.
④ Data: The Hidden Advantage of China’s Super-App Ecosystem
Data is the core fuel for AI.
From this perspective, China has a different type of strength from the United States.
In the United States, major applications are relatively specialized.
Facebook and Instagram are centered on social networking.
Amazon is centered on e-commerce.
Google is centered on search and advertising.
In contrast, China’s super-app ecosystem, including WeChat and Alipay, integrates daily life into a single platform.
Payments, shopping, messaging, reservations, transportation, finance, and public services are all connected.
This creates deeper data density.
For example, purchasing a banana milk drink at a convenience store is only a simple transaction record.
But if the time of purchase, accompanying items, payment method, movement patterns, and broader consumption behavior are all connected, the data becomes far more valuable.
China’s super-app ecosystem is structurally well positioned to collect and connect this type of multi-layered data.
This is a significant asset for AI service development.
⑤ Research Capability: China’s Fundamental AI Research Has Advanced Rapidly
AI competition should not be judged only by current products.
Long-term competitiveness comes from papers, patents, researcher depth, and basic R&D capability.
In the past, U.S. AI research dominated global citation rankings.
However, in the 2020s, China’s AI citation share has reportedly surpassed that of the United States.
According to the source text, China’s global AI paper citation share is around 21%, and it has exceeded the United States since 2019.
China’s scale advantage in patents and publications is also expanding.
This suggests that China is moving beyond a simple imitation phase.
It is building long-term competitiveness in foundational research as well.
4. The Market Impact of Falling AI Model Prices
The spread of Chinese AI models is putting pressure on AI service pricing.
As open-source and low-cost models expand, token prices tend to decline.
According to the source text, after Chinese model launches, a token price index fell roughly 40% from its peak and declined to around $1.20.
Token pricing is one of the core revenue drivers for AI companies.
Lower prices benefit users, but they can weaken profitability for AI providers.
This is one reason recent earnings expectations for hyperscalers have softened.
The market is asking whether:
- AI companies can continue to sustain large-scale data center investment,
- profitability can hold if token prices keep falling,
- hyperscalers will continue purchasing Nvidia GPUs and HBM at scale, and
- the AI semiconductor supercycle can remain intact.
These questions are influencing semiconductor valuations and the outlook for the KOSPI.
5. Model Mixing and the New AI Cost Structure
Companies are unlikely to rely on a single AI model going forward.
The use of model mixing, combining high-performance and lower-cost models, is likely to expand.
For example, GPT, Claude, or Gemini may be used for complex strategic analysis, while lower-cost models may handle customer service or internal document processing.
This approach can materially reduce total operating costs.
According to the source text, model mixing could reduce task costs by as much as 90%.
This changes the economics of the AI industry.
The market is moving from an era of universal use of premium models to one of task-specific model selection.
6. AI Semiconductor Demand Will Not Necessarily Decline
This is where many investors may make a mistaken assumption.
They may think that falling AI model prices automatically mean weaker demand for AI semiconductors.
That is not necessarily the case.
This is where Jevons paradox becomes relevant.
Jevons paradox suggests that as technology becomes more efficient and unit costs decline, total usage may increase rather than decrease.
Just as lower electricity prices can lead to higher electricity consumption, lower AI token prices can lead to broader AI adoption.
This can sharply increase overall AI usage.
In other words, lower token prices may pressure near-term profitability for AI companies, but they can also expand AI usage over time.
And as usage rises, demand for computing power and AI semiconductors may increase again.
7. The Next Demand Wave May Come from Physical AI and Robotics
The next major source of AI demand may not be chatbots.
It is more likely to come from physical AI.
Physical AI refers to AI operating in the real world.
This includes robotics, autonomous driving, smart factories, drones, humanoids, and industrial automation.
China is also strong in this area.
It already has a large ecosystem in advanced manufacturing, robotics, batteries, electric vehicles, bio-manufacturing, and industrial automation.
Companies such as Unitree are emblematic of China’s physical AI momentum.
Lower AI model prices can also reduce the cost of embedding AI into robots and manufacturing environments.
This could accelerate the adoption of physical AI.
If so, demand could rise simultaneously for AI semiconductors, sensors, memory, power semiconductors, and communication modules.
8. The U.S. and China AI Ecosystems Are Likely to Split
The global AI market is likely to become less integrated and more divided into U.S.-centered and China-centered ecosystems.
The United States has the Nvidia CUDA ecosystem, hyperscaler cloud platforms, and model ecosystems led by OpenAI, Google, and Anthropic.
China is building a separate ecosystem around Huawei’s CANN, Alibaba’s platform base, domestic AI models, and local standards.
China is also expanding international cooperation frameworks such as the Shanghai AI Cooperation Organization.
The objective is to expand the reach of Chinese AI standards to other countries.
This is not simply a technology competition.
It is a geoeconomic supply chain fragmentation process.
Going forward, some countries may use U.S. AI systems while others adopt Chinese AI systems.
9. Implications for Korean Semiconductors and the KOSPI
For Korea, this should not be viewed emotionally.
The issue is not alignment with the United States or China, but where opportunities emerge within the value chain.
Samsung Electronics and SK hynix are directly linked to U.S. AI infrastructure investment.
HBM, high-performance DRAM, NAND, and advanced packaging are key areas of demand.
If hyperscaler data center investment in the United States slows, it may weigh on short-term share prices.
However, if AI usage expands in line with Jevons paradox, semiconductor demand could strengthen again over the medium to long term.
Physical AI and robotics could also increase demand for memory, sensors, and power semiconductors.
Chinese companies such as CXMT may create long-term competitive pressure for Korean memory makers.
This is especially relevant in commodity DRAM and lower-end memory markets.
However, Korean firms still maintain a strong technological edge in HBM and advanced memory products.
In the outlook for the KOSPI, the key issue is not whether the semiconductor cycle is over.
The more important question is where demand is shifting.
- Existing demand: cloud data centers and generative AI
- Emerging demand: physical AI, robotics, autonomous driving, and smart factories
- Bottleneck variable: power infrastructure and the pace of data center expansion
- Pricing variable: token price declines and AI company profitability
- Competitive variable: China’s semiconductor value chain localization
10. The Most Important Point Often Missing in Other Coverage
First, the next decisive area in the AI rivalry may be power infrastructure rather than GPUs.
Even if companies want to build more data centers, expansion is constrained without sufficient electricity.
On this front, China appears to have a meaningful structural advantage.
Second, the real threat from Chinese AI is not necessarily top-end performance, but cost disruption.
Even if performance is slightly below the U.S. benchmark, significantly lower prices can still lead enterprises to adopt Chinese models.
This puts pressure on the profitability of global AI providers.
Third, falling token prices may not signal a collapse in semiconductor demand, but rather the start of demand expansion.
As in Jevons paradox, lower AI usage costs can prompt broader adoption across industries and users.
That could ultimately increase demand for AI semiconductors.
Fourth, China’s super-app ecosystem has an advantage in data density.
The key is not simply user scale, but the integration of payment, lifestyle, mobility, and consumption data within a single platform.
This data structure is highly favorable for AI service development.
Fifth, Korea should not respond by making an emotional choice between the United States and China.
Korea should evaluate U.S. AI infrastructure value chains, China’s manufacturing AI expansion, and physical AI demand simultaneously.
A strategy spanning semiconductors, robotics, power equipment, batteries, and smart factories is necessary.
< Summary >
The U.S.-China AI rivalry is shifting from clear U.S. dominance to rapid Chinese convergence.
The United States still leads in computing power and data centers.
China is strengthening its position in power infrastructure, cost-efficient AI models, super-app data, and foundational research.
DeepSeek R1 and Kimi K3 are emblematic of how U.S. restrictions have accelerated China’s AI self-sufficiency.
Lower AI model prices pressure global AI profitability, but may also expand long-term usage through Jevons paradox.
The next demand wave is likely to come from physical AI and robotics.
For Korea, the key is to assess not only the AI semiconductor cycle centered on Samsung Electronics and SK hynix, but also power infrastructure, robotics, and China’s progress in semiconductor localization.
[Related Articles…]
- AI Semiconductor Supercycle and Data Center Investment Trends
- Physical AI and the Next Growth Cycle in Robotics
*Source: [ 경제 읽어주는 남자(김광석TV) ]
– 미중 AI 패권전쟁 : 중국이 미국의 AI-반도체 벨류체인을 추격하고 있다 [즉시분석]
● China Semiconductor Crisis, Rare Earth Shakeup, Korea Wins
China’s Semiconductor Stress and the Restructuring of Rare Earth Supply Chains: The Real Opportunity for Korean Manufacturing
The key point is not simply that “Chinese semiconductors are failing.”
The more important issue is that China has poured substantial subsidies into the sector without yet resolving its core bottlenecks.
At the same time, Korea’s effort to diversify rare earth supply chains is intersecting with U.S.-China technology competition, causing another shift in the global supply chain structure.
This is particularly relevant because AI semiconductors, memory chips, rare earth materials, manufacturing competitiveness, and supply chain security are now closely linked. For Korean companies and investors, the implications are material.
1. Core Takeaway: The Real Nature of China’s Semiconductor Stress
The central claim in the source material is that China’s semiconductor industry is under significant strain.
In particular, ongoing losses at ChangXin Memory Technologies (CXMT), a Chinese memory semiconductor company, were highlighted.
While the framing is sharp, the economic interpretation is that the cost of China’s semiconductor self-sufficiency drive is proving far higher than expected.
China views semiconductors not as a conventional industry but as a strategic sector tied to national security.
As a result, it continues to deploy government subsidies, policy financing, and local government support even when profitability remains weak.
The “throwing water into a bottomless pit” description reflects this capital-intensive structure.
From China’s perspective, semiconductors must be produced within domestic supply chains even at a loss.
That is because higher external dependence increases vulnerability to U.S. export controls and equipment restrictions.
Accordingly, China is prioritizing self-reliance over economic efficiency in its semiconductor strategy.
2. Why China Is Struggling: A Sector That Cannot Be Solved by Capital Alone
Semiconductors are not built simply by constructing fabs and importing equipment.
Yield, design capability, equipment access, material quality, process know-how, and customer trust must all align.
This is why China has not been able to close the gap quickly despite large-scale capital investment.
First, access to advanced equipment is restricted.
The United States and its allies have tightened export controls to limit China’s advanced chipmaking capabilities.
Extreme ultraviolet lithography equipment is effectively inaccessible to China, and even deep ultraviolet tools face increasing export constraints.
Second, memory semiconductors require scale and accumulated expertise.
Established leaders such as Samsung Electronics, SK hynix, and Micron have built process technology and manufacturing experience over decades.
China faces not only a technology gap but also challenges in production stability and customer qualification.
Third, the gap may widen in the AI semiconductor era.
As AI servers and data centers expand, high-bandwidth memory, high-performance DRAM, and advanced packaging are becoming more important.
Even if China narrows the gap in standard memory products, major bottlenecks remain across the AI semiconductor value chain.
Fourth, semiconductors depend on materials and components as much as equipment.
Photoresists, specialty gases, wafers, CMP slurries, and packaging materials all require stable quality.
China is trying to raise self-sufficiency, but in advanced process nodes, material quality and long-term reliability remain decisive.
3. The Meaning of CXMT Losses: Not Failure, but a Signal of Rising Costs
CXMT’s losses should not be interpreted as evidence that China’s semiconductor industry is collapsing.
Rather, they indicate that the cost of achieving semiconductor self-sufficiency is extremely high.
Memory semiconductors are highly cyclical.
Margins can be strong during price upcycles, but losses can widen quickly when supply exceeds demand.
Chinese firms must absorb both this cycle and the cost of technology catch-up and equipment constraints.
Ultimately, the issue is not that China lacks money, but that many parts of the industry still require time even after the money is spent.
Government support can build fabs, but it cannot create global customer confidence overnight.
4. Why China Cannot Give Up Semiconductors
China’s commitment to semiconductors is straightforward.
Semiconductors are essential to smartphones, electric vehicles, data centers, defense technology, AI, robotics, and communications equipment.
Heavy dependence on foreign suppliers would leave future industries exposed to external shocks.
As U.S.-China technology competition intensifies, China can no longer rely on the assumption that it can simply buy what it needs abroad.
This is due to tighter export controls on advanced AI chips, semiconductor equipment, EDA software, and high-performance computing technologies.
As a result, China must continue producing within its own supply chain even at a loss.
That is the core meaning of the statement that supply must be secured within its domestic ecosystem.
5. Rare Earths: A Strategic Lever That May Lose Strength
The source also refers to Korea’s efforts to reduce reliance on Chinese rare earths.
Rare earths are essential to electric vehicle motors, wind power, defense, semiconductor equipment, displays, batteries, and robotics.
For many years, China has maintained dominant influence over rare earth mining, refining, and processing.
The key issue is not the rarity of the ores themselves.
The real bottleneck lies in refining, separation, and processing after extraction.
These steps are environmentally intensive and technically demanding, which allowed China to dominate the supply chain for a long period.
For Korea to replace Chinese rare earths, sourcing ore from another country is not enough.
Alternative material development, recycling technology, refining capacity, non-China supply chains, and long-term procurement contracts are all required.
The emphasis on substitute materials and new processes that can deliver comparable performance reflects this reality.
6. What Changes If Korea Secures Rare Earth Supply Chains
If Korea diversifies its rare earth supply chains, stability should improve first for electric vehicles, batteries, and semiconductor equipment.
This would reduce exposure to export restrictions and price spikes from a single country.
Second, Korean manufacturing competitiveness could strengthen.
Korea already has global manufacturing capabilities in semiconductors, batteries, shipbuilding, automobiles, displays, and specialty chemicals.
Adding more stable access to core materials would improve the bargaining power of the entire manufacturing value chain.
Third, global companies may view Korea as a more reliable production partner.
Global supply chains are increasingly favoring locations that can deliver stable output over those that are merely low-cost.
If Korea reduces materials and components risk, its attractiveness as a manufacturing base could rise.
7. Korea’s Manufacturing Strength: A Different Competitive Model from China
Korea is not a resource-rich country, but it has strong manufacturing execution.
It has established global competitiveness in industries that require advanced process control, including semiconductors, batteries, automobiles, shipbuilding, and displays.
China’s advantages are scale, a large domestic market, and policy support.
Korea’s strengths are precision manufacturing, quality control, global customer responsiveness, and technical reliability.
In semiconductors, production experience and yield management are especially important.
As AI chip demand rises, the strategic value of Korean companies with strong memory and advanced packaging capabilities may increase.
8. Semiconductors and Rare Earths Through the AI Lens
This issue should be understood not as a narrow China semiconductor story, but as part of infrastructure competition in the AI era.
As AI models scale, demand rises across data centers, GPUs, HBM, power infrastructure, cooling systems, and rare earth materials.
AI is not only a software story.
It also depends on chips, servers, power grids, materials, equipment, and manufacturing capacity.
Accordingly, semiconductors and rare earth supply chains are likely to remain central variables in the global economy.
China will need advanced semiconductors to expand its AI capabilities.
However, as U.S. restrictions make access to high-performance AI chips more difficult, China may be forced to increase spending on domestic chip development.
That could keep losses elevated in the near term.
9. The Most Important Point That Is Often Missed
The key issue is not whether China’s semiconductor industry collapses immediately.
The real point is that China is likely to accept low or negative profitability for a prolonged period in order to build domestic capacity.
This could have significant implications for the global semiconductor market.
If Chinese firms continue to expand standard chip output with government backing, pricing pressure may emerge in some product categories.
At the same time, in advanced semiconductors, U.S. restrictions are likely to limit China’s pace of catch-up.
As a result, the semiconductor market may become more bifurcated.
Standard products may face greater price pressure from Chinese supply growth, while high-performance AI chips and HBM may continue to command a technology premium led by Korea, the United States, and Taiwan.
Rare earths are another critical point.
They have long been a strategic lever for China, but if Korea and other major economies build alternative supply chains, that leverage weakens.
This shift has significance far beyond near-term headlines.
10. Key Watchpoints for Korean Companies and Investors
First, monitor HBM and AI memory demand.
If AI data center investment continues to expand, demand for high-performance memory should rise structurally.
This would further strengthen the strategic position of Korean semiconductor companies.
Second, watch for China’s expansion in standard chips.
If China increases production with state support, competition could intensify in some memory and system semiconductor segments.
For Korean firms, a focus on high-value-added products will be more important than volume competition.
Third, pay attention to rare earth and critical mineral supply chains.
Electric vehicles, robotics, defense, wind power, and semiconductor equipment are all sensitive to rare earth risk.
Projects outside China, recycling technologies, and substitute-material companies may draw more attention over the medium to long term.
Fourth, manufacturing competitiveness may regain a premium.
As the global economy shifts from efficiency to resilience, countries and companies with trusted manufacturing bases are likely to gain value.
Korea is well positioned in this environment.
Fifth, U.S.-China technology competition should be viewed as a long-duration cycle.
China’s semiconductor sector is unlikely to collapse quickly; instead, it is likely to continue a long-term catch-up strategy supported by capital.
That process suggests ongoing supply chain restructuring.
11. Likely Scenarios Ahead
Scenario 1: China expands standard chip supply
China may continue increasing production to serve domestic demand.
In that case, global pricing pressure could rise in some categories.
Scenario 2: Advanced semiconductor gaps remain
Leadership in AI chips, HBM, advanced foundry services, and advanced packaging is likely to remain with current leaders.
The barriers created by equipment and process know-how remain very high.
Scenario 3: Rare earth supply chain diversification accelerates
Korea, the United States, Japan, and Europe are likely to keep reducing dependence on China.
Australia, Canada, Vietnam, India, and parts of Africa may receive greater attention as alternative supply sources.
Scenario 4: Korea’s strategic value in manufacturing rises
Korea could benefit from supply chain realignment across semiconductors, batteries, electric vehicles, defense, shipbuilding, and robotics.
However, this will require reducing dependence on imported materials and components.
12. One-Sentence Conclusion
China’s semiconductor stress is not simply a corporate earnings issue; it is a structural outcome of the intersection between U.S.-China technology competition and global supply chain reorganization.
If Korea strengthens rare earth diversification and advances its AI semiconductor capabilities, it may secure a more important position in the global manufacturing order.
< Summary >
China’s semiconductor industry continues to face challenges in advanced equipment access, yield, materials, and customer trust despite heavy state support.
CXMT’s losses suggest that the cost of China’s self-sufficiency strategy is rising.
Because China cannot give up semiconductors, it is likely to continue absorbing losses in order to build domestic supply chains.
Korea’s rare earth diversification efforts can reduce dependence on China and improve manufacturing resilience.
AI semiconductors, HBM, advanced packaging, and rare earth materials are likely to remain central to global industrial competition.
The real issue is not a sudden collapse in Chinese semiconductors, but a market split in which standard chips face Chinese supply expansion while advanced chips retain a technology premium centered on Korea, the United States, and Taiwan.
[Related Articles…]
China Semiconductor Risk and the Restructuring of Global Supply Chains
Rare Earth Supply Chain Shifts and Korea’s Manufacturing Competitiveness
*Source: [ 달란트투자 ]
– 사상 최악의 대참사 터졌다 쫄딱 망해버린 중국 반도체 | 조현승 박사 3부
