前言 一边是中国大幅放开稀土出口，7月对美出口量从46吨猛增至619吨，展现十足合作诚意。 另一边却是美国驻华使馆公然给中国贴上"敌对国家"标签，在舆论、经济、教育三条线同时发力，完全一副翻脸不认人的架势。 这种"拿货就翻脸"的操作打的什么算盘？中美稀土较量将走向哪里？ 作者-山 75%暴增的稀土刚到手，美国就连下三道狠手 数据不会撒谎。7月中国稀土磁铁出口达到5577吨，比6月份足足增长了75%。 这个数字背后，是中国对美国释放的实实在在的善意。从5月的46吨到6月的353吨，再到7月的619吨，对美稀土出口呈现火箭般的增长轨迹。 要知道，稀土被称为"工业维生素"，美国80%的稀土需求依赖进口，这些"泥土"般不起眼的金属，却是F-35战机、爱国者导弹、特斯拉电动车都离不开的关 键原料。 按理说，解了燃眉之急的美国应该心怀感激才对。 然而现实却让人大跌眼镜。 更过分的是，美方还无端指责中国向美国输出"非法电子烟产品"，甚至暗示中国涉及什么"基因工程改造"，这种毫无根据的造谣抹黑，已经突破了正常外交 的底线。 舆论战只是开胃菜，真正的重拳还在后面。 从舆论抹黑到教育围堵，再到经济制裁，美国这三记重拳招招致命，时 ...

前言 中国主动释放善意，7月稀土出口暴增75%，用实际行动为中美关系降温。 美国的回应却让人大跌眼镜，先是使馆公然叫嚣"敌对国家"，接着制裁企业，最后连留学生都不放过。 拿到稀土就翻脸，美国这波操作暴露了什么？中美关系走到了哪一步？ 75%暴增送稀土，换来三记闷棍 7月一个数字震惊了华盛顿。 5577吨稀土磁铁涌向美国，75%的暴增幅度创下年内新高，这是中国海关总署8月20日公布的官方数据。 这个数字意味着什么？要知道稀土就像工业"味精"，看着不起眼，却是每个高科技产品都离不开的核心材料。从F-35战斗机到特斯拉电动车，从iPhone到风 力发电机，全都要用到中国稀土。 中国掌握着全球90%的稀土加工能力，这可不是谁想替代就能替代的。 更关键的是，这批稀土的主要买家德国、美国和越南中，很多运到越南的产品只是"过个手"，最终还是流向了美国企业。 从5月到7月，中国对美稀土出口呈现阶梯式增长：46吨、353吨、619吨，每个月都在加码。 这清楚说明，美国对中国稀土的依赖已经深深扎根在全球供应链的每一个环节里。中国这次是拿出了十足的诚意。 但这份诚意，华盛顿似乎并不买账。 就在稀土数据公布前后，美国对中国连续出了三 ...

Core Viewpoint - The article discusses the controversy surrounding NVIDIA's H20 AI chip, which is perceived to potentially contain backdoor features that could compromise security, particularly in the Chinese market. The CEO, Jensen Huang, faces the challenge of proving the chip's safety while navigating complex geopolitical tensions between the U.S. and China [1][4][6]. Group 1: Chip Performance and Market Impact - The H20 chip's performance is only about 20% of NVIDIA's flagship H100, leading to it being referred to as a "crippled version," yet it is still allowed to enter the Chinese market [6][19]. - China contributes significantly to NVIDIA's revenue, accounting for 22% of its global income, which amounts to $17.1 billion, with a growth rate of 66% [6][19]. - Following the controversy, NVIDIA's revenue in China plummeted by 42%, while orders for Huawei's Ascend chips surged by 300% [19]. Group 2: Regulatory and Geopolitical Tensions - The Chinese government raised concerns about the H20 chip potentially having tracking and remote shutdown capabilities, demanding technical proof from NVIDIA within 48 hours [6][11]. - U.S. legislation requires regulated chips to include location verification and remote shutdown features, raising suspicions about the H20 chip's design [9][11]. - Jensen Huang's visit to Washington resulted in a deal where NVIDIA committed to a $50 billion investment in the U.S. in exchange for expedited export approval for the H20 chip [17]. Group 3: Trust and Market Dynamics - Chinese companies, including Baidu and iFlytek, have halted H20 chip purchases and are exploring domestic alternatives, indicating a rapid shift in market dynamics [18][19]. - The share of domestic chips in AI computing procurement is projected to rise from 5% in 2022 to 40% by 2025, while NVIDIA's market share in China is expected to drop from 95% to 50% [21]. - The article emphasizes that trust is the most critical factor in the chip market, suggesting that even high-performance chips cannot restore confidence once it is lost [22].

Core Viewpoint - The recent surge in the semiconductor sector, particularly with companies like Cambrian and Shanghai Hualing, is driven by the anticipation surrounding the release of DeepSeek-R2, marking a significant moment for domestic computing power in China [1][3]. Group 1: Market Dynamics - Cambrian and Shanghai Hualing experienced a 20% increase in stock prices, indicating strong market interest [1][3]. - The release window for DeepSeek-R2 (August 15-30) has created a speculative environment, with investors betting on its potential impact [3]. - Daily Interactive, which previously surged due to DeepSeek, saw a 12% increase, suggesting that smart money is positioning itself ahead of the expected announcement [3]. Group 2: Technological Advancements - DeepSeek-R2 boasts 1.2 trillion parameters, with a dynamic activation of only 78 billion, outperforming GPT-4 by 20% in inference speed while reducing costs to $0.07 per million characters, just 2.7% of GPT-4's cost [3]. - The training of DeepSeek-R2 is based on Huawei's Ascend 910B chip, achieving 91% of the performance of Nvidia's A100 with an 82% utilization rate, indicating a significant advancement in domestic technology [3]. - The model supports multi-modal integration, achieving a 98.1% accuracy rate in medical diagnostics, surpassing top-tier doctors [3]. Group 3: Risks and Challenges - The release of DeepSeek-R2 has faced delays, originally scheduled for May, due to performance issues and supply chain disruptions, which could affect market sentiment if further postponed [4]. - The recent launch of GPT-5 poses a competitive threat, showcasing superior capabilities in various applications, necessitating DeepSeek-R2 to demonstrate its effectiveness through real-world results [5]. Group 4: Investment Opportunities - Key players in the semiconductor core circle include Cambrian, which collaborates with the Ascend ecosystem, and Shanghai Hualing, a critical supplier of materials for Ascend chips [7]. - Companies like Tuowei Information and Digital China are positioned to benefit from increased orders driven by the Ascend server market, with Digital China expected to see an 80% increase in orders by 2025 [8]. - Technology empowerment firms such as Softcom Power and Guangxun Technology are integrating R2 technology into their platforms, enhancing product quality and positioning themselves strategically in the computing power landscape [9]. Group 5: Strategic Recommendations - Investors are advised to focus on leading stocks like Cambrian and Tuowei Information if R2 is released on schedule, while considering a shift to more stable alternatives like SMIC and Changdian Technology if delays occur [10]. - Long-term strategies should include monitoring the commercial feedback of R2 in industrial and medical applications, as well as tracking server shipment volumes from Ascend partners to confirm market health [11].

Core Viewpoint - The article discusses the supply-demand imbalance of NVIDIA's H20 AI chips in the Chinese market following the lifting of export restrictions, highlighting the significant demand from Chinese tech companies and the challenges in meeting this demand due to geopolitical factors and supply chain constraints [2][3][4]. Supply and Demand Dynamics - After the lifting of export restrictions, there is a strong demand for H20 AI chips in China, with NVIDIA placing an order for 300,000 units from TSMC due to this demand [2]. - Jefferies estimates that the total demand for H20 chips in the Chinese market is around 1.8 million units, while NVIDIA's inventory is projected to be only between 600,000 to 900,000 units by early 2025, indicating a significant supply gap [2]. - Chinese tech giants have already placed pre-orders worth over $12 billion for H20 chips to secure future supply [3]. Geopolitical Influence - The H20 chip is not a standard product but rather a result of geopolitical influences, particularly U.S. export control policies that have fluctuated, impacting its availability in China [4]. - The U.S. government updated its export control list in April 2025, leading to a temporary halt in sales to China, which resulted in NVIDIA writing down inventory values by $4.5 billion to $5.5 billion in its Q1 2026 financial report [6]. Production and Supply Chain Challenges - NVIDIA's CEO announced in July 2025 that the company received U.S. government approval to resume shipments of H20 chips to China, but the supply chain is unable to respond immediately due to prior adjustments [7]. - TSMC faces challenges in fulfilling NVIDIA's orders, with a potential nine-month timeline required to restart the supply chain for H20 chips, which may not meet the immediate needs of Chinese customers [10][11]. Market Positioning and Competitiveness - The H20 chip is positioned to provide a legal and compatible high-performance computing solution for Chinese tech companies, especially in AI inference and large model training, amidst restrictions on advanced AI chips from the U.S. [5]. - Despite the H20's limitations in performance compared to the H100, its specifications, such as 96GB HBM3 memory and 4TB/s bandwidth, make it suitable for many applications in the Chinese market [12][13]. Challenges for Domestic Alternatives - Domestic AI chips struggle to replace the H20 due to the extensive reliance on NVIDIA's CUDA platform, which requires significant code rewriting and adaptation for alternative frameworks [12]. - The pricing strategy of the H20 chip is designed to create a competitive edge, being priced between $6,500 and $8,000, which is strategically positioned above the government procurement price of Huawei's Ascend 910B [14]. - The current state of domestic AI chips is at a critical point, requiring ongoing technological innovation and policy support to achieve broader market acceptance [15].

Core Viewpoint - The article discusses the potential risks associated with foreign chips, particularly the H20 chip promoted by NVIDIA's CEO Jensen Huang, in the context of China's national security concerns and the implications for the domestic chip industry [1][3][5]. Group 1: Risks of Foreign Chips - The Chinese Ministry of National Security warns that foreign chips may contain "backdoors" that can compromise data security and privacy [1][4]. - Three types of backdoors are identified: 1. "Factory-installed malware" that can activate devices remotely [3]. 2. "Maintenance channels turned espionage tools" that can be exploited by hackers [4]. 3. "Supply chain poisoning" where malicious code is introduced during software updates [4]. Group 2: H20 Chip Analysis - The H20 chip, marketed as a solution for China's AI needs, has only 15% of the performance of the H100 chip and does not support training trillion-parameter models [5]. - The U.S. strategy appears to be to sell lower-performance chips to maintain market presence in China while preventing the transfer of core technology [5]. Group 3: Domestic Chip Development - Domestic chips are improving but still lag behind foreign counterparts like the H20 in performance [6]. - Companies like Huawei are innovating with techniques such as "four-chip packaging" to enhance performance while reducing costs [6]. - The article emphasizes the importance of developing a robust domestic ecosystem to compete with established foreign technologies like NVIDIA's CUDA [6][7]. Group 4: Future Strategies - A dual approach is suggested: using domestic chips for sensitive applications while gradually integrating foreign technology where necessary [6]. - Long-term goals include fostering an open-source framework for domestic chips and investing in AI talent to reduce reliance on foreign technology [7].

Core Viewpoint - The recent approval of the H20 chip for sale in China marks a significant shift in U.S. chip policy, reflecting the complex interplay between political interests and commercial benefits in the ongoing U.S.-China tech rivalry [1][3]. Group 1: Policy Changes - In April 2025, the U.S. imposed a ban on the H20 chip, leading to a $5.5 billion inventory write-down for Nvidia and a 6% drop in its stock price [3]. - By July 15, 2025, the policy was reversed, allowing the H20 chip to enter the Chinese market, resulting in a surge in related A-share stocks, with Zhongji Xuchuang rising by 16.58% in a single day [3][5]. Group 2: Strategic Motivations - The reversal of the U.S. policy is driven by three main pressures: technological competition, capital market impacts, and political calculations [5][7]. - The technological gap between U.S. and Chinese chips is narrowing, with Huawei's Ascend 910B chip achieving 148 TFLOPS, only slightly behind the H20's 160 TFLOPS [5]. - The capital impact of the ban led to a $155 billion loss in Nvidia's market value and a 30% drop in ASML's lithography machine orders, prompting the semiconductor industry to lobby the White House [5]. Group 3: Commercial Dynamics - Nvidia's CEO, Jensen Huang, indicated a deal with the U.S. government involving a $500 billion investment in AI data centers over four years in exchange for H20 export approval, highlighting the tension between U.S. efforts to contain China and the need for commercial gains [7]. - The H20 chip's technical limitations, such as only achieving 28% of the FP16 performance of the H100 and using less efficient interconnect technology, suggest a strategic attempt to limit China's technological advancement [8]. Group 4: Market Implications - Despite the H20's limitations, demand in China's medical and educational sectors remains high due to the dominance of the CUDA ecosystem, which 90% of global AI developers rely on [10]. - The dependency on the H20 chip poses long-term risks for China's tech independence, as local alternatives like Huawei's Ascend series and Cambricon's MLU370 are still developing their software ecosystems [10][11]. Group 5: Future Outlook - The U.S. may continue to implement "dynamic controls," allowing the export of older generation products while China must focus on breaking the CUDA monopoly, accelerating chiplet technology commercialization, and building a self-sufficient semiconductor equipment system [11]. - Huang's statement about China accounting for 20% of Nvidia's revenue underscores the ultimate logic of this geopolitical struggle: commercial interests may eventually override political barriers [11].

Core Viewpoint - The U.S. government has approved the sale of H20 chips to China, which are specifically designed with limited performance to create a dependency on American technology, raising questions about the underlying motives behind this decision [1][3]. Group 1: H20 Chip Characteristics and Implications - The H20 chip, developed by NVIDIA for the Chinese market, is an AI accelerator with intentionally restricted capabilities, unable to support large model training, thus limiting its use to mid- to low-computational scenarios [3]. - This design choice is not due to technical limitations but is a strategic move by the U.S. government to keep China's AI industry reliant on American technology [3]. - The recent lifting of the export ban on H20 chips is part of a broader geopolitical strategy, where the U.S. aims to exchange technology for critical resources from China [3]. Group 2: China's Response and Future Directions - In the short term, the availability of H20 chips may alleviate China's computational power shortages, with major tech companies like Alibaba and Tencent considering procurement [5]. - However, long-term reliance on these "customized" chips could undermine the motivation for independent R&D among Chinese AI firms, potentially placing them at a disadvantage in technological competition [5]. - China is accelerating the development of domestic alternatives, such as Huawei's Ascend 910B chip, which matches the performance of H20 while offering better energy efficiency [5]. - The Chinese government is promoting initiatives for AI technology self-sufficiency, including the "Large Model + Industry" project and the "Domestic Replacement Alliance" [5]. Group 3: Historical Context and Strategic Outlook - Historical precedents indicate that excessive reliance on foreign technology can be detrimental, as seen in Japan's semiconductor industry in the 20th century [7]. - In contrast, China is pursuing a path of self-innovation to break free from technological constraints, leveraging its position as the largest semiconductor consumer market to reshape the technological landscape [7]. - The ongoing U.S.-China technological rivalry is complex, with no clear resolution in sight, but China's efforts are aimed at gaining more leverage for the future [7].

Core Insights - Nvidia is set to launch a new AI chip for the Chinese market, priced significantly lower than the recently restricted H20 model, with production expected to start as early as June [1] - The new GPU will belong to Nvidia's latest Blackwell architecture series, with a price range of $6,500 to $8,000, compared to the H20's previous price of $12,000 to $15,000 [1] - The new chip will utilize traditional GDDR7 memory instead of advanced high-bandwidth memory (HBM) and will not employ TSMC's advanced semiconductor packaging technology [1] Group 1: Market Impact and Regulations - New export restrictions have introduced bandwidth constraints on GPU memory, limiting it to 1.7-1.8 TB/s, while the H20 could achieve 4 TB/s [3] - The new GPU is expected to meet the export control limit with a bandwidth of approximately 1.7 TB/s [3] - Nvidia's CEO criticized the U.S. government's AI chip diffusion control regulations, stating that they hinder the company's ability to adapt to market needs [4] Group 2: Financial Implications - The H20 ban has forced Nvidia to write down $5.5 billion in inventory, potentially leading to a revenue loss of $14 billion to $18 billion for the year [5] - Nvidia's revenue from China reached $17.108 billion in the fiscal year ending January 2024, marking a 66% increase from the previous year [5] - The company's market share in China has dropped from 95% in 2022 to 50% currently, with Huawei emerging as a key competitor [4]

Core Viewpoint - Nvidia is launching a new AI chip for the Chinese market, significantly reducing the price compared to the previously restricted H20 model, as a response to U.S. export controls and competition from Huawei [2][3]. Group 1: New Product Launch - Nvidia plans to introduce a new AI chip priced between $6,500 and $8,000, which is a substantial decrease from the H20's price range of $10,000 to $12,000 [2]. - The new chip is based on the Blackwell architecture and utilizes the RTX Pro 6000D server-grade processor along with conventional GDDR7 memory [2]. - Production for the new chip is expected to start in June, following the ban of the H20 model [2]. Group 2: Market Impact and Competition - Following the ban of the H20, Nvidia's market share in China plummeted from 95% to 50%, with Huawei's Ascend 910B chip rapidly gaining market presence [2]. - Nvidia's CEO Jensen Huang warned that continued U.S. restrictions could lead to more Chinese customers shifting to Huawei [2]. Group 3: Financial Implications - The discontinuation of the H20 resulted in Nvidia recording a $5.5 billion inventory loss and forfeiting $15 billion in potential orders [3]. - Nvidia is also planning to mass-produce another Blackwell architecture chip, potentially named B40, aimed at the Chinese market in September [3]. Group 4: Regulatory Challenges - Nvidia is awaiting final approval from the U.S. government for the new product designs, as the company navigates compliance with U.S. export regulations [3]. - Industry analysis indicates that the new U.S. regulations aim to suppress China's AI computing power by limiting memory bandwidth, while Nvidia seeks to find market opportunities by adjusting chip configurations [3].