Technology + Politics 技术 + 政治 Trump Team Internally Floats Idea of Selling Nvidia H200 Chips to China 特朗普团队内部曾提出向中国出售英伟达 H200 芯⽚的想法 Tap to unmute Advertisement⼴ 告 0:27 Trump Team Internally Floats Selling Nvidia H200 Chips to China 特朗普团队内部讨论向中国出售英伟达 H200 芯⽚。 By Mackenzie Hawkins and Jenny Leonard 作者： ⻨肯⻬·霍⾦斯和珍妮·伦纳德 November 22, 2025 at 2:49 AM GMT+8 Updated on 更新于 November 22, 2025 at 4:47 AM GMT+8 Save 节 省 Listen听 6:40 Translate 翻译 Takeaways by Bloomberg AI 据知情⼈⼠透露，美国官员正在就英伟达公司是否向中国出售其 H200 ⼈⼯智能芯⽚进⾏初步讨论。 N ...

Core Viewpoint - Four individuals have been indicted for attempting to illegally export Nvidia chips valued at millions of dollars to China and Hong Kong, violating U.S. export restrictions [1][2][3] Group 1: Indictment Details - The defendants are charged with conspiracy to violate the Export Control Reform Act of 2018, specifically related to the export of Nvidia chips to China and Hong Kong after routing them through Malaysia and Thailand [2][3] - The indictment highlights that the chips involved, including Nvidia's A100 and H200 GPUs, are highly restricted due to their applications in artificial intelligence and supercomputing [3][4] - The alleged scheme began in September 2023, with the indictment filed on November 13 in U.S. District Court in Tampa, Florida [3][4] Group 2: Individuals Involved - Brian Curtis Raymond, identified as the chief technology officer of an AI cloud company, was involved in the conspiracy and had previously owned a technology products distributor licensed to sell Nvidia GPUs [5][9] - Mathew Ho, another defendant, acted as an intermediary for unlawful exports and submitted false documentation regarding the shipments [6][7] - The other defendants, Jing Chen and Cham Li, were also arrested and are charged with similar offenses, including conspiracy and violations of the Export Control Reform Act [11][12][13] Group 3: Financial Transactions - Raymond faces multiple charges, including seven counts of money laundering related to wire transfers exceeding $3.4 million from a Chinese company to his business [10] - Ho is charged with nine counts of money laundering connected to $4 million in wire transfers from a Chinese company to his and Raymond's businesses [12]

Core Insights - Cambricon Technologies, founded by Chen Tianshi, has seen a significant increase in its market value and revenue, positioning itself as a leading player in the AI chip market in China, often referred to as "China's Nvidia" [1][2] Financial Performance - Cambricon reported a 14-fold increase in quarterly revenue, achieving a net profit of $79.6 million (567 million yuan), a substantial turnaround from a net loss of $27.2 million (194 million yuan) a year ago, marking a 1,332% increase [1] - Following the earnings report, Cambricon's stock surged by 15%, contributing to a $2.4 billion increase in Chen Tianshi's net worth, which now stands at approximately $24.1 billion [2] Market Context - The company's success reflects China's strategic push to develop domestic semiconductor alternatives amid escalating U.S. trade restrictions, particularly the ban on advanced AI chip exports to China [3] - Cambricon's growth is seen as a response to the need for domestic companies to reduce reliance on Nvidia products, creating opportunities for local chipmakers [3][4] Company Background - Cambricon was founded in 2016 as a spinoff from the Chinese Academy of Sciences by Chen Tianshi and his brother Chen Yunji, both of whom have strong academic backgrounds in mathematics and computer science [4] - The company went public on Shanghai's STAR Market in July 2020, with shares increasing by 230% on debut, but it faced seven consecutive years of annual losses until achieving its first quarterly profit in late 2024 [5] Competitive Landscape - Cambricon supplies AI chips to major Chinese tech firms such as Alibaba, Tencent, and Baidu, but faces stiff competition from Huawei, which shipped between 300,000 and 400,000 Ascend AI chips last year compared to Cambricon's over 10,000 units [6] - Analysts project that Cambricon could deliver 80,000 units through the remainder of 2025 and potentially double that in 2026, indicating growth potential in the competitive AI chip market [6]

Core Viewpoint - Meta is set to acquire RISC-V chip startup Rivos to enhance its chip development team and reduce reliance on Nvidia GPU hardware [2][6][12] Group 1: Acquisition Details - The acquisition of Rivos is aimed at accelerating Meta's internal AI chip development, highlighting the company's increasing investment in custom chip design [6][7] - Rivos was recently valued at $2 billion, and the acquisition price is expected to be in the nine to ten-figure range [3] - The deal is not finalized, and the current status of negotiations is unclear [3] Group 2: Strategic Importance - The acquisition signifies a trend among major tech companies towards vertical integration and custom chip development, emphasizing the critical role of dedicated hardware in unleashing advanced AI potential [8][9] - Meta's CEO Mark Zuckerberg has expressed the need to accelerate internal chip development, recognizing the strategic importance of custom chips for AI advancements [7][8] Group 3: Market Impact - The acquisition could trigger a chain reaction in the semiconductor market, potentially challenging the dominance of established players like Nvidia [6][9] - Competitors in social media and AI may accelerate their own custom hardware development in response to Meta's move [9] Group 4: Future Outlook - Post-acquisition, Meta will focus on integrating Rivos' talent and technology into its existing hardware and AI departments, aiming for faster development and deployment of custom AI chips [10][12] - The long-term goal is to significantly reduce capital expenditures related to third-party AI hardware procurement, allowing for more resources to be allocated to AI research and development [10][12]

Group 1: Nvidia's Investment Strategy - Nvidia holds significant positions in two AI stocks: CoreWeave and Arm, with 91% of its $4.3 billion portfolio allocated to CoreWeave and 4% to Arm [1][8] Group 2: CoreWeave Overview - CoreWeave specializes in cloud infrastructure and software services tailored for AI workloads, operating 33 data centers across the U.S. and Europe [3] - The company has a strong relationship with Nvidia, allowing it to launch new chips ahead of competitors, including being the first to offer Nvidia's H100, H200 GPUs, and GB200 superchips [4] Group 3: CoreWeave Financial Performance - CoreWeave's Q2 revenue surged 206% to $1.2 billion, with non-GAAP operating income rising 134% to $200 million, although the non-GAAP net loss widened to $131 million when including interest payments [5][6] - The company is heavily reliant on Microsoft, which contributed 71% of its revenue in the quarter, and anticipates capital expenditures exceeding $20 billion this year [6] Group 4: CoreWeave Valuation and Market Outlook - CoreWeave trades at 12 times sales, with revenue expected to grow at 88% annually through 2027, and stock price targets range from $32 to $180 per share [7] Group 5: Arm Holdings Overview - Arm designs CPU architectures and licenses its intellectual property to companies, capturing 99% market share in smartphones and increasing demand in data centers for AI workloads [8][9] Group 6: Arm Financial Performance - Arm's total sales increased 12% to $1 billion, but it missed sales estimates due to lower licensing and royalty revenue, with non-GAAP net income falling 13% to $0.35 per diluted share [10] - The company expects sales growth to accelerate to about 25% in the current quarter [10] Group 7: Arm's Licensing Strategy - Arm has begun licensing compute subsystems, which has more than doubled its customer base, leading to increased royalty revenue potential [11] Group 8: Arm Market Expectations - Wall Street anticipates Arm's adjusted earnings to grow at 23% annually through March 2027, although its current valuation of 87 times adjusted earnings appears high [12]

Here in this remote northwestern corner of China, is a town at the center of the country's AI ambitions. We are going to go there to see how the construction going and basically get a better understanding of how these data centers fit in with the overall strategy, for China to build its AI capabilities The Xinjiang region is sensitive. China has been accused of human rights abuses against its ethnic Uyghur population.Foreign journalists who go here are monitored. It seems to be a white car following us. I'm ...

Core Insights - The latest MLPerf benchmark results indicate that Nvidia's GPUs continue to dominate the market, particularly in the pre-training of the Llama 3.1 403B large language model, despite AMD's recent advancements [1][2][3] - AMD's Instinct MI325X GPU has shown performance comparable to Nvidia's H200 in popular LLM fine-tuning benchmarks, marking a significant improvement over its predecessor [3][6] - The MLPerf competition includes six benchmarks targeting various machine learning tasks, emphasizing the industry's trend towards larger models and more resource-intensive pre-training processes [1][2] Benchmark Performance - The pre-training task is the most resource-intensive, with the latest iteration using Meta's Llama 3.1 403B, which is over twice the size of GPT-3 and utilizes a four times larger context window [2] - Nvidia's Blackwell GPU achieved the fastest training times across all six benchmarks, with the first large-scale deployment expected to enhance performance further [2][3] - In the LLM fine-tuning benchmark, Nvidia submitted a system with 512 B200 processors, highlighting the importance of efficient GPU interconnectivity for scaling performance [6][9] GPU Utilization and Efficiency - The latest submissions for the pre-training benchmark utilized between 512 and 8,192 GPUs, with performance scaling approaching linearity, achieving 90% of ideal performance [9] - Despite the increased requirements for pre-training benchmarks, the maximum GPU submissions have decreased from over 10,000 in previous rounds, attributed to improvements in GPU technology and interconnect efficiency [12] - Companies are exploring integration of multiple AI accelerators on a single large wafer to minimize network-related losses, as demonstrated by Cerebras [12] Power Consumption - MLPerf also includes power consumption tests, with Lenovo being the only company to submit results this round, indicating a need for more submissions in future tests [13] - The power consumption for fine-tuning LLMs on two Blackwell GPUs was measured at 6.11 gigajoules, equivalent to the energy required for heating a small house in winter [13]

Core Insights - The article discusses the emergence of GPU cloud providers outside of traditional giants like AWS, Microsoft Azure, and Google Cloud, highlighting a significant shift in AI infrastructure [1] - Parasail, founded by Mike Henry and Tim Harris, aims to connect enterprises with GPU computing resources, likening its service to that of a utility company [2] AI and Automation Context - Customers are seeking simplified and scalable solutions for deploying AI models, often overwhelmed by the rapid release of new open-source models [2] - Parasail leverages the growth of AI inference providers and on-demand GPU access, partnering with companies like CoreWeave and Lambda Labs to create a contract-free GPU capacity aggregation [2] Cost Advantages - Parasail claims that companies transitioning from OpenAI or Anthropic can save 15 to 30 times on costs, while savings compared to other open-source providers range from 2 to 5 times [3] - The company offers various Nvidia GPUs, with pricing ranging from $0.65 to $3.25 per hour [3] Deployment Network Challenges - Building a deployment network is complex due to the varying architectures of GPU clouds, which can differ in computation, storage, and networking [5] - Kubernetes can address many challenges, but its implementation varies across GPU clouds, complicating the orchestration process [6] Orchestration and Resilience - Henry emphasizes the importance of a resilient Kubernetes control plane that can manage multiple GPU clouds globally, allowing for efficient workload management [7] - The challenge of matching and optimizing workloads is significant due to the diversity of AI models and GPU configurations [8] Growth and Future Plans - Parasail has seen increasing demand, with its annual recurring revenue (ARR) exceeding seven figures, and plans to expand its team, particularly in engineering roles [8] - The company recognizes a paradox in the market where there is a perceived shortage of GPUs despite available capacity, indicating a need for better optimization and customer connection [9]

Core Viewpoint - The latest MLCommons machine learning benchmark results indicate that Nvidia's new Blackwell GPU architecture outperforms all other computers, while AMD's latest Instinct GPU MI325 competes closely with Nvidia's H200 [1][3][10]. Benchmark Testing - MLPerf has introduced three new benchmark tests to better reflect the rapid advancements in machine learning, bringing the total to 11 server benchmarks [1][11]. - The new benchmarks include two large language models (LLMs), with the Llama2 70B being a mature benchmark and the new "Llama2-70B Interactive" requiring computers to generate at least 25 tokens per second and respond within 450 milliseconds [2][12]. Performance Insights - Nvidia continues to dominate MLPerf benchmarks through submissions from itself and 15 partners, with its Blackwell architecture GPU B200 being the fastest, outperforming the previous Hopper architecture [8][14]. - The B200 GPU features 36% more high-bandwidth memory than the H200 and can perform critical machine learning operations with precision as low as 4 bits, enhancing AI computation speed [8][14]. Comparative Performance - In the Llama3.1 405B benchmark, Supermicro's 8-core B200 system achieved nearly four times the token throughput of Cisco's 8-core H200 system [15]. - The fastest system reported in this round of MLPerf is Nvidia's B200 server, delivering 98,443 tokens per second [15]. AMD's Position - AMD's latest Instinct GPU MI325X is positioned to compete with Nvidia's H200, featuring increased high-bandwidth memory and bandwidth [15][17]. - In Llama2 70B tests, the MI325X system's speed is comparable to the H200, with only a 3% to 7% difference [17]. Intel and Other Competitors - Intel's Xeon 6 chips showed significant performance improvements, achieving about 80% better results compared to previous models, although Intel appears to be stepping back from the AI accelerator chip competition [18]. - Google's TPU v6e chips also performed well, achieving a 2.5 times improvement over their predecessors, although their performance is roughly equivalent to Nvidia's H100 in similar configurations [18].