AI_Chart_Weekly_12_12_Chip_Wars

Oct 12, 2026

Upload by Fahsai

The report argues that computational power is the core US advantage in AI competitiveness, driven by leadership across GPUs and the broader microchip supply chain ecosystem, including software, chip design, and cluster performance. It examines the US decision on Dec 8 to allow limited exports of the Hopper H200 while keeping newer NVIDIA Blackwell B300/B30 under restrictions, and contrasts this with China’s advancing infrastructure and talent pipeline. It also compares NVIDIA and Huawei AI chip capabilities via processing power and memory bandwidth.

LanguageEnglish

CategoryPresentations

FormatPDF

Pages1

AccessFree to download

US computational power and AI competitiveness

US-China metrics for AI advantage

Nvidia vs Huawei AI chip capabilities

Comparison tables and processing power (TPP)

US-aligned semiconductor supply chain advantage

Computational Power is Core US Competitive Advantage Across the numerous metrics for evaluating AI competitiveness, the United States enjoys the largest, order-of-magnitude advantage in AI computational power through its leadership in not only GPUs (i.e., NVIDIA), but the entire microchip supply chain and ecosystem including software, design and cluster performance. Against this backdrop, the US government’s decision on Dec 8th to permit commercial sale to China of the most powerful chip in the older generation Hooper architecture (H200) has been viewed by many as a strategic misstep that has notably diluted US AI competitiveness and national security. The newest generation NVIDIA Blackwell chips (B300, B30) are still subject to US export restrictions. Metrics for evaluating US-China AI competitiveness | | Energy | | | • US leads on energy production, but China moving much more rapidly and efficiently in adding power and grid capacity aligned with AI data center expansion | | --- | --- | --- | --- | --- | | | Computational Power | | | • Most dominant area of US competitive advantage led by new generation of Nvidia Blackwell GPUs (and related system and software architecture)
• Significant US restrictions on export of Blackwell chips to China; older Hopper generation (H20, H200) permitted | | | Infrastructure | | | • US leadership in data centers, software and hardware; China moving rapidly with central planning and local gov’t tax credits and subsidies | | | | | Engineering Talent / Innovation | • US institutions host largest share of world class AI talent, high quality research and labs; more depth and innovation in US private sector AI ecosystem (hyperscalers, start-ups)
• China’s growing pipeline of engineering talent and scale creating significant longer-term competitive advantage | | | | Language Models | | • Significant US leadership in frontier models, innovation and foundational research; China narrowing the gap with several notable cost-efficient breakthroughs (DeepSeek) | | | | | | | | | Data
Business Applications | | | • Significant Chinese structural advantages in data volume and ease of large-scale harvesting; more permissive legal environment; surveillance data; strong consumer apps
• US strong on high quality, globally sourced enterprise data
• US advantages in high-value enterprise AI, but China more entrepreneurial and rapid with integration across a plethora of business applications | Competit ive Advantage Nvidia vs . Huawei AI Chip Capabilities The US Government has explicit restrictions on the sale of both NVIDIA’s most advanced Blackwell generation B300 AI GPU to China, as well as the less powerful NVIDIA B30A GPU (often dubbed “half a B300), which NVIDIA customized with China in mind. However, on Dec 8, 2025, the US Government did approve exports of NVIDIA’s most advanced Hopper generation H200 chip, the most advanced US AI GPU approved for commercial use in China to date. While not nearly as high performing as the B300, NVIDIA’s H200 is generally more capable than Huawei’s most advanced chip, the Ascend 910C, and nearly on par with NVIDIA’s B30A GPU. Nvidia & Huawei AI chip capabilities Total processing power (TPP) 60k 50k 40k 30k 20k 10k 0k | | | | | | | | | | | | Nvidi
(expo | B300
a’s most advanced Brt to China prohibite | lackwell GPU d) | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | B30A* | | | | | | | | | | | | | | | | ( | Nvidia’s proposed China not approved for ex | -specific GPU port to China) | | | | | | | | B200 | | | | | | | | | | | | | | | | | B100 | | | | | | | | | | | | | | | | | | | | | | Ascen Huawei’s m | | | d 910C
ost advanced AI GPU | |
H100 | | | | | | |

FAQs

为什么文中认为计算能力是美国的核心AI竞争优势？

文中指出美国在GPU领导力之外，还在微芯片供应链与生态系统（软件、设计与集群性能）方面占据优势，从而形成数量级的计算能力领先。

Dec 8的政策决定对NVIDIA芯片出口意味着什么？

文中表示美国允许商业销售更老一代Hooper架构的H200到中国，但对更新一代Blackwell芯片（B300、B30）仍受出口限制。

文中如何对比NVIDIA与华为的AI芯片能力？

文中通过总处理能力（TPP）、热设计功耗（TDP）以及内存带宽等指标，对NVIDIA（H20、H200、B30A、B300）与华为Ascend 910C等进行对照。

Oct 12, 2026

Upload by Fahsai