电子：见证AI变革时刻，“世界最强大芯片”Blackwell问世

Investment Ratings - The report assigns an "Increase" rating to 卓胜微 (Zhuoshengwei) and 江波龙 (Jiangbolong), and a "Buy" rating to 芯原股份 (Xinyuan) [1] Core Insights - The report highlights the emergence of generative AI as a significant milestone in the new industrial revolution, with the introduction of the "world's most powerful chip" Blackwell, which offers AI performance five times that of its predecessor Hopper [2] - The Blackwell architecture GPU B200 integrates 208 billion transistors and provides up to 20 PFLOPS of FP4 performance, significantly enhancing computational capabilities for AI applications [2] - Nvidia's new AI software NIMs allows businesses to deploy custom applications rapidly, reducing deployment time from weeks to minutes, which is expected to attract more customers to Nvidia's ecosystem [2] - The collaboration with Microsoft Azure to launch AI Foundry enables companies to build and deploy custom generative AI models on the cloud, enhancing the accessibility of AI technologies for enterprises [3] Summary by Sections Section: Chip Technology - The Blackwell chip architecture represents a leap in AI processing power, with a 30-fold performance increase in large language model inference workloads compared to previous models [2] - The GB200 super chip, consisting of two B200 GPUs and one Grace GPU, reduces costs and energy consumption by 25% while improving performance [2] Section: AI Services - Nvidia's introduction of enterprise-level generative AI microservices, including NIMs, aims to optimize inference for over 20 popular AI models, streamlining the application development process for businesses [2] - The AI Foundry service allows startups and enterprises using Microsoft Azure to create tailored AI models, enhancing their operational efficiency [3] Section: Robotics and Simulation - Nvidia's Omniverse platform is being integrated with Apple's Vision Pro, showcasing advancements in simulation technology for climate prediction and robotics [4] - The introduction of the Project GR00T humanoid robot model and upgrades to the Isaac Robotics platform indicate Nvidia's commitment to advancing robotics technology [4]