国泰海通证券研究·2025-08-31 13:59Core Insights - Nvidia's Q2 earnings exceeded expectations, with revenue reaching $46.7 billion, a 56% year-over-year increase, and data center revenue at $41.1 billion, also up 56% year-over-year and 5% quarter-over-quarter. The company plans to invest significantly in AI infrastructure, projecting data center infrastructure spending to reach $3-4 trillion by 2030 [2][3] Group 1: Nvidia - Nvidia's Q2 2026 revenue was $46.7 billion, marking a 56% year-over-year growth, with data center revenue at $41.1 billion, also up 56% year-over-year and 5% quarter-over-quarter [2] - The company reported double-digit growth in its networking segment, with Spectrum-X Ethernet achieving annualized revenue exceeding $10 billion [2] - Nvidia's GB300 is now in mass production, with a capacity of 1,000 racks per week, indicating strong demand for its products [2] Group 2: Microsoft - Microsoft launched two self-developed models: MAI-Voice-1 for voice generation and MAI-1-preview, a mixture of experts model, enhancing the capabilities of its Copilot ecosystem [3] - MAI-Voice-1 is designed for high-fidelity audio content generation and is integrated into Copilot Daily and podcast features [3] - The MAI-1-preview model utilizes 15,000 Nvidia H100 GPUs, showcasing Microsoft's commitment to leveraging advanced hardware for AI development [3] Group 3: xAI - xAI introduced Grok Code Fast 1, a new programming model emphasizing speed and affordability, available for free on major coding platforms [4] - The model is built from scratch with a focus on programming-related content, optimizing for real-world coding tasks [4] - Pricing for Grok Code Fast 1 is competitive, with costs set at $0.20 per million input tokens and $1.50 per million output tokens, appealing to small and medium developers [4]