【中泰电子】AI全视角-科技大厂财报专题英伟达Q2解读：Q2业绩新高，数据中心拉动增长

Investment Rating - The report does not explicitly mention an investment rating for the industry or the specific company [1][2][3] Core Views - NVIDIA's FY25Q2 revenue reached $30 04 billion, a 122% year-over-year (yoy) increase and 15% quarter-over-quarter (qoq) growth, surpassing market expectations of $28 6 billion by 4 9% [2][3] - Data center revenue was the primary driver, contributing $26 27 billion, a 154% yoy increase and 16% qoq growth, exceeding market expectations by 5 7% [2][3] - The company announced an additional $50 billion stock repurchase plan, with $15 4 billion returned to shareholders in the first half of FY25 through buybacks and dividends [3] - NVIDIA expects FY25Q3 revenue to be $32 5 billion, with a GAAP gross margin of 74 4% [3] Financial Performance FY25Q2 Financial Highlights - Revenue: $30 04 billion, yoy +122%, qoq +15%, exceeding market expectations by 4 9% [2][3] - Data center revenue: $26 27 billion, yoy +154%, qoq +16%, accounting for 87% of total revenue [2][3] - Gaming revenue: $2 88 billion, yoy +16%, qoq +9%, accounting for 10% of total revenue [2][3] - Professional visualization revenue: $450 million, yoy +20%, qoq +6%, accounting for 2% of total revenue [2][3] - Automotive revenue: $350 million, yoy +37%, qoq +5%, accounting for 1% of total revenue [2][3] - GAAP net income: $16 6 billion, yoy +168%, qoq +12%, exceeding market expectations [2] - GAAP gross margin: 75 1%, yoy +5 0%, qoq -3 3%, in line with market expectations [2][3] - Capital expenditures: $980 million, yoy +238%, qoq +165%, significantly exceeding expectations [16] Regional Revenue Breakdown - US revenue: $13 billion, yoy +115%, qoq -4%, accounting for 43% of total revenue [14][15] - Mainland China (including Hong Kong) revenue: $3 7 billion, yoy +34%, qoq +47%, accounting for 12% of total revenue [14][15] - Taiwan revenue: $5 7 billion, yoy +102%, qoq +31%, accounting for 19% of total revenue [14][15] - Other regions revenue: $7 6 billion, yoy +304%, qoq +34%, accounting for 25% of total revenue [14][15] Business Segments Data Center - Data center revenue reached $26 27 billion, driven by strong demand for the Hopper GPU computing platform [3][20] - Compute revenue within data center: $22 6 billion, yoy +162%, qoq +17%, accounting for 86% of data center revenue [22][23] - Networking revenue within data center: $3 67 billion, yoy +114%, qoq +16%, driven by InfiniBand and AI Ethernet growth [22][23] Gaming - Gaming revenue: $2 88 billion, yoy +16%, qoq +9%, driven by GeForce RTX 40 series GPUs and gaming console SoCs [25][26] Professional Visualization - Professional visualization revenue: $450 million, yoy +20%, qoq +6%, driven by RTX GPU demand [28] Automotive - Automotive revenue: $350 million, yoy +37%, qoq +5%, driven by the NVIDIA DRIVE platform [29] Business Developments Blackwell GPU - Blackwell GPU samples were delivered to customers in FY25Q2, with production ramp-up expected to start in Q4 and continue through FY2026 [3][32] - Blackwell is expected to contribute billions in revenue starting in FY25Q4 [3][32] - Blackwell offers 4x faster training and 30x faster inference speeds compared to the H100, with 25x lower TCO and energy consumption [35] NVIDIA NIM and CUDA-X Microservices - NVIDIA NIM inference microservices and CUDA-X microservices help developers quickly build and deploy AI and accelerated applications [50] - Companies like Adobe, Cadence, CrowdStrike, and SAP are among the first to adopt these microservices [50] AI and Accelerated Computing in Various Industries - NVIDIA AI and accelerated computing are being expanded in industries such as healthcare, industrial digitization, and EDA/CAE [51] - NVIDIA cuLitho accelerates computational lithography by 40-60x, with production already underway at TSMC and Synopsys [51] Automotive and Robotics - NVIDIA DRIVE Thor, based on Blackwell, is being adopted by leading automotive companies like BYD, XPeng, and NIO, with production expected to start in 2025 [54] - NVIDIA's GROOT project supports humanoid robotics companies like 1X Technologies and Agility Robotics [54] Hopper Inference Performance - NVIDIA Hopper achieved leading performance in the latest MLPerf benchmarks, with NVIDIA TensorRT-LLM improving GPT-J LLM performance by nearly 3x compared to six months ago [55] Cloud Provider Capital Expenditures - Combined capital expenditures of the four major cloud providers (Meta, Amazon, Google, Microsoft) reached $53 2 billion in 24Q2, a 58% yoy increase and 19% qoq growth [45][47] - Meta revised its 2024 capital expenditure guidance to $37-40 billion, while Amazon expects to spend over $35 billion in the second half of 2024 [48][49] - Total capital expenditures for the four major cloud providers are expected to exceed $200 billion in 2024, driven by AI-related investments [48][49]