腾讯，重磅发布！马化腾发声

Core Viewpoint - Tencent's 2024 financial report shows a revenue of 660.3 billion RMB, an 8% year-on-year growth, and a NON-IFRS net profit of 222.7 billion RMB, up 41% from the previous year, indicating strong financial performance and growth potential [1] Financial Performance - Tencent's Q4 revenue reached 1724.5 billion RMB, marking an 11% year-on-year increase, with gross profit and operating profit (NON-IFRS) growing by 17% and 21% respectively, surpassing revenue growth for nine consecutive quarters [1] - The company plans to continue share buybacks in 2025, with an expected scale of at least 800 billion HKD, and a cash dividend increase of 32% to approximately 410 billion HKD, projecting total shareholder returns of at least 1210 billion HKD for 2025 [1] AI Strategy - Tencent's AI strategy has entered a phase of heavy investment, with R&D spending reaching 70.69 billion RMB in 2024, and a cumulative investment of 340.3 billion RMB over seven years. Capital expenditures have seen a three-digit percentage increase for four consecutive quarters, with annual capital expenditure exceeding 76.7 billion RMB, a 221% year-on-year growth, setting a historical high [2] - The company has restructured its AI team to focus on rapid product innovation and deep model development, increasing capital expenditure related to AI and enhancing R&D and marketing efforts for native AI products [2] AI Model Development - Tencent is embracing a multi-model strategy that combines self-developed core technologies with open-source models, aiming to create practical and evolving AI products and solutions based on user needs [5][6] - The company launched the mixed Yuan model in 2023, utilizing the MoE architecture, with flagship model parameters reaching trillion-level. Recently, it introduced the new generation fast-thinking model, mixed Yuan Turbo S, and is set to release the mixed Yuan T1 model, which excels in deep reasoning [6] - Tencent maintains an open and compatible approach towards open-source models, supporting dual model calls for various applications and platforms, and providing comprehensive support for enterprises to train their industry-specific large models [7]