Core Insights - The global computing infrastructure is entering a new round of "computing efficiency" strategic competition, with SenseTime's large device division leading in AI infrastructure and proposing a Chinese-style "computing-energy" collaborative construction paradigm [2][4] Group 1: Company Developments - SenseTime's large device division plans to launch a self-controlled "computing-energy collaborative intelligent scheduling platform" by July 2025, based on its long-term operational practices at its AIDC [2][4] - The platform integrates computing and electricity data, achieving significant economic and social benefits in its actual operation [2][4] Group 2: Technological Innovations - SenseTime's self-developed energy model utilizes a multi-modal MoE architecture, trained on vast industry knowledge texts, structured energy data, and computing monitoring indicators [2][4] - The model can predict computing load trends in advance and dynamically solve across systems by considering factors such as electricity price signals, green electricity ratios, energy storage status, and grid demand [2][4] Group 3: Performance Metrics - The energy demand prediction accuracy of the model has reached over 88%, with decision-making accuracy at over 93% [2][4] - Continuous optimization of algorithms and energy storage devices is expected to improve prediction accuracy to 90%-95% and decision-making accuracy to over 95%, establishing industry-leading standards [2][4]
商汤科技林海：我们正迎来新一轮“算力效能”战略竞争