论证人工智能工厂作为利润中心的可行性.pdf

《论证人工智能工厂作为利润中心的可行性.pdf》由会员分享，可在线阅读，更多相关《论证人工智能工厂作为利润中心的可行性.pdf（23页珍藏版）》请在三个皮匠报告上搜索。

1、Making AI factories a Profit center with Reduce,Reuse,Re-SourceChinmay KulkarniProduct Manager Data center APAC/Danfoss COOLING ENVIRONMENTSOutline(Optional)54321What are AI factories and its EconomyImportance of Reduce,Reuse,ResourceIncreasing Energy efficiency with Liquid cooling Optimizing Heat r

2、euse With Liquid cooling ESG Secondary income source through ESG ProjectsAs NVIDIA CEO Jensen Huang puts it:“Infrastructure is no longer just supporting productsit becomes the product.”AI factories eco system Revenue streams Core contributors Energy efficiency and heat re-useEnergy efficiency liquid

3、 cooling.Agenda/StorylineData centers to AI FactoriesAI factories are optimized for HPC workloads and accelerator hardware(GPUs,TPUs),unlike traditional enterprise infrastructure,which prioritized uptime over value generation.AI factories deploy high-density compute hardware such as GPUs and TPUs,re

4、sulting in rack-level power consumption exceeding 50 kW,compared to 510 kW in conventional CPU-based data centers.This reflects the significantly higher energy and thermal demands of AI workloads.Every inference request(via API,copilot,or embedded feature)produces monetizable tokens.Infrastructure n

5、ow directly powers revenue-generating workflows.Diagrams/ChartsThe Token EconomyMax AI token value=Powerful Model(Maximize)Cost infrastructure(Minimize)Powerful AI Model helps to increase value per task or value per token=primary revenue streamCost of infrastructure negatively impact overall value o

6、f AI token increases with increase in data,Though the number of internet users has more than doubled since 2010,global internet traffic has grown“20-fold.”Revenue is now tied directly to infrastructure throughput,not just software licensing.3 main components of infrastructureHigh Rack density low ph