ABI Research：GTC 2025：Token经济重新定义AI算力疆界白皮书（英文版）（15页）.pdf

1、GTC 2025:REDEFINING AIS COMPUTATIONAL HORIZON WITH TOKEN ECONOMYGTC 2025:REDEFINING AIS COMPUTATIONAL HORIZON WITH TOKEN ECONOMYNVIDIAs GPU Technology Conference(GTC 2025)marked a watershed moment,with in-person attendance surging to 25,000 attendees representing 800 global companiesup from 16,000 t

2、he previous yearwhile virtual participation held steady at approximately 300,000.The exhibition floor expanded to nearly 400 companies,with notable growth in the startup pavilion.The evident excitement throughout the event evoked memories of the early Mobile World Congress gatherings of the 2000s or

3、 Microsofts Professional Developers Conferences in their mid-2000s heyday.Jensen Huangs keynote articulated a paradigm shift transcending Moores Law,presenting a computational architecture vision operating at unprecedented scale to support next-gen-eration Artificial Intelligence(AI)frameworks,inclu

4、ding agentic systems.Despite its technical complexity,the message was clear:we are witnessing a fundamental transition from general purpose to accelerated computing,driven by exponential AI demand.This transformation redefines AI computational efficiency around quality token generationthe foundation

5、 of Generative Artificial Intelligence(Gen AI)and Agentic AI systems.The new paradigm integrates three critical elements:raw Graphics Processing Unit(GPU)perfor-mance,sophisticated GPU-GPU interconnect communications,and optimized storage-GPU interactions,all enhanced by workload-specific libraries(

6、CUDA-X)tailored to targeted ver-ticals and specific use cases.While competitors remain fixated on benchmarking individual chips using traditional metrics like Trillions of Operations Per Second(TOPS)or Floating Point Operations Per Second(FLOPS),NVIDIA has shifted the evaluation paradigm from data p