Unverified50% confidenceFactExact time
AI推理工作负载要求GPU保持超过80%的持续占用率，一座超大规模数据中心用电量可达数百兆瓦
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedIn real-world projects, 80% of the time is spent on data preparation.69% similarUnverifiedSupernode computing power connects thousands to tens of thousands of GPUs/NPUs into a unified computing pool via high-speed interconnect networks, forming the necessary infrastructure for training large models with hundreds of billions to trillions of parameters.69% similarUnverifiedA large enterprise's AI system can consume hundreds of billions of tokens annually, and cost optimization per token directly impacts operational costs of hundreds of thousands or even millions of dollars.67% similarUnverifiedBy 2024, NVIDIA held more than 80% of the AI computing market share.66% similarUnverified据Canalys预测，到2028年AI PC将占全球PC出货量的70%以上65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561878API
curl https://kongchang.com/api/v1/knowledge/claims/561878MCP
get_claim(id=561878)