Search
OpenAI, Surrounded by Crises
6 hour ago /
Read about 0 minute
Author：小编
It is estimated that the pre-training computational load of DeepSeek V4-Pro is approximately 1e25 FLOPs. OpenAI possesses 100,000 GB200 devices, and even with a utilization rate of only 15%, it would take just 37 hours to complete training on the same scale, meaning it could 'replicate' DeepSeek V4 within a day and a half. This demonstrates OpenAI's astonishing computational reserves as of April 2026. However, the issues facing OpenAI cannot be resolved solely through computational power.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
