FundaAI Says GPT-6 Training Compute May Be About 10 Times GPT-5

Pequity Research, citing FundaAI, said GPT-6 may use about 10 times the training compute of the GPT-5 generation. According to ChainCatcher, FundaAI also said that compute used for experiments, reinforcement learning, synthetic data generation, and related infrastructure could total several times the main pretraining run, potentially reaching as much as 10 times. FundaAI said the next round of pretraining acceleration could begin once TPU 8t and Vera Rubin ship at scale. It added that large-scale interconnect and optics would benefit the most during that cycle.