2 stories in this blend

DataCamp executive leadership revealed that scaling interactive AI tutoring across millions of student hours could cost between 10 million and 40 million dollars per year in model fees. Although open models like Google Gemma 4 31B outperform closed frontier models on internal quality tests, a lack of specialized hosting infrastructure prevents DataCamp from migrating live traffic to open weights.

SemiAnalysis founder Dylan Patel discussed hardware supply chains and computing requirements on a podcast with Dwarkesh Patel. He argued that steep capital costs and GPU allocation could soon leave just two major laboratories in control of most AI capabilities. The conversation covers how hardware trends shape market dominance across the technology sector.