Tencent predicts ample GPU supply by year-end, but don't get too optimistic yet

Last week or so, I wrote a post breaking down Jeff Dean's BP, where I mentioned one thing: compute is currently the single largest capital expenditure for all AI companies. Today, I saw news about Tencent's Q2 earnings, where an exec said they expect GPU resources to be more abundant from the end of this year into early next year. In the context of an earnings report, my first reaction was: let me translate that—this is actually a signal of the inflection point where compute shifts from training to inference.