CAN AI SCALING CONTINUE THROUGH 2030?

Summary from the TLDR Founders newsletter

By 2030, it may be technically feasible for AI training runs to reach scales up to 10,000 times larger than current models, potentially continuing the historical trend of 4x annual growth in compute. The main constraints to this scaling are likely to be power availability and chip manufacturing capacity, though data scarcity and latency issues could also become limiting factors. Achieving such scales would require massive investments in infrastructure and could lead to significant advances in AI capabilities comparable to the progress seen in recent years.