“Major innovations in AI algorithms and architectures, such as the Transformer and FlashAttention, are concentrated in labs with abundant GPU resources rather than in compute-poor environments like academia.”