Comment by augusto-moura

Not just shortage, any improvements to LLMs/electricity/servers memory footprint is becoming much more valuable as the time goes. If we can get 10% faster, you can easily get a lead in the LLM race. The incentives to transparently improving performance are tremendous