The "lead" in AI, even with google's TPU advantage, may not make economic sense for a company. Can you make back the training investment on a competitive frontier model, before everyone switches to a newer frontier model within a year?