Alex Imas, Director of AGI Economics at Google DeepMind, posits that AGI can be economically transformative while remaining a mediocre business for standalone model labs due to converging capabilities. Competition is shifting from absolute intelligence to cost per completed task, leading to shortening frontier pricing power and compressing API margins. Labs need to achieve recursive self-improvement to lower costs and widen capability gaps before economics fully commoditize.
As AI capabilities converge, competition shifts to cost per task, compressing API margins and making successive training runs harder to fund for model labs.