based on current trends, it's going to be a parallel manifestation like Newton/Leibnitz calculus, the invention of radio, the 80's PC revolution, etc.
one could argue that the largest entities with the largest scale of compute can ship privatized AGI the soonest, but then again, frameworks like the Alberta Plan (https://www.youtube.com/watch?v=PvJ14d0r3CM ) are open-sourced, as are models like stablediffusion, and from what's been observed already, AGI may depend on the training data (which is less scalable than the compute). No, anyone doing the research can conclude that this is not a black swan.