It might be they’re not trying anymore.

All the common AI tasks that by end-users and small businesses can be done better by local software rather than through compute on a hyperscale platform. The only reason to need big AI is for large-scale projects. They’re still selling compute at a huge loss, and their profits are dwarfed by their expenses.

(Some argue the expenses are due to the massive buildout, which is reaching a capability point of diminishing returns. But even the running expenses outweigh the profits, even with government contracts.)

You know, like controlling armies of killer drones, or processing mass surveillance data to find dissidents that are an actual threat. Or doing cyber-ops on a rival nation.

I think that is why the stock market is still investing in big AI, not because we want better slop, but because they want garrison large populations and wreck rival societies.