The only thing that ever does: a comparative advantage.
It may be that no model supplier is in this category and the only winners are hardware suppliers, if e.g. open weights models are good enough for approximately all* problems.
* Has to be very close to all problems, because an AI which is wildly better than anything else at what it does, even when what it does is only a percent or so of the economic problems, that's in the same space as "software" generally today, and that's a big enough market to supply multiple trillion-dollar-market-cap companies.
It may be that no model supplier is in this category and the only winners are hardware suppliers, if e.g. open weights models are good enough for approximately all* problems.
* Has to be very close to all problems, because an AI which is wildly better than anything else at what it does, even when what it does is only a percent or so of the economic problems, that's in the same space as "software" generally today, and that's a big enough market to supply multiple trillion-dollar-market-cap companies.