If all the talk is true that something approximating human intelligence can be scaled simply by building more datacenters.
How can an "approximation of human intelligence" be "scaled by building more datacenters"? That doesn't even make sense. You don't need more datacenters to demonstrate that you have an "approximation of human intelligence", all you need is a lab and at most 1 (one) datacenter. Only after you've demonstrated you know how to deliver the goods, you can think about scaling, retail and marketing.
The demand for the current "approximation" is out there for all to see, it's nothing like the "most desired commodity in the world", in fact, it needs huge inflationary subsidies in order to sustain some sort of demand-for-show.